Search

The popularity of egocentric cameras and their always-on nature has lead to the abundance of day long first-person videos. The highly redundant nature of these videos and extreme camera-shakes make them difficult to watch from beginning to end. These videos require efficient summarization tools for consumption. However, traditional summarization techniques developed for static surveillance videos or highly curated sports videos and movies are either not suitable or simply do not scale for such hours long videos in the wild. On the other hand, specialized summarization techniques developed for egocentric videos limit their focus to important objects and people. This paper presents a novel unsupervised reinforcement learning framework to summarize egocentric videos both in terms of length and the content. The proposed framework facilitates incorporating various prior preferences such as faces, places, or scene diversity and interactive user choice in terms of including or excluding the particular type of content. This approach can also be adapted to generate summaries of various lengths, making it possible to view even 1-minute summaries of one's entire day. When using the facial saliency-based reward, we show that our approach generates summaries focusing on social interactions, similar to the current state-of-the-art (SOTA). The quantitative comparisons on the benchmark Disney dataset show that our method achieves significant improvement in Relaxed F-Score (RFS) (29.60 compared to 19.21 from SOTA), BLEU score (0.68 compared to 0.67 from SOTA), Average Human Ranking (AHR), and unique events covered. Finally, we show that our technique can be applied to summarize traditional, short, hand-held videos as well, where we improve the SOTA F-score on benchmark SumMe and TVSum datasets from 41.4 to 46.40 and 57.6 to 58.3 respectively. We also provide a Pytorch implementation and a web demo at https://pravin74.github.io/Int-sum/index.html.

Temporal networks are ubiquitous in nature and society, and tracking the dynamics of networks is fundamental for investigating the mechanisms of systems. Dynamic communities in temporal networks simultaneously reflect the topology of the current snapshot (clustering accuracy) and historical ones (clustering drift). Current algorithms are criticized for their inability to characterize the dynamics of networks at the vertex level, independence of feature extraction and clustering, and high time complexity. In this study, we solve these problems by proposing a novel joint learning model for dynamic community detection in temporal networks (also known as jLMDC) via joining feature extraction and clustering. This model is formulated as a constrained optimization problem. Vertices are classified into dynamic and static groups by exploring the topological structure of temporal networks to fully exploit their dynamics at each time step. Then, jLMDC updates the features of dynamic vertices by preserving features of static ones during optimization. The advantage of jLMDC is that features are extracted under the guidance of clustering, promoting performance, and saving the running time of the algorithm. Finally, we extend jLMDC to detect the overlapping dynamic community in temporal networks. The experimental results on 11 temporal networks demonstrate that jLMDC improves accuracy up to 8.23% and saves 24.89% of running time on average compared to state-of-the-art methods.

The localization and segmentation of the novel coronavirus disease of 2019 (COVID-19) lesions from computerized tomography (CT) scans are of great significance for developing an efficient computer-aided diagnosis system. Deep learning (DL) has emerged as one of the best choices for developing such a system. However, several challenges limit the efficiency of DL approaches, including data heterogeneity, considerable variety in the shape and size of the lesions, lesion imbalance, and scarce annotation. In this article, a novel multitask regression network for segmenting COVID-19 lesions is proposed to address these challenges. We name the framework MT-nCov-Net. We formulate lesion segmentation as a multitask shape regression problem that enables partaking the poor-, intermediate-, and high-quality features between various tasks. A multiscale feature learning (MFL) module is presented to capture the multiscale semantic information, which helps to efficiently learn small and large lesion features while reducing the semantic gap between different scale representations. In addition, a fine-grained lesion localization (FLL) module is introduced to detect infection lesions using an adaptive dual-attention mechanism. The generated location map and the fused multiscale representations are subsequently passed to the lesion regression (LR) module to segment the infection lesions. MT-nCov-Net enables learning complete lesion properties to accurately segment the COVID-19 lesion by regressing its shape. MT-nCov-Net is experimentally evaluated on two public multisource datasets, and the overall performance validates its superiority over the current cutting-edge approaches and demonstrates its effectiveness in tackling the problems facing the diagnosis of COVID-19.

To utilize a huge amount of observation data based on real-world events, a data assimilation process is needed to estimate the state of the system behind the observed data. The Kalman filter is a very commonly used technique in data assimilation, but it has a problem in terms of practical use from the viewpoint of processing efficiency and estimating the deterioration in precision when applied to particularly large-scale datasets. In this paper, we propose a method that simultaneously addresses these problems and demonstrate its usefulness. The proposed method improves the processing efficiency and suppresses the deterioration in estimation precision by introducing correction processes focusing on the non-negative nature and sparseness of data in wavelet space. We show that the proposed method can accurately estimate population dynamics (MAE ≤ 3, RMSE ≤ 7) on the basis of an evaluation done using population data generated from cellular networks. In addition, the possibility of wide area abnormality detection using the proposed method is shown from a situation analysis of when Category 5 typhoon Hagibis made landfall in Japan. The proposed method has been deployed in a commercial service to estimate real-time population dynamics in Japan.

Unmanned Aerial Vehicles (UAV) have many applications in both commerce and recreation. However, irresponsibly operated UAVs will pose a threat to public safety. Therefore, developing our understanding of UAVs and their uses is of particular interest. This paper considers tracking UAVs, which provide multifaceted information around location, paths and trajectories. To facilitate research on this topic, we introduce a new benchmark, herein referred to as Anti-UAV, which provides a novel direction for UAV tracking with more than 300 video pairs containing over 580k manually annotated bounding boxes. Addressing anti-UAV research challenges could help to design anti-UAV systems, which in turn may improve surveillance. Accordingly, we have proposed a simple yet effective approach, called dual-flow semantic consistency (DFSC) is proposed for UAV tracking. Modulated by the semantic flow across video sequences, tracker learns more robust class-level semantic information and obtains more discriminative instance-level features. Experiments highlight significant performance gain with the proposed approach over state-of-the-art trackers and the challenging aspects of Anti-UAV. The Anti-UAV benchmark and the code for the proposed approach have been made publicly available at https://github.com/ucas-vg/Anti-UAV.

In order to solve the problem of poor fusion between the spots of deformation camouflage and the background, a small-spot deformation camouflage design algorithm based on background texture matching is proposed in this research. The combination of spots and textures improved the fusion of the spot pattern and the background. An adversarial autoencoder convolutional network was designed to extract background texture features. The image adversarial loss was added and the reconstruction loss was improved to improve the clarity of the generated texture pattern and the generalization ability of the model. The digital camouflage was formed by obtaining the mean value of the square area and replacing the main color. At the same time, the spots in the square area with a side length of 2 s were subjected to simple linear iterative clustering to form irregular small-spot camouflage. A dataset with a scale of 1050 was established in the experiment. The training results of three different loss functions were investigated. The results showed that the proposed loss function could enhance the generalization of the model and improve the quality of the generated texture image. A variety of digital camouflages with main colors and irregular small-spot camouflage were generated, and their efficiency was tested. On the one hand, intuitive evaluation was given by personnel observing the camouflage pattern embedded in the background and its contour map calculated by the canny operator. On the other hand, objective comparison result was formed by calculating the 4 evaluation indexes between the camouflage pattern and the background. Both results showed that the generated pattern had a high degree of fusion with the background. This model could balance the relationship between the spot size, the number of main colors and the actual effect according to actual needs.

