Search

Ear count serves as a crucial parameter for estimating wheat yield, holding great significance in crop phenotypic parameter calculations, yield predictions and wheat ear density management. To accurately determine the number of wheat ears, this study utilized the Global Wheat Head Detection 2021 dataset, and introduced a wheat ear counting method based on the improved YOLOX model. An attention mechanism was integrated into the feature extraction network to bolster the semantic information of the feature map. This enhancement aimed to improve the capability of effectively detecting wheat ears. By using GIoU loss to optimize the loss function, the positioning accuracy of bounding box was improved. Experimental results show that, compared with other models, the improved YOLOX model has significantly improved the detection effect of different targets. Among them, the precision of the improved YOLOX model is 97.24%, the average precision is 94.15%, the accuracy is 99.76%, and the log average miss detection rate is 0.25. This study provides a reliable method for precision wheat ear identification, offering valuable applications in high-throughput investigation of wheat ear traits. [ABSTRACT FROM AUTHOR]

The unique geographical location of the land-sea transition makes the ecological environment of the Yellow River Delta very fragile and vulnerable to human activities. As one of the characteristics of anthropogenic activities, monitoring the spatiotemporal changes of impervious surface area (ISA) is of great significance to the protection of the ecological environment in the Yellow River Delta (YRD). Based on the Landsat historical images and computing resources provided by Google Earth Engine (GEE), an ISA mapping method was developed through combining spectral, texture features and random forest algorithm, and subsequently was applied to generate the spatiotemporal distribution data of ISA of the YRD for 1992, 1998, 2004, 2010, 2016 and 2021. The experimental results demonstrated that the proposed method achieved satisfactory accuracy, with an average overall accuracy of 92.23% and an average Kappa coefficient of 0.9090. Through further time-series analysis of ISA, it found that the area of ISA in the YRD increased from the initial 394.87 km2 to 1081.74 km2 during study periods, and the annual growth rate broke through new highs, ranging from the initial 1.01 km2/year to 67.87 km2/year. According to the research results, urban development activities in the region should be strictly restricted in order to protect the ecological environment of the YRD. [ABSTRACT FROM AUTHOR]

Landslides are one of the main geological disasters in the riparian zone of the Three Gorges Reservoir Area (TGRA) in China, causing massive losses in terms of industrial and agricultural production and people's lives and property. Due to the difficulty of data collection and lack of facade information, the remote sensing methods currently used for landslide mapping and detection are insufficient. The current landslide information acquisition method is mainly based on visual interpretation and automatic computer classification from aerial orthophotographs, which is subjective and time-consuming and yields low accuracy. New shipborne photogrammetry has unique advantages in small and steep slope landslide detection. In recent years, the deep learning technology represented by convolutional neural networks (CNNs) exhibits more powerful feature learning ability in image recognition than traditional machine learning technology. Meanwhile, transfer learning can make the network suitable for small datasets through fine-tuning, which could transfer the parameters of the pretrained model to the new model in a certain way, thereby increasing the learning efficiency of the model. In this paper, the training of multiple deep learning models is realized through transfer learning, and automatic identification of landslides based on shipborne images is achieved through decision-level fusion. The experimental results show that through the integration of decision-level fusion and transfer learning, it performs well in the classification of landslides based on shipborne images. [ABSTRACT FROM AUTHOR]

Due to their important role in maintaining temperature and soil moisture, agricultural plastic covers have been widely utilized around the globe for improving crop-growing conditions, which include both plastic-covered greenhouses (PCGs) and plastic-mulched farmlands (PMFs). However, it is a challenging and long-neglected issue to separate PCGs from PMFs due to their spectral similarity. The objective of this study is to propose a deep semantic segmentation model for accurate PCG and PMF mapping based on very high-resolution satellite images and to improve the model's spatial generalization capability using a transfer learning strategy. Specifically, the proposed semantic segmentation model has an encoder-decoder structure, where the encoder is composed of a new convolutional neural network for discriminative spatial feature learning, while the decoder utilizes a multi-task strategy to improve the predictions on the boundaries. Meanwhile, a transfer learning framework is adopted to increase mapping performance and generalization ability under limited samples. Experimental results in several typical regions across the Eurasian continent show that the proposed model could separate PCGs from PMFs accurately with a mean overall accuracy of 94.49% and an average mIoU of 0.8377. Ablation studies verify the role of encoder-decoder and transfer learning strategy in improving classification performance. [ABSTRACT FROM AUTHOR]