Even when environments deteriorate gradually, ecosystems may shift abruptly from one state to another. Such catastrophic shifts are difficult to predict and reverse (hysteresis). While well studied in simplified contexts, we lack a general understanding of how catastrophic shifts spread in realistic spatial contexts. For different types of landscape structure, including typical terrestrial modular and riverine dendritic networks, we here investigate landscape-scale stability in metapopulations made of bistable patches. We find that such metapopulations usually exhibit large scale catastrophic shifts and hysteresis, and that the properties of these shifts depend strongly on metapopulation spatial structure and dispersal rate: intermediate dispersal rates and a riverine spatial structure can largely reduce hysteresis size. Interestingly, our study suggests that large-scale restoration is easier with spatially clustered restoration efforts and in populations characterized by an intermediate dispersal rate.

Pavement performance models are an essential component of a pavement management system (PMS) and have a direct impact on future pavement condition prediction, selection of pavement maintenance and rehabilitation (M&R) methods, and budget planning and allocations. Therefore, it is critical to develop and maintain pavement performance models as accurate as possible. The California Department of Transportation (Caltrans) has implemented a modern pavement management system called PaveM. To maintain the intended functions of PaveM, regular updates of its databases and key components are necessary, including pavement performance models. This paper aims to evaluate the network-level pavement performance models in PaveM by utilizing the concept of deterioration rate and the most recent automated pavement condition survey (APCS) data. First, the concept of pavement deterioration rate was defined. The actual deterioration rates were calculated using two latest cycles of APCS data and then compared to the predicted deterioration rates obtained using APCS data and the current configurations of PaveM. Two typical pavement distresses (International Roughness Index and Caltrans’ asphalt pavement alligator B cracking) for one selected pavement treatment (thin overlay) were predicted using PaveM and compared to the actual APCS measurements. The results from this study show that the concept of deterioration rate was effective in evaluating the overall quality of performance models in PaveM as well as identifying the time spans in which the models may over- or under-predict pavement performance.

People can infer the weather from clouds. Various weather phenomena are linked inextricably to clouds, which can be observed by meteorological satellites. Thus, cloud images obtained by meteorological satellites can be used to identify different weather phenomena to provide meteorological status and future projections. How to classify and recognize cloud images automatically, especially with deep learning, is an interesting topic. Generally speaking, large-scale training data are essential for deep learning. However, there is no such cloud images database to date. Thus, we propose a large-scale cloud image database for meteorological research (LSCIDMR). To the best of our knowledge, it is the first publicly available satellite cloud image benchmark database for meteorological research, in which weather systems are linked directly with the cloud images. LSCIDMR contains 104 390 high-resolution images, covering 11 classes with two different annotation methods: 1) single-label annotation and 2) multiple-label annotation, called LSCIDMR-S and LSCIDMR-M, respectively. The labels are annotated manually, and we obtain a total of 414 221 multiple labels and 40 625 single labels. Several representative deep learning methods are evaluated on the proposed LSCIDMR, and the results can serve as useful baselines for future research. Furthermore, experimental results demonstrate that it is possible to learn effective deep learning models from a sufficiently large image database for the cloud image classification.

The M5 competition uncertainty track aims for probabilistic forecasting of sales of thousands of Walmart retail goods. We show that the M5 competition data faces strong overdispersion and sporadic demand, especially zero demand. We discuss resulting modeling issues concerning adequate probabilistic forecasting of such count data processes. Unfortunately, the majority of popular prediction methods used in the M5 competition (e.g. lightgbm and xgboost GBMs) fails to address the data characteristics due to the considered objective functions. The distributional forecasting provides a suitable modeling approach for to the overcome those problems. The GAMLSS framework allows flexible probabilistic forecasting using low dimensional distributions. We illustrate, how the GAMLSS approach can be applied for the M5 competition data by modeling the location and scale parameter of various distributions, e.g. the negative binomial distribution. Finally, we discuss software packages for distributional modeling and their drawback, like the R package gamlss with its package extensions, and (deep) distributional forecasting libraries such as TensorFlow Probability.

Small-object detection has long been a challenge. High-megapixel cameras are used to solve this problem in industries. However, current detectors are inefficient for high-resolution images. In this work, we propose a new module called Pre-Locate Net, which is a plug-and-play structure that can be combined with most popular detectors. We inspire the use of classification ideas to obtain candidate regions in images, greatly reducing the amount of calculation, and thus achieving rapid detection in high-resolution images. Pre-Locate Net mainly includes two parts, candidate region classification and behavior classification. Candidate region classification is used to obtain a candidate region, and behavior classification is used to estimate the scale of an object. Different follow-up processing is adopted according to different scales to balance the variance of the network input. Different from the popular candidate region generation method, we abandon the idea of regression of a bounding box and adopt the concept of classification, so as to realize the prediction of a candidate region in the shallow network. We build a high-resolution dataset of aircraft and landing gears covering complex scenes to verify the effectiveness of our method. Compared to state-of-the-art detectors (e.g., Guided Anchoring, Libra-RCNN, and FASF), our method achieves the best mAP of 94.5 on 1920 × 1080 images at 16.7 FPS.

As a typical big data representative, the information of global Cellular Signal Strength (CSS), defined as the signal power received by mobile phones, plays an important role in geographic flow analysis, because the density of such information can reflect the urbanization variables such as population, gross domestic product, built-up area, electric power consumption, and etc. Despite the importance, the real-time analysis of global CSS distribution remains a challenging problem due to the large data scale. In this paper, a Display-driven Computing (DisDC) technique is designed and applied to provide efficient large scale interactive CSS visualization, generating results by calculating the value of each pixel that directly for display that can dramatically improve system capacity in big data handling. Specifically, we present an efficient CSS measurement algorithm, which introduces spatial indexes and a corresponding query strategy; besides, an optimized parallel computing architecture is proposed to ensure the ability of real-time visualization. Experiments show that our approach obviously outperforms traditional methods and is capable of handling more than 40 million base stations in real-time. Moreover, an online demonstration is provided at https://github.com/MemoryMmy/CSSMap.

Seizure prediction from intracranial electroencephalogram (iEEG) has great potentials to improve the life quality of epileptic patients, but faces big challenges. One major difficulty lies in that brain signal changes occasionally during long-term monitoring, due to electrode movements or the nonstationary brain dynamics. This leads to a serious situation that a predictor learned from historical data usually only works well in a short time period as long as the data do not change much. While in a long time span, the performance of the learned features decreases or even become totally invalid. To deal with the problem, we propose a domain adaptation convolutional neural network to learn robust preictal features that is invariant across different time periods. Specifically, the preictal feature is learned and enhanced by multi-scale temporal convolutions in the neural network. Based on this, a domain adaptation method is adopted to constrain that the learned features should be invariant across different time periods. Experimental results demonstrate that our approach can effectively improve seizure prediction performance against signal changes.

In the past ten years, many high-quality datasets have been released to support the rapid development of deep learning in the fields of computer vision, voice, and natural language processing. Nowadays, deep learning has become a key research component of the Sixth-Generation wireless systems (6G) with numerous regulatory and defense applications. In order to facilitate the application of deep learning in radio signal recognition, in this work, a large-scale real-world radio signal dataset is created based on a special aeronautical monitoring system - Automatic Dependent Surveillance-Broadcast (ADS-B). This paper makes two main contributions. First, an automatic data collection and labeling system is designed to capture over-the-air ADS-B signals in the open and real-world scenario without human participation. Through data cleaning and sorting, a high-quality dataset of ADS-B signals is created for radio signal recognition. Second, we conduct an in-depth study on the performance of deep learning models using the new dataset, as well as comparison with a recognition benchmark using machine learning and deep learning methods. Finally, we conclude this paper with a discussion of open problems in this area.

Breast cancer diagnosis from biopsy tissue images conducted manually by pathologists is costly, time-consuming, and disagreements among specialists. Nowadays, the advancement of the Computer-Aided Diagnosis (CAD) system allows pathologists to identify breast cancer more reliably and quickly.For this reason, interest in CAD-based deep learning models has been increased significantly. In this study, we propose a “MultiNet” framework based on the transfer learning concept to classify different breast cancer types using two publicly available datasets that include 7909 and 400 microscopic breast images, respectively. The proposed “MultiNet” framework is designed to provide fast and accurate diagnostics for breast cancer with binary classification (benign and malignant) and multi-class classification (benign, in situ, invasive, and normal). In the proposed framework, features from microscopy images are extracted using three well-known pre-trained models, including DenseNet-201, NasNetMobile, and VGG16. The extracted features are then fed into the concatenate layer, making a robust hybrid model. The proposed framework yields an overall classification accuracy of 99% in classifying two classes. It also achieves 98% classification accuracy in classifying four classes. Such promising results will provide the opportunity to use “MultiNet” framework as a diagnostic model in clinics and health care.

Landscape and quality attributes are major ecosystem characteristics closely associated with soil conservation service (SCS). However, the intrinsic mechanisms by which these two attributes influence SCS are still unclear. Therefore, this study quantitatively analyzed the landscape pattern, ecological quality, and SCS in the Lianshui River watershed (a typical soil and water loss area of red soil in southern China) and its sub-watersheds in 2019. The boosted regression tree model was used to explore the influence of 15 factors (i.e., landscape and quality attributes) on SCS at the sub-watershed scale. According to the results, compared with the landscape attribute, the quality attribute of the watershed ecosystem could better explain the spatial heterogeneity of SCS across 66 sub-watersheds. The overall degree of influence of five quality factors on SCS reached 57.81%, with the highest being the normalized differential build-up and bare soil index (NDBSI), at 25.11%. Among 10 landscape factors, aggregation had the greatest influence on SCS, at 28.64%. The relationships between key influencing factors and SCS were non-monotonic and non-linear, with threshold effects. For example, NDBSI values of 0.18–0.41 had a positive influence on SCS, while NDBSI values of 0.41–0.65 had a negative influence on SCS. The findings broaden our understanding of the response of SCS to changes in landscape and quality attributes at the sub-watershed scale, and could offer comprehensive support for soil erosion management in the watershed ecosystem.

Strategic maintenance is essential for sustainable road infrastructure development. Accurate estimation of road maintenance effects can support the assessment of maintenance strategies and reasonable allocation of budgets and resources. Road deterioration is affected by sophisticated factors, but accurate investigation of the integrated deterioration factors is limited. This study developed a dynamic trade-off model (DTOM), a hybrid nonlinear and machine learning method, for quantifying temporally varied impacts of factors and examining maintenance effects at the network level. Pavement deterioration factors are classified into three categories: (i) historical observations of roughness, (ii) pavement age, and (iii) traffic, climate and environment factors. Their respective impacts on pavements are estimated using a non-linear least square regression, a joinpoint regression and a random forest model, respectively. Vehicle-based laser scanner monitored high-resolution deterioration data was collected for a large spatial scale road network in Western Australia from 2007 to 2018. Results show that the resurfacing and rehabilitation are essential for strategic reduction of deterioration. Twelve-year maintenance activities reduced the distress of roughness by 7.5% and increased road performance (the percentage of roads with roughness lower than 2.085 IRI) by 14.5% for the whole road network. The DTOM has great potentials in accurately assessing infrastructure maintenance effects and predicting deterioration scenarios.

Pedestrian trajectory prediction in dynamic, multi-agent traffic junction scene is an important problem in the context of self-driving cars. Accurately predicting the trajectory of the agent, especially the pedestrian with high randomness, is of great significance to autonomous driving technology. In this paper, we propose Multi-PPTP, a novel trajectory prediction model that first utilizes a composite rasterized map to model the complex representations and interactions of road components, including dynamic obstacles (e.g., pedestrians, vehicles, and cyclists) and static road information (e.g., lanes, traffic lights, and sidewalks). Then it uses MapNet and AgentNet to extract spatio-temporal features by deep convolutional networks and LSTM to automatically derive relevant features, and next exploits Interaction-AttNet aggregate features to learn the high interactions among all components by affine transformation and multi-head attention mechanism. Additionally, we also propose a series of unique loss functions to predict multiple possible trajectories of each pedestrian while estimating their probabilities. Following extensive offline evaluation and comparison to the state-of-the-art models, our approach outperforms significantly the state-of-the-art models on our large scale in-house and public nuScenes dataset.

Soil moisture estimates at high spatial and temporal resolution are of great value for optimizing water and agricultural management. To fill the gap between local ground observations and coarse spatial resolution remote sensing products, we use Soil Moisture Active Passive (SMAP) and Sentinel-1 data together with a unique data set of ground-based soil moisture estimates by cosmic ray neutron sensors (CRNS) and capacitance probes to test the possibility of downscaling soil moisture to the sub-kilometre resolution. For a high-latitude study area within a highly heterogeneous landscape and diverse land use in Denmark, we first show that SMAP soil moisture and Sentinel-1 backscatter time series correlate well with in situ CRNS observations. Sentinel-1 backscatter in both VV and VH polarizations shows a strong correlation with CRNS soil moisture at higher spatial resolutions (20–400 m) and exhibits distinct and meaningful signals at different land cover types. Satisfactory statistical correlations with CRNS soil moisture time series and capacitance probes are obtained using the SMAP Sentinel-1 downscaling algorithm. Accounting for different land use in the downscaling algorithm additionally improved the spatial distribution. However, the downscaling algorithm investigated here does not fully account for the vegetation dependency at sub-kilometre resolution. The study suggests that future research focussing on further modifying the downscaling algorithm could improve representative soil moisture patterns at a fine scale since backscatter signals are clearly informative. Highlights. Backscatter produces informative signals even at high resolutions. At the 100 m scale, the Sentinel-1 VV and VH polarizations are soil moisture dependent. The downscaling algorithm is improved by introducing land-cover-dependent clusters. The downscaled satellite and CRNS soil moisture agree best at the agricultural site.