The urbanization worldwide leads to the rapid increase of solid waste, posing a threat to environment and people's wellbeing. However, it is challenging to detect solid waste sites with high accuracy due to complex landscape, and very few studies considered solid waste mapping across multi-cities and in large areas. To tackle this issue, this study proposes a novel deep learning model for solid waste mapping from very high resolution remote sensing imagery. By integrating a multi-scale dilated convolutional neural network (CNN) and a Swin-Transformer, both local and global features are aggregated. Experiments in China, India and Mexico indicate that the proposed model achieves high performance with an average accuracy of 90.62%. The novelty lies in the fusion of CNN and Transformer for solid waste mapping in multi-cities without the need for pixel-wise labelled data. Future work would consider more sophisticated methods such as semantic segmentation for fine-grained solid waste classification. [ABSTRACT FROM AUTHOR]

During clinical practice, radiologists often use attributes, e.g., morphological and appearance characteristics of a lesion, to aid disease diagnosis. Effectively modeling attributes as well as all relationships involving attributes could boost the generalization ability and verifiability of medical image diagnosis algorithms. In this paper, we introduce a hybrid neuro-probabilistic reasoning algorithm for verifiable attribute-based medical image diagnosis. There are two parallel branches in our hybrid algorithm, a Bayesian network branch performing probabilistic causal relationship reasoning and a graph convolutional network branch performing more generic relational modeling and reasoning using a feature representation. Tight coupling between these two branches is achieved via a cross-network attention mechanism and the fusion of their classification results. We have successfully applied our hybrid reasoning algorithm to two challenging medical image diagnosis tasks. On the LIDC-IDRI benchmark dataset for benign-malignant classification of pulmonary nodules in CT images, our method achieves a new state-of-the-art accuracy of 95.36% and an AUC of 96.54%. Our method also achieves a 3.24% accuracy improvement on an in-house chest X-ray image dataset for tuberculosis diagnosis. Our ablation study indicates that our hybrid algorithm achieves a much better generalization performance than a pure neural network architecture under very limited training data. [ABSTRACT FROM AUTHOR]

Accurate and timely crop yield prediction over large spatial regions is critical to national food security and sustainable agricultural development. However, designing a robust model for crop yield prediction over a large spatial region remains challenging due to inadequate surveyed samples and an under-development of deep-learning frameworks. To tackle this issue, we integrated multi-source (remote sensing, weather, and soil properties) data into a dual-stream deep-learning neural network model for winter wheat in China's major planting regions. The model consists of two branches for robust feature learning: one for sequential data (remote sensing and weather series data) and the other for statical data (soil properties). The extracted features by both branches were aggregated through an adaptive fusion model to forecast the final wheat yield. We trained and tested the model by using official county-level statistics of historical winter wheat yields. The model achieved an average R2 of 0.79 and a root-mean-square error of 650.21 kg/ha, superior to the compared methods and outperforming traditional machine-learning methods. The dual-stream deep-learning neural network model provided decent in-season yield prediction, with an error of about 13% compared to official statistics about two months before harvest. By effectively extracting and aggregating features from multi-source datasets, the new approach provides a practical approach to predicting winter wheat yields at the county scale over large spatial regions. [ABSTRACT FROM AUTHOR]