Recovering the three-dimensional structure information from a monocular camera is significant for automated driving, robot navigation, and traffic safety assessment. Recent work has solved various tight issues on self-supervised monocular depth estimation of leveraging on-board videos, such as occlusion/disocclusion, dynamic objects, and scale inconsistent. Nevertheless, rare work focuses on the model's prediction confidence and underlying relations between depths, while they are essential for a decision-making system and performance improvement, respectively. This paper proposes a novel scheme, that of correlation-aware structure, to dig into the relations between depths, converting the independent depths into a graph-like connected depth map. Subsequently, a Gaussian estimator is devised to predict the depth map and uncertainty map concurrently. The uncertainty map can show us problematic regions where it is difficult to predict, from which we further develop uncertainty-based strategies to improve the performance. Specifically, we propose a simple image preprocessing method to overcome the gradient locality issue caused by low-texture, especially in smooth roads and shadows. Also, to avoid the influence of high-uncertainty regions, we propose a solidity-aware mask to recognize the reliable pixels for training in the image. The experiments on the KITTI dataset show that our method results in a competitive performance in both depth and ego-motion estimation tasks compared with the state-of-the-art methods. Besides, additional experiments on the Make3D and Cityscapes datasets demonstrate our method's strong generalization capability and practicality.

With the advent of big data and major technological achievements in information and communications technology, abundant human trajectory data are collected by location-aware devices such as mobile phones. This information has spurred the development of several methods for measuring spatial interaction, such as the gravity model and radiation model. However, prior studies have mainly measured the scale of spatial interaction at an aggregated level and have derived static or unidirectional interaction. As an improvement, an individual differentiated multidimensional (IDMD) Hawkes model, which can capture individual mobility differences as well as the dynamic mutual interaction between geographical units, is proposed herein. Internet protocol detail records (IPDRs) of Jinhua, Zhejiang, China, covering 23 days, are used as a case study to infer the spatial structure of urban interaction. Compared to the traditional spatial interaction model, the IDMD Hawkes model demonstrates unique advantages in capturing spatial interactions. The results reveal the level of interaction between regions and identify the closely related districts; this can be verified by the economic exchange and transportation convenience between these regions. Our approach provides a novel perspective for measuring spatial interaction considering individual mobility, which can assist policymakers to better understand the spatial structure of a city and plan an efficient city configuration.

Repair scheduling and routing and logistics support are interdependent and critical for post-disaster distribution system restoration (PDSR), which is also influenced by the geographical characteristics of outage area. Hence, we develop a co-optimization model for the PDSR with logistics support and geographical characteristics. A hybrid improved bacterial colony chemotaxis algorithm is proposed to solve the model, in which A* algorithm is employed to route repair crews and material delivery in the transportation network considering geographical characteristics, and an improved bacterial colony chemotaxis algorithm is proposed to determine the repair scheduling and material allocation in the distribution system. Different scale of distribution system instances with different damage levels and different geographical characteristics are used to demonstrate the effectiveness of the proposed methodology.

Urban sprawl is a phenomenon with a negative effect which, over time, consumes the living environment . The aim of this research is to analyze and evaluate the urban sprawl of Annaba situated at the northeastern part of Algeria in a diachronic approach. The studied area contains twelve agglomerations. Through the supervised classification of a time series of data provided by multi spectral sensors mounted on LANDSAT 8 Operati onal Land Imager (OLI), LANDSAT 7 Thematic Mapper Plus satellite platforms (ETM +) and L ANDSAT 45 Thematic Mapper (TM) the urban sprawl was examined in different periods of 1987, 2013 and 2017, on different scales. This study displayed that a retrospective understanding of the timing and magnitude of Annaba sprawl in a spatio temporal dimension allows updating the phenomenon to better manage it in the future on either macro or micro scale as a decision support tool.

The use of Artificial Neural Network (ANN) approaches has gained a significant role over the last decade in the field of predicting the distribution of effects triggered by natural forcing, this being particularly relevant for the development of adequate risk mitigation strategies. Among the most critical features of these approaches, there are the accurate geolocation of the available data as well as their numerosity and spatial distribution. The use of an ANN has never been tested at a national scale in Italy, especially in estimating earthquake-triggered landslides susceptibility. Based on the statistics deductible from the most up-to-date national inventory of earthquake-induced ground effects, i.e. the CEDIT catalogue, it results that over 56\% of the ground effects triggered by earthquakes in Italy are represented by landslides. Therefore, a landslide dataset with such high geolocation precision was suitable to evaluate the efficiency of an ANN to explain the distribution of landslides over the Italian territory. An ex-post evaluation of the ANN-based susceptibility model was also performed, using a sub-dataset of historical data with lower geolocation precision. The ANN training highly performed in terms of spatial prediction, by partitioning the Italian landscape into slope units. The obtained results returned a distribution of potentially unstable slope units with maximum concentrations primarily distributed in the central-northern Apennines and secondarily in the southern Apennines. Moreover, the Alpine sector clearly appeared to be divided into two areas, a western one with relatively low susceptibility to earthquake-triggered landslides and the eastern sector with a higher susceptibility. However, the scale of the analysis carried out to train the ANN does not allow it to be applied for planning purposes or for seismic microzonation studies, for which training on a smaller spatial scale will be required.

The eddy-covariance technique measures carbon, water, and energy fluxes between the land surface and the atmosphere at hundreds of sites globally. Collections of standardised and homogenised flux estimates such as the LaThuile, Fluxnet2015, National Ecological Observatory Network (NEON), Integrated Carbon Observation System (ICOS), AsiaFlux, AmeriFlux, and Terrestrial Ecosystem Research Network (TERN)/OzFlux data sets are invaluable to study land surface processes and vegetation functioning at the ecosystem scale. Space-borne measurements give complementary information on the state of the land surface in the surroundings of the towers. They aid the interpretation of the fluxes and support the benchmarking of terrestrial biosphere models. However, insufficient quality and frequent and/or long gaps are recurrent problems in applying the remotely sensed data and may considerably affect the scientific conclusions. Here, we describe a standardised procedure to extract, quality filter, and gap-fill Earth observation data from the MODIS instruments and the Landsat satellites. The methods consistently process surface reflectance in individual spectral bands, derived vegetation indices, and land surface temperature. A geometrical correction estimates the magnitude of land surface temperature as if seen from nadir or 40∘ off-nadir. Finally, we offer the community living data sets of pre-processed Earth observation data, where version 1.0 features the MCD43A4/A2 and MxD11A1 MODIS products and Landsat Collection 1 Tier 1 and Tier 2 products in a radius of 2 km around 338 flux sites. The data sets we provide can widely facilitate the integration of activities in the eddy-covariance, remote sensing, and modelling fields.

In general, vehicle images have varying resolutions due to vehicles’ movements and different camera settings. However, most existing vehicle re-identification models are single-resolution deep networks trained with pre-uniformly resizing vehicle images, which underestimates adverse effects of varying resolutions and leads to unsatisfactory performance. A straightforward solution for dealing with varying resolutions is to train multiple vehicle re-identification models. Each model is independently trained with images of a specific resolution. However, this straightforward solution requires significant overhead and ignores intrinsic associations among different resolution images. For that, an efficient multi-resolution network (EMRN) is proposed for vehicle re-identification in this paper. Firstly, EMRN embeds a newly-designed multi-resolution feature dimension uniform module (MR-FDUM) behind a traditional backbone network (i.e., ResNet-50). As a result, the whole model can extract fixed dimensional features from different resolution images so that it can be trained with one loss function of fixed dimensional parameters rather than training multiple models. Secondly, a multi-resolution image randomly feeding strategy is designed to train EMRN, making each mini-batch data of a random resolution during the training process. Consequently, EMRN can implicitly learn collaborative multi-resolution features via only a unitary deep network. Experiments on three large scale datasets, i.e., VeRi776, VehicleID, and VRIC, demonstrate that EMRN is superior to state-of-the-art vehicle re-identification methods.

Images contain not only individual entities, but also abundant visual relationships between entities. Therefore, conditioned on visual relationship triples that can be viewed as structured texts, entities (subjects or objects) can be localized in images without ambiguity. However, it is challenging to efficiently model visual relationships since a specific relationship usually has dramatic intra-class visual differences when involved with different entities, quite a number of which are in small scale. In addition, the subject and the object in a relationship triple may have different best scales, and matching the subject and the object with different appropriate scales may improve prediction. To address these issues, a multi-scale conditional relationship graph network is proposed in this paper to localize entities based on visual relationships. Specifically, an attention pyramid network is first introduced to generate multi-scale attention maps to capture entities with various sizes for entity matching. Then, a conditional relationship graph network is further designed to aggregate and refine multi-scale attention features to localize entities via passing relationship contexts between entity attention maps, which sufficiently utilizes the entity attention maps with the best scales. In order to mitigate the negative effects of intra-class visual differences of relationships, vision-agnostic relationship features are utilized in the proposed conditional relationship graph network to indirectly model relationship contexts. The experiments demonstrate the superiority of the proposed method compared with the previous powerful frameworks on three challenging benchmark datasets including CLEVR, Visual Genome and VRD. The project page can be found in https://mic.tongji.edu.cn/d9/5c/c9778a186716/page.htm.

The ability to sense and understanding the driving environment is a key technology for ADAS and autonomous driving. Human drivers have to pay more visual attention to important or target elements and ignore unnecessary ones present in their field of sight. A model that computes this visual attention of targets in a specific driving environment is essential and useful in supporting autonomous driving, object-specific tracking & detection, driving training, car collision warning, traffic sign detection, etc. In this paper, we propose a new framework of visual attention that can predict important objects in the driving scene using a conditional generative adversarial network. A large scale Visual Attention Driving Database (VADD) of saliency heat-maps is built from existing driving datasets using a saliency mechanism. The proposed framework model takes its strength from these saliency heat-maps as conditioning label variables. The results show that the proposed approach makes us able to predict heat-maps of most important objects in a driving environment.

Insights into the rates at which ecosystems and vegetation respond to a changing climate is fundamental to anticipating impacts of projected climate change. Characterization of vegetation change over millennia to centuries has potential to make an important contribution toward this goal, and regional scale syntheses of fossil pollen data can provide the foundation for this understanding. However, challenges of data analysis and integration are nontrivial. Here we present a case study in which publicly available fossil pollen data for California are assessed and analyzed. The data are selected according to a clearly defined selection criteria, and a Rate of Change (RoC) value is calculated to assess rates of vegetation change in California from ~15k BP (before A.D. 1950) to the present. Our results highlight several challenges presented by the extant data sets, including temporal sampling variation within and between records, large age control uncertainties, and sparse, geographically biased coverage. Recommendations for methodological refinements to better characterize ecological rates of change include increasing sampling frequency, maintaining a consistent temporal spacing within records, and applying probabilistic approaches to existing pollen data sets.

In recent years, learning-based single image dehazing networks have been comprehensively developed. However, performance improvement is limited due to domain shift between trained synthetic hazy images and untrained real-world hazy images. To alleviate this issue, this paper proposes a real-world dehazing targeted training scheme which nearly realizes paired real-world data training. As a result, a Twofold Multi-scale Generative Adversarial Network (TMS-GAN) consisting of a Haze-generation GAN (HgGAN) and a Haze-removal GAN (HrGAN) is designed. HgGAN attributes real haze properties to synthetic images and HrGAN removes haze from both synthetic and generated fake realistic data under supervision. Thus, the proposed method can better adapt to real-world image dehazing using this cooperative training scheme. Meanwhile, several structural advances of TMS-GAN also improve dehazing performance. Specifically, a haze residual map based on atmospheric scattering model is deduced in HgGAN for fake realistic data generation. The dual-branch generator in HrGAN draws attention to detail restoration by one branch along with another color-branch. A plug-and-play Multi-attention Progressive Fusion Module (MAPFM) is proposed and inserted in both HgGAN and HrGAN. MAPFM incorporates multi-attention mechanism to guide multi-scale feature fusion in a progressive manner, in which Adjacency-attention Block (AAB) can capture contributing features of each level and Self-attention Block (SAB) can establish non-local dependency of feature fusion. Experiments on mainstream benchmarks show that the proposed framework is superior especially on real-world hazy images among single image dehazing methods.

Continuous monitoring of road surface quality is necessary to maintain high functionality of the entire road networks and allow users to perceive road roughness. Although several fundamental methods have been used to evaluate road roughness by employing smart probe cars, the accuracy of these methods is affected by factors, such as vehicle speed, sensor positions, and vehicle suspension systems. Also, the long-term application of these methods to the entire road network is also limited due to the high cost. Further, the widely used roughness indices do not reflect road users' perception about road roughness as these indices are also affected by the accuracy of the roughness evaluation methods. Therefore, we propose a crowdsourcing based road roughness evaluation model which uses power spectral density accompanied with blind source separation technique to eliminate the vehicle effects. We also propose a road surface ranking based on majority voting algorithm for comparing roads based on their surface quality. Finally, the road surface roughness index is derived to widen the range of quality scale and capture roughness at fine granularity. The models are tested by real world experiments on different roads in Taiwan. The results show that the proposed model can accurately measure the roughness of roads, rank these roads and index them accordingly in a way that shows tiny differences in the road surface quality.

Under the condition of network operation of high-speed railways (HSRs), the influence of disruptions on the train control and dispatching at the current line and related lines is more and more significant. This article focuses on the timetable cooperative rescheduling problem with multidispatching sections of HSRs from a macroscopic perspective in the case of large disruptions. The problem is formulated as a mixed-integer linear program (MILP) model on the objectives of minimizing the weighted sum of the arrival delay time of trains, the delay time of depart trains at the handover station, and the number of delays of trains at all stations. The strategies of retiming and reordering are adopted to generate the rescheduling scheme and reduce delay propagation by making full use of three kinds of buffer time reserved in the timetable, i.e., buffer times of train operation in the station, train running in the section, and electric multiple unit (EMU) connection. A case study of the timetable rescheduling at the two adjacent dispatching sections of the Beijing-Shanghai HSRs line is conducted to evaluate the performance of the proposed integrated rescheduling approach. The relationship between computing time and the quality of rescheduling schemes is also investigated. The computational results show that the proposed approach can generate a conflict-free timetable with the minimum arrival delay time and the number of delays of all trains at all stations compared to the nonintegrated rescheduling approach and the benchmark solution of the first-come-first-serve (FCFS) approach. The propagation of delay between dispatching sections is also greatly suppressed. The results can provide support for dispatchers to making reasonable rescheduling decisions in the case of large disruptions.