Grassland aboveground biomass is crucial for evaluating grassland desertification, degradation, and grassland and livestock balance. Given the lack of understanding of mechanical processes and limited simulation accuracy for grassland aboveground biomass estimation, especially at the regional scale, this study investigates a new method combining remote sensing data assimilation technology and a grassland process-based model to estimate regional grassland biomass, focusing on improving the simulation accuracy by modeling and revealing the mechanism interpretability of grassland growth processes. Xilinhot City of Inner Mongolia was used as the study area. The ModVege model was selected as the grass dynamic simulation model. A likelihood function was constructed composed of the LAI, grassland aboveground biomass, and daily measurements wherein the accumulated temperature reached ST2 (the temperature sum defining the end of reproductive growth). Then, the Markov chain Monte Carlo (MCMC) methodology was adapted to calibrate the ModVege model by maximizing the likelihood function. The time-series LAI from MOD15A3H was assimilated into the ModVege model, and the model parameters ST2 and BMGV0 (initial biomass and green vegetative tissues, respectively) were optimized at a 500 m pixel scale based on the four-dimensional variational method (4DVar) method. Compared with August 15th, the RMSE and MAPE of aboveground biomass were 242 kg/ha and 10%, respectively, after calibration. Data assimilation improved this accuracy, with the RMSE decreasing to 214 kg/ha. Overall, the aboveground grassland biomass of Xilinhot City shows spatial distribution patterns of high value in the northeast and low value in the central and southeast areas. Generally, the method implemented in this study provides an important reference for the aboveground biomass estimation of regional grassland. [ABSTRACT FROM AUTHOR]

Semi-arid regions belong to important ecological transitional zones with a high ecosystem vulnerability and sensitivity. In recent years, due to the influence of climate change and human activities, the land covers of semi-arid regions have experienced dramatic changes. Therefore, timely and accurate land cover mapping of semi-arid regions is of great significance. Deep learning has been a hotspot in land cover mapping; however, it requires a large number of labelled samples. To tackle this data-hunger issue, we resolve to a multi-gate semi-supervised learning (SSL) method that could use a huge number of unlabelled samples to improve the classification performance. Specifically, the proposed method consists of a probability gate, an uncertainty gate and a consistency gate, aiming to generate high-quality pseudo-labels from the unlabelled datasets. Experimental results from Zhangbei, China indicate that the proposed method yields a good performance with an overall accuracy of 90.31%, which is 4.31% higher than that of traditional supervised learning (SL) methods. Ablation studies also verify the effectiveness of the gated mechanism in selecting unlabelled data. [ABSTRACT FROM AUTHOR]

Wheat is a major staple food crop in China. Accurate and cost-effective wheat mapping is exceedingly critical for food production management, food security warnings, and food trade policy-making in China. To reduce confusion between wheat and non-wheat crops for accurate growth stage wheat mapping, we present a novel approach that combines a random forest (RF) classifier with multi-sensor and multi-temporal image data. This study aims to (1) determine whether an RF combined with multi-sensor and multi-temporal imagery can achieve accurate winter wheat mapping, (2) to find out whether the proposed approach can provide improved performance over the traditional classifiers, and (3) examine the feasibility of deriving reliable estimates of winter wheat-growing areas from medium-resolution remotely sensed data. Winter wheat mapping experiments were conducted in Boxing County. The experimental results suggest that the proposed method can achieve good performance, with an overall accuracy of 92.9% and a kappa coefficient (κ) of 0.858. The winter wheat acreage was estimated at 33,895.71 ha with a relative error of only 9.3%. The effectiveness and feasibility of the proposed approach has been evaluated through comparison with other image classification methods. We conclude that the proposed approach can provide accurate delineation of winter wheat areas. [ABSTRACT FROM AUTHOR]