Real-life systems involving interacting objects are typically modeled as graphs and can often grow very large in size. Revealing the community structure of such systems is crucial in helping us better understand their complex nature. A critical relevant challenge is the discovery of the overlapping communities of a given node in a billion-node graph. In this work, we propose a scalable local community detection approach to efficiently unfold the communities of individual target nodes. Our goal is to reveal the clusters formed around nodes by leveraging the relations within all different contexts these nodes participate in. To this end, we consider the similarity of pairs of links in the graph as well as the extent of their participation in multiple contexts. Then, we determine the order in which we should group the pairs of links so that we form meaningful hierarchical communities. We are not affected by constraints existing in previous techniques such as the need for several seed nodes. Our experimental evaluation on large real-world networks shows that we significantly outperform state-of-the-art methods on both accuracy and efficiency. Moreover, we show that we uncover very effectively the hierarchical structure of overlapping communities by producing detailed dendrograms.

Recently, deep convolutional neural networks (CNNs) have emerged as powerful tools for the correspondence problem in stereo matching task. However, the existence of multiscale objects and inevitable ill-conditioned regions such as textureless regions in real-world scene images continue to challenge current CNN architectures. In this paper, we present a robust scale-aware stereo matching network which aims to predict multiscale disparity maps and fuse them to achieve a more accurate disparity map. To this end, powerful feature representations are extracted from stereo images and are concatenated into a 4D feature volume. The feature volume is then fed into a series of connected encoder-decoder cost aggregation structures for the construction of multi-scale cost volumes. Following this, we regress multi-scale disparity maps from the multi-scale cost volumes and feed them into a fusion module to predict final disparity map. However, uncertainty estimations at each scale and complex disparity relationships among neighboring pixels pose a challenge on the disparity fusion. To overcome this challenge, we design a robust learning based scale-aware disparity map fusion model which seeks to map multi-scale disparity maps onto the ground truth disparity map by leveraging their complementary strengths. Experimental results show that the proposed network is more robust and outperforms recent methods on standard stereo evaluation benchmarks.

Shallow, light to moderate magnitude earthquakes in stable continental regions can have a damaging impact on vulnerable surface constructions. In the coal area of the Hainaut province in Belgium, a century of shallow seismic activity occurred from the end of the 19th century until the late 20th century. This seismicity is the second largest source of seismic hazard in northwestern Europe, after the Lower Rhine Embayment. The present study synthesises the impact and damage caused by this unique shallow seismicity. Reviewing intensity data provided in official macroseismic surveys held by the Royal Observatory of Belgium, press reports, and contemporary scientific studies resulted in a complete macroseismic intensity dataset. The strong shaking of five seismic events with moment magnitudes Mw around 4.0, which occurred on 3 June 1911, 3 April 1949, 15 December 1965, 16 January 1966, and 28 March 1967, locally caused widespread moderate damage to buildings corresponding to maximum intensity VII in the EMS-98 scale. For 28 earthquakes, detailed macroseismic maps were created. Our study highlights the capability of shallow, small-magnitude earthquakes to generate damage. Subsequently, using the Hainaut intensity dataset, we modelled a new Hainaut intensity attenuation law and created relationships linking magnitude, epicentral intensity and focal depth. Using these relationships, we estimated the location and magnitude of pre-1985 earthquakes that occurred prior to deployment of the modern digital Belgian seismic network. Estimated focal depths allowed discriminating between two different types of earthquakes. Some events were very shallow, only a few hundred metres deep, suggesting a close link to mining activities. Other earthquakes, including the largest and most damaging events, occurred at depths greater than 2 km but no deeper than 6 km, which would exclude a direct relationship with mining, but yet still might imply a triggering causality. This work results in a new updated earthquake catalogue including 123 seismic events. Our attenuation modelling moreover suggests that current hazard maps overestimated ground motion levels in the Hainaut area due to the use of inadequate ground motion prediction equations. Our Hainaut attenuation model is hence useful to evaluate the potential impact of current and future, e.g. geothermal energy, projects in the Hainaut area and other regions with a similar geological configuration.

Soil erodibility (K factor) mapping has been accomplished mainly by soil map-linked or geo-statistical interpolation. However, the resulting maps usually have coarse spatial resolution at a regional scale. The objectives of this study were a) to map the K factors using a set of environmental variables and random forest (RF) model, and b) to identify the important environmental variables in the predictive mapping on a regional scale. We collected 101 surface soil samples across southeast China in the summer of 2019. For each sample, we measured the particle size distribution and organic matter content, and calculated the K factors using the nomograph equation. The hyperparameters of RF were optimized through 5-fold cross validation (mtry = 2, ntree = 500, p = 63), and a digital map with 250 m resolution was generated for the K factor. The lower and upper limits of a 90% prediction interval were also produced for uncertainty analysis. It was found that the important environmental variables for the K factor prediction were relief, climate, land surface temperature and vegetation indexes. Since the existing K factor map has an average polygonal area of 6.8 km2, our approach dramatically improves the spatial resolution of the K factor to 0.0625 km2. The new method captures more distinct differences in spatial details, and the spatial distribution of the K factor derived from RF prediction followed a similar pattern with kriging interpolation. This suggests the presented approach in this study is effective for mapping the K factor with limited sampling data.

UAV swarms have triggered wide concern due to their potential application values in recent years. While there are studies proposed in terms of the architecture design for UAV swarms, two main challenges still exist: (1) Scalability, supporting a large scale of vehicles; (2) Versatility, integrating diversified missions. To this end, a multi-layered and distributed architecture for mission oriented miniature fixed-wing UAV swarms is presented in this paper. The proposed architecture is built on the concept of modularity. It divides the overall system to five layers: low-level control, high-level control, coordination, communication and human interaction layers, and many modules that can be viewed as black boxes with interfaces of inputs and outputs. In this way, not only the complexity of developing a large system can be reduced, but also the versatility of supporting diversified missions can be ensured. Furthermore, the proposed architecture is fully distributed that each UAV performs the decision-making procedure autonomously so as to achieve better scalability. Moreover, different kinds of aerial platforms can be feasibly extended by using the control allocation matrices and the integrated hardware box. A prototype swarm system based on the proposed architecture is built and the proposed architecture is evaluated through field experiments with a scale of 21 fixed-wing UAVs. Particularly, to the best of our knowledge, this paper is the first work which successfully demonstrates formation flight, target recognition and tracking missions within an integrated architecture for fixed-wing UAV swarms through field experiments.