Virtual geographic environments (VGEs) are extensively used to explore the relationship between humans and environments. Crowd simulation provides a method for VGEs to represent crowd behaviors that are observed in the real world. The social force model (SFM) can simulate interactions among individuals, but it has not sufficiently accounted for inter-group and intra-group behaviors which are important components of crowd dynamics. We present the social group force model (SGFM), based on an extended SFM, to simulate group behaviors in VGEs with focuses on the avoiding behaviors among different social groups and the coordinate behaviors among subgroups that belong to one social group. In our model, psychological repulsions between social groups make them avoid with the whole group and group members can stick together as much as possible; while social groups are separated into several subgroups, the rear subgroups try to catch up and keep the whole group cohesive. We compare the simulation results of the SGFM with the extended SFM and the phenomena in videos. Then we discuss the function of Virtual Reality (VR) in crowd simulation visualization. The results indicate that the SGFM can enhance social group behaviors in crowd dynamics. [ABSTRACT FROM AUTHOR]

This paper presents an approach to process raw unmanned aircraft vehicle (UAV) image-derived point clouds for automatically detecting, segmenting and regularizing buildings of complex urban landscapes. For regularizing, we mean the extraction of the building footprints with precise position and details. In the first step, vegetation points were extracted using a support vector machine (SVM) classifier based on vegetation indexes calculated from color information, then the traditional hierarchical stripping classification method was applied to classify and segment individual buildings. In the second step, we first determined the building boundary points with a modified convex hull algorithm. Then, we further segmented these points such that each point was assigned to a fitting line using a line growing algorithm. Then, two mutually perpendicular directions of each individual building were determined through a W-k-means clustering algorithm which used the slop information and principal direction constraints. Eventually, the building edges were regularized to form the final building footprints. Qualitative and quantitative measures were used to evaluate the performance of the proposed approach by comparing the digitized results from ortho images. [ABSTRACT FROM AUTHOR]

As an important ecosystem, wetlands play a crucial role in both regional and global environments. Accurate land-cover classification can facilitate the management and understanding of wetlands. Considering the timely and cost-effective characteristics of remote sensing, this technique was used to obtain land-cover information for the Yellow River Delta (YRD) wetland in this investigation. Landsat-8 Operational Land Imager (OLI) sensor data were selected for the data set in this study. A combined approach of multiple end-member spectral mixture analysis (MESMA) and Random Forest (RF) was developed for land-cover classification mapping of the YRD wetland. This study aimed (1) to determine whether the MESMA technique in combination with RF significantly improves the accuracy of classification in complex landscapes such as the YRD wetland, (2) to determine whether the RF classifier shows good performance in land-cover classification of the YRD wetland, and (3) to compare the proposed method with the traditional Maximum Likelihood Classifier (MLC). The proposed hybrid method showed good performance, with an overall accuracy of 89.5% and a kappa coefficient (κ) of 0.88. The inclusion of fractional information derived from MESMA can improve the classification accuracy by 2–3%. In addition, through a comparison with traditional maximum likelihood (ML) methodology, the effectiveness of the proposed approach was evaluated. Overall, the proposed approach in this study can relatively accurately delineate a land-cover classification map of the YRD wetland with Landsat-8 OLI remotely sensed data. [ABSTRACT FROM PUBLISHER]

Accurate estimation of phytoplankton chlorophyll-a(chl-a) concentration from remote sensing data is challenging due to the complex optical properties of case II waters. Recently, a novel semi-analytical four-band model was developed to estimate chl-aconcentration in turbid productive waters. The objective of this study was to evaluate the performance of the four-band model and extend its application to hyperspectral satellite data for estimating chl-aconcentration in Qiandao Lake of China. Based on field spectral measurements and in situ water sampling, the four-band model expressed as [Rrs−1(661.6) –Rrs−1(706.7)] [Rrs−1(714.8) –Rrs−1(682.2)]−1was calibrated after band tuning, whereRrs−1represents the reciprocal of the remote sensing reflectance. The spectral-based four-band model accounted for more than 88% of variance in chl-aconcentration with a root mean square error (RMSE) of 1.47 μg l−1. To justify the potential of this model with satellite data, comparable wavelengths selected from HJ-1A Hyperspectral Imager (HSI) imagery were utilized to calibrate the four-band model. The HSI-based model explained about 80% of chl-avariation with an RMSE of 1.35 μg l−1. Experimental results also showed that the four-band model outperformed its three-band counterpart. The results validated the rationale of the four-band model and demonstrated the effectiveness of this model for estimating chl-aconcentration from both in situ spectral data and HJ-1A hyperspectral satellite imagery. [ABSTRACT FROM AUTHOR]

Grasslands, as an important part of terrestrial ecosystems, are facing serious threats of land degradation. Therefore, the remote monitoring of grasslands is an important tool to control degradation and protect grasslands. However, the existing methods are often disturbed by clouds and fog, which makes it difficult to achieve all-weather and all-time grassland remote sensing monitoring. Synthetic aperture radar (SAR) data can penetrate clouds, which is helpful for solving this problem. In this study, we verified the advantages of the fusion of multi-spectral (MS) and SAR data for improving classification accuracy, especially for cloud-covered areas. We also proposed an adaptive feature fusion method (the SK-like method) based on an attention mechanism, and tested two types of patch construction strategies, single-size and multi-size patches. Experiments have shown that the proposed SK-like method with single-size patches obtains the best results, with 93.12% accuracy and a 0.91 average f1-score, which is a 1.02% accuracy improvement and a 0.01 average f1-score improvement compared with the commonly used feature concatenation method. Our results show that the all-weather, all-time remote sensing monitoring of grassland is possible through the fusion of MS and SAR data with suitable feature fusion methods, which will effectively enhance the regulatory capability of grassland resources. [ABSTRACT FROM AUTHOR]

Agricultural greenhouse (AG), one of the fastest-growing technology-based approaches worldwide in terms of controlling the environmental conditions of crops, plays an essential role in food production, resource conservation and the rural economy, but has also caused environmental and socio-economic problems due to policy promotion and market demand. Therefore, long-term monitoring of AG is of utmost importance for the sustainable management of protected agriculture, and previous efforts have verified the effectiveness of remote sensing-based techniques for mono-temporal AG mapping in a relatively small area. However, currently, a continuous annual AG remote sensing-based dataset at large-scale is generally unavailable. In this study, an annual AG mapping method oriented to the provincial area and long-term period was developed to produce the first Landsat-derived annual AG dataset in Shandong province, China from 1989 to 2018 on the Google Earth Engine (GEE) platform. The mapping window for each year was selected based on the vegetation growth and the phenological information, which was critical in distinguishing AG from other misclassified categories. Classification for each year was carried out initially based on the random forest classifier after the feature optimization. A temporal consistency correction algorithm based on classification probability was then proposed to the classified AG maps for further improvement. Finally, the average User's Accuracy, Producer's Accuracy and F1-score of AG based on visually-interpreted samples over 30 years reached 96.56%, 86.64% and 0.911, respectively. Furthermore, we also found that the ranked features via calculating the importance of each tested feature resulted in the highest accuracy and the strongest stability in the initial classification stage, and the proposed temporal consistency correction algorithm improved the final products by approximately five percent on average. In general, the resultant AG sequence dataset from our study has revealed the expansion of this typical object of "Human–Nature" interaction in agriculture and has a potential application in use of greenhouse-related technology and the scientific planning of protected agriculture. [ABSTRACT FROM AUTHOR]