Individual differences exist in the ability to create an accurate mental survey representation (i.e., a cognitive map) of a novel environment, yet the mechanisms underlying differences in cognitive map accuracy are still under investigation. To determine whether differences in overt attention allocation contribute to these individual differences, the current study examined whether looking times to landmarks and other objects while navigating in a dynamic virtual environment were related to cognitive map accuracy. Participants completed a battery of spatial tests; some tests assessed spatial skills prior to the navigation task (the Santa Barbara Sense of Direction Scale and the Spatial Orientation Test), and others tested memory of the virtual environment Silcton after an exploration period (a landmark recognition task, a direction estimation task, a map-building task, and a route construction task). Individuals with inaccurate cognitive maps of Silcton, as measured by the direction estimation and map-building tasks, showed equivalent eye fixations to buildings and objects when exploring Silcton as those with accurate maps. Despite similar looking times, the inaccurate mappers were significantly worse at judgments of relative direction between landmarks in Silcton and showed poorer memory for landmarks in Silcton than accurate mappers. These findings suggest that cognitive mechanisms, such as mental perspective-taking, occurring after attention allocation underlie differences in cognitive map accuracy. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Fine-grained vehicle categorization has evolved into a significant subject of study due to its importance in the Intelligent Transportation System. A highly accurate and real-time vehicle categorization system will help to support many applications not only in the security aspect but also many walks of life. In this paper, facing the growing importance of this study, we present an image dataset named Frontal-103 to promote the development of the vision-based research on the vehicle, and particularly for the task of fine-grained vehicle categorization. This paper provides a detailed analysis of Frontal-103 in its current state: 1,759 fine-grained vehicle models in 103 vehicle makes and 65,433 web-nature images in total. Apart from the specific viewpoint and vehicle hierarchy, Frontal-103 is superior to the other state-of-the-art vehicle image datasets not only in the scale and diversity but also the accuracy and fine-grained level. We further discuss the peculiar challenges and issues lies in the task of fine-grained vehicle categorization and illustrate the usefulness of our dataset in addressing those problems. We hope Frontal-103 will be beneficial to the vision-based vehicle analysis and contribute to the computer vision community.

Deforestation is cutting trees of forests on a huge scale, often resulting in loss of habitat of millions of wild animals. About 30% of earths land is still covered with forests but due to deforestation we are losing them at the rate of about half the size of England per year. In forests, tree cutting activities are illegal but due to shortage of manpower and other resources, governments are not very successful in curbing this menace. One way to stop this is to detect the tree cutting process in an early stage so that timely measures can be taken to stop the same. The simplest method of early detection of tree cutting is to regularly monitor the forest area either manually or using some automatic techniques. As tree cutting generates lot of noise, it can be detected by regularly monitoring the acoustic signals inside the forest. An acoustic signature can provide valuable information about the activities of any intruder inside the forest. This paper proposes an algorithm for automatic detection of tree cutting in forest. The proposed algorithm is based on distance between parameters, along with K-means clustering, GMM and PCA for comparison. The efficiency of proposed algorithm is 92%.

With the integration of the large-scale wind farms, the conventional pilot protection is facing new challenges. It suffers from the performance degradation can even refuse operation for some extreme faut scenarios. To address this problem, this paper presents a novel pilot protection based on cosine similarity. The protection setting calculation and industrial application testing is investigated. The functional structure is described and the logic of the cosine similarity based protection is evaluated under different fault scenarios. Compared with the conventional differential protection, application of the proposed protection can greatly improve the relay performance. The proposed protection is embedded into an industry relay device, the standard hardware-in-loop tests are carried out using the real time digital simulator (RTDS) before the industrial application. Now the protection device based on cosine similarity is installed in the wind farms and put into operation in Inner Mongolia.

Predictions of how vegetation responds to spatial and temporal differences in climate rely on established links with plant functional traits and vegetation types that can be encoded into Dynamic Global Vegetation Models. Individual traits have been linked to climate at species level and at community level within regions. However, a recent global assessment of aggregated community level traits found unexpectedly weak links with macroclimate, bringing into question broadscale trait–climate associations and implicating local-scale environmental differences in the filtering of communities. To further evaluate patterns in light of these somewhat contradictory results, we quantified the power of macro-environmental variables to explain aggregated plant community traits, taking advantage of new trait data for leaf area, plant height and seed mass combined with a national survey that records cover-abundance using consistent methods for a large number of plots across Australia. In contrast to the global study, we found that abundance-weighted community mean and variance of leaf area and maximum height were correlated with macroclimate. Height and leaf area were highest in wet (especially warm, wet) environments, with actual evapotranspiration explaining 30% of variation in leaf area and 26% in maximum height. Seed mass was weakly related to environment, with no variable explaining more than 5% of variance. Considering all three traits together in a redundancy analysis, the complete set of environmental variables explained 43% of variation in site-mean traits and 29% of within-site trait variance. While significant trait variation remains unexplained, the trait–environment relationships reported here suggest climatically-driven filtering plays a strong role in assembling these vegetation communities. Regional assessments using standardised species abundances can therefore be used to predict aspects of vegetation function. Our quantification of plant community trait patterns along macroclimatic gradients at continental scale thereby contributes a much-needed functional basis for Australian vegetation.

Albeit great success has been achieved in image defocus blur detection, there are still several unsolved challenges, e.g., interference of background clutter, scale sensitivity and missing boundary details of blur regions. To deal with these issues, we propose a deep neural network which recurrently fuses and refines multi-scale deep features (DeFusionNet) for defocus blur detection. We first fuse the features from different layers of FCN as shallow features and semantic features, respectively. Then, the fused shallow features are propagated to deep layers for refining the details of detected defocus blur regions, and the fused semantic features are propagated to shallow layers to assist in better locating blur regions. The fusion and refinement are carried out recurrently. In order to narrow the gap between low-level and high-level features, we embed a feature adaptation module before feature propagating to exploit the complementary information as well as reduce the contradictory response of different feature layers. Since different feature channels are with different extents of discrimination for detecting blur regions, we design a channel attention module to select discriminative features for feature refinement. Finally, the output of each layer at last recurrent step are fused to obtain the final result. We collect a new dataset consists of various challenging images and their pixel-wise annotations for promoting further study. Extensive experiments on two commonly used datasets and our newly collected one are conducted to demonstrate both the efficacy and efficiency of DeFusionNet.

This paper investigates the spatiotemporal characteristics and predictability of the emerging modern traffic behavior, ridesourcing. We collect a comprehensive data set of Didi ridesourcing cars on a large geographical scale of a capital city in China, including both the temporal order information and the GPS-recorded spatial trajectories. To extract the features of this kind of traffic behavior, we construct a large-scale network by considering every traffic flow of the orders. Therein, a driver consecutively visiting different regions of the city connects the relationship of these sites. The weighted ridesourcing network shows a consistency of the distribution of trip orders and the Clark model for population distribution. The network also has spatial and temporal features with power laws, sometimes with exponential truncations and log-normal distributions. Furthermore, we propose a general analytical method to quantify the predictability of this kind of behavior by calculating the entropy at a collective level, which can be extended to quantify other traffic behaviors. Finally, by considering the traffic congestion factor, we propose a better neural network based model for predicting dwelling time of the ridesourcing behavior. We suggest that the traffic behavior of ridesourcing cars indicates specific non-Markovian characteristics, which can be systematically analyzed from the viewpoint of network sciences.