High-quality training samples are essential for accurate land cover classification. Due to the difficulties in collecting a large number of training samples, it is of great significance to collect a high-quality sample dataset with a limited sample size but effective sample distribution. In this paper, we proposed an object-oriented sampling approach by segmenting image blocks expanded from systematically distributed seeds (object-oriented sampling approach) and carried out a rigorous comparison of seven sampling strategies, including random sampling, systematic sampling, stratified sampling (stratified sampling with the strata of land cover classes based on classification product, Latin hypercube sampling, and spatial Latin hypercube sampling), object-oriented sampling, and manual sampling, to explore the impact of training sample distribution on the accuracy of land cover classification when the samples are limited. Five study areas from different climate zones were selected along the China–Mongolia border. Our research identified the proposed object-oriented sampling approach as the first-choice sampling strategy in collecting training samples. This approach improved the diversity and completeness of the training sample set. Stratified sampling with strata defined by the combination of different attributes and stratified sampling with the strata of land cover classes had their limitations, and they performed well in specific situations when we have enough prior knowledge or high-accuracy product. Manual sampling was greatly influenced by the experience of interpreters. All these sampling strategies mentioned above outperformed random sampling and systematic sampling in this study. The results indicate that the sampling strategies of training datasets do have great impacts on the land cover classification accuracies when the sample size is limited. This paper will provide guidance for efficient training sample collection to increase classification accuracies. [ABSTRACT FROM AUTHOR]

Obtaining water body images quickly and reliably is important to guide human production activities and study urban change. This paper presents a fast and accurate method to identify water bodies in complex environments based on UAV high-resolution images. First, an improved U-Net (SU-Net) model is proposed in this paper. By increasing the number of connections in the middle layer of the neural network, more image features can be retained through S-shaped circular connections. Second, aiming at the interference of mixed ground objects and dark ground objects on water detection, the fusion of a deep learning network and visual features is investigated. We analyse the influence of a wavelet transform and grey level cooccurrence matrix (GLCM) on water extraction. Using a confusion matrix to evaluate accuracy, the following conclusions are drawn: (1) Compared with existing methods, the SU-Net method achieves a significant improvement in accuracy, and the overall accuracy (OA) is 96.25%. The kappa coefficient (KC) is 0.952. (2) SU-Net combined with the GLCM has a higher accuracy (OA is 97.4%) and robustness in distinguishing mixed and dark objects. Based on this method, a distinct water boundary in urban areas, which provides data for urban water vector mapping, can be obtained. [ABSTRACT FROM AUTHOR]

Sentinel-2 images have been widely used in studying land surface phenomena and processes, but they inevitably suffer from cloud contamination. To solve this critical optical data availability issue, it is ideal to fuse Sentinel-1 and Sentinel-2 images to create fused, cloud-free Sentinel-2-like images for facilitating land surface applications. In this paper, we propose a new data fusion model, the Multi-channels Conditional Generative Adversarial Network (MCcGAN), based on the conditional generative adversarial network, which is able to convert images from Domain A to Domain B. With the model, we were able to generate fused, cloud-free Sentinel-2-like images for a target date by using a pair of reference Sentinel-1/Sentinel-2 images and target-date Sentinel-1 images as inputs. In order to demonstrate the superiority of our method, we also compared it with other state-of-the-art methods using the same data. To make the evaluation more objective and reliable, we calculated the root-mean-square-error (RSME), R 2 , Kling–Gupta efficiency (KGE), structural similarity index (SSIM), spectral angle mapper (SAM), and peak signal-to-noise ratio (PSNR) of the simulated Sentinel-2 images generated by different methods. The results show that the simulated Sentinel-2 images generated by the MCcGAN have a higher quality and accuracy than those produced via the previous methods. [ABSTRACT FROM AUTHOR]