Deep learning methods, especially convolutional neural networks, have been successfully applied to lesion segmentation in breast ultrasound (BUS) images. However, pattern complexity and intensity similarity between the surrounding tissues (i.e., background) and lesion regions (i.e., foreground) bring challenges for lesion segmentation. Considering that such rich texture information is contained in background, very few methods have tried to explore and exploit background-salient representations for assisting foreground segmentation. Additionally, other characteristics of BUS images, i.e., 1) low-contrast appearance and blurry boundary, and 2) significant shape and position variation of lesions, also increase the difficulty in accurate lesion segmentation. In this paper, we present a saliency-guided morphology-aware U-Net (SMU-Net) for lesion segmentation in BUS images. The SMU-Net is composed of a main network with an additional middle stream and an auxiliary network. Specifically, we first propose generation of saliency maps which incorporate both low-level and high-level image structures, for foreground and background. These saliency maps are then employed to guide the main network and auxiliary network for respectively learning foreground-salient and background-salient representations. Furthermore, we devise an additional middle stream which basically consists of background-assisted fusion, shape-aware, edge-aware and position-aware units. This stream receives the coarse-to-fine representations from the main network and auxiliary network for efficiently fusing the foreground-salient and background-salient features and enhancing the ability of learning morphological information for network. Extensive experiments on five datasets demonstrate higher performance and superior robustness to the scale of dataset than several state-of-the-art deep learning approaches in breast lesion segmentation in ultrasound image.

Automatic modulation classification (AMC) is one of the major challenges for cognitive radio (CR), which can enhance the spectrum utilization efficiency. In this study, a hybrid signal and image-based deep learning model is designed for AMC in CR. A convolutional neural network (CNN) is applied in both the deep learning models. The signal-based CNN (SBCNN) is designed with the optimal filter size for the prediction accuracy, which is used as a pre-training deep learning network to extract features with size 24×1. The features extracted by SBCNN are converted into heat map images, which showed RGB images in the scale range of -30 to +30. Finally, the images are utilized for training and testing the image-based CNN (IBCNN). The dataset used for the experiment is DeepSig:RADIOML2018.01A, which is the latest version. For the IBCNN, the prediction accuracy is 1.96%, 7.99%, and 4.63% higher at signal-to-noise ratio (SNR) 10 dB, and 3.26%, 6.4%, and 4.13% higher at SNR 0 dB as compared to conventional models: ECNN, SCGNet, and LCNN, respectively.

Increasing heatwaves and extreme temperatures have recently been observed across Central Asia (CA). Accurately assessing and projecting the changing climate extremes at the local (station) scale required for climate risk management are therefore highly important. However, global and regional climate models often fail to represent the statistical distributions of observed daily extreme variables and hence extremes in complex terrain. In this work, we developed a statistical downscaling (SD) model to project summer daily maximum temperature (Tmax) and heatwave indices for 65 meteorological stations in CA toward 2100. The SD model involves first-order autoregression and multiple linear regression using large-scale Tmax and circulation indices (CIs) as predictors, and the model is cross-validated against historical observations. The local Tmax and heatwave indices are then projected for 2015–2100 driven by the output of a global climate model (CNRM-CM6-1) under four Shared Socioeconomic Pathways (SSP126, SSP245, SSP370, and SSP585). The application of the SD model significantly improves forecasting of the probability distribution (10th/90th percentiles) of Tmax at stations, particularly across mountainous regions. The model also captures interannual variability and the long-term trend in Tmax, consistent with synoptic-scale inputs. SD projections demonstrate strong warming trends of summer Tmax in CA toward 2100 with rates between 0.35–0.64 °C per decade based on the SSP245 and SSP370 scenarios. Consequently, heatwave occurrence is projected to rise by 1.0–5.0 and 2.0–7.0 d per decade under the SSP245 and SSP370 scenarios, respectively, by 2100. Duration, intensity, and amplitude of heatwaves rise at greater rates under higher-emission scenarios, particularly in southeastern CA. The proposed SD model serves as a useful tool for assessing local climate extremes, which are needed for regional risk management and policymaking for adaption to climate change.

The small spatial and temporal scales at which flash floods occur make predicting events challenging, particularly in data-poor environments where high-resolution weather models may not be available. Additionally, the uptake of warnings may be hampered by difficulties in translating the scientific information to the local context and experiences. Here we use social science methods to characterise local knowledge of flash flooding among vulnerable communities along the flat Lake Malawi shoreline in the district of Karonga, northern Malawi. This is then used to guide a scientific analysis of the factors that contribute to flash floods in the area using contemporary global datasets; including geomorphology, soil and land-use characteristics, and hydro- meteorological conditions. Our results show that communities interviewed have detailed knowledge of the impacts and drivers of flash floods (deforestation, sedimentation), early warning signs (changes in clouds, wind direction and rainfall patterns), and distinct hydro-meteorological processes that lead to flash flood events at the beginning and end of the wet season. Our analysis shows that the scientific data corroborates this knowledge, and that combining local and scientific knowledge provides improved understanding of flash flood processes within the local context. We highlight the potential in linking large-scale global datasets with local knowledge to improve the usability of flash flood warnings.

We present a system to extract architectural assets from large-scale collections of panoramic imagery. We automatically rectify and crop parts of the panoramic image that contain dominant planes, and then use object detection to extract assets such as façades and windows. We also provide various tools to identify attributes of the assets to determine the asset quality and index the assets for search. In addition, we propose a User Interface (UI) to visualize and query assets. Finally, we present applications for urban modeling and texture synthesis.

Nowcasts of strong convective precipitation and radar-based quantitative precipitation estimations have always been hot yet challenging issues in meteorological sciences. Data-driven machine learning, especially deep learning, provides a new technical approach for the quantitative estimation and forecasting of precipitation. A high-quality, large-sample, and labeled training dataset is critical for the successful application of machine-learning technology to a specific field. The present study develops a benchmark dataset that can be applied to machine learning for minute-scale quantitative precipitation estimation and forecasting (QpefBD), containing 231,978 samples of 3,185 heavy precipitation events that occurred from April to October during the years 2016 to 2018 in six provinces in central and eastern China. Each individual sample consists of eight products of weather radars at 6-min intervals within the time window of the corresponding event and products of 27 physical quantities at hourly intervals that describe the atmospheric dynamic and thermodynamic conditions. Two data labels, i.e., ground precipitation intensity and areal coverage of heavy precipitation at 6-minute intervals, are also included. The present study describes the basic components of the dataset and data processing and provides metrics for the evaluation of model performance on precipitation estimation and forecasting. Based on these evaluation metrics, some simple, commonly used methods are applied to evaluate precipitation estimates and forecasts. The results can serve as the benchmark reference for the performance evaluation of machine learning models using this dataset. This paper also gives some suggestions and scenarios of the QpefBD application. We believe that the application of this benchmark dataset will promote interdisciplinary collaboration between meteorological sciences and artificial intelligence sciences, providing a new way for the identification and forecast of heavy precipitation.