With the rapid process of both urban sprawl and urban renewal, large numbers of old buildings have been demolished in China, leading to wide spread construction sites, which could cause severe dust contamination. To alleviate the accompanied dust pollution, green plastic mulch has been widely used by local governments of China. Therefore, timely and accurate mapping of urban green plastic covered regions is of great significance to both urban environmental management and the understanding of urban growth status. However, the complex spatial patterns of the urban landscape make it challenging to accurately identify these areas of green plastic cover. To tackle this issue, we propose a deep semi-supervised learning framework for green plastic cover mapping using very high resolution (VHR) remote sensing imagery. Specifically, a multi-scale deformable convolution neural network (CNN) was exploited to learn representative and discriminative features under complex urban landscapes. Afterwards, a semi-supervised learning strategy was proposed to integrate the limited labeled data and massive unlabeled data for model co-training. Experimental results indicate that the proposed method could accurately identify green plastic-covered regions in Jinan with an overall accuracy (OA) of 91.63%. An ablation study indicated that, compared with supervised learning, the semi-supervised learning strategy in this study could increase the OA by 6.38%. Moreover, the multi-scale deformable CNN outperforms several classic CNN models in the computer vision field. The proposed method is the first attempt to map urban green plastic-covered regions based on deep learning, which could serve as a baseline and useful reference for future research. [ABSTRACT FROM AUTHOR]

Timely and accurate forecasting of crop yields is crucial to food security and sustainable development in the agricultural sector. However, winter wheat yield estimation and forecasting on a regional scale still remains challenging. In this study, we established a two-branch deep learning model to predict winter wheat yield in the main producing regions of China at the county level. The first branch of the model was constructed based on the Long Short-Term Memory (LSTM) networks with inputs from meteorological and remote sensing data. Another branch was constructed using Convolution Neural Networks (CNN) to model static soil features. The model was then trained using the detrended statistical yield data during 1982 to 2015 and evaluated by leave-one-year-out-validation. The evaluation results showed a promising performance of the model with the overall R 2 and RMSE of 0.77 and 721 kg/ha, respectively. We further conducted yield prediction and uncertainty analysis based on the two-branch model and obtained the forecast accuracy in one month prior to harvest of 0.75 and 732 kg/ha. Results also showed that while yield detrending could potentially introduce higher uncertainty, it had the advantage of improving the model performance in yield prediction. [ABSTRACT FROM AUTHOR]

Vegetable mapping from remote sensing imagery is important for precision agricultural activities such as automated pesticide spraying. Multi-temporal unmanned aerial vehicle (UAV) data has the merits of both very high spatial resolution and useful phenological information, which shows great potential for accurate vegetable classification, especially under complex and fragmented agricultural landscapes. In this study, an attention-based recurrent convolutional neural network (ARCNN) has been proposed for accurate vegetable mapping from multi-temporal UAV red-green-blue (RGB) imagery. The proposed model firstly utilizes a multi-scale deformable CNN to learn and extract rich spatial features from UAV data. Afterwards, the extracted features are fed into an attention-based recurrent neural network (RNN), from which the sequential dependency between multi-temporal features could be established. Finally, the aggregated spatial-temporal features are used to predict the vegetable category. Experimental results show that the proposed ARCNN yields a high performance with an overall accuracy of 92.80%. When compared with mono-temporal classification, the incorporation of multi-temporal UAV imagery could significantly boost the accuracy by 24.49% on average, which justifies the hypothesis that the low spectral resolution of RGB imagery could be compensated by the inclusion of multi-temporal observations. In addition, the attention-based RNN in this study outperforms other feature fusion methods such as feature-stacking. The deformable convolution operation also yields higher classification accuracy than that of a standard convolution unit. Results demonstrate that the ARCNN could provide an effective way for extracting and aggregating discriminative spatial-temporal features for vegetable mapping from multi-temporal UAV RGB imagery. [ABSTRACT FROM AUTHOR]

The greenhouse is the fastest growing food production approach and has become the symbol of protected agriculture with the development of agricultural modernization. Previous studies have verified the effectiveness of remote sensing techniques for mono-temporal greenhouse mapping. In practice, long-term monitoring of greenhouse from remote sensing data is vital for the sustainable management of protected agriculture and existing studies have been limited in understanding its spatiotemporal dynamics. This study aimed to generate multi-temporal greenhouse maps in a typical protected agricultural region (Shouguang region, north China) from 1990 to 2018 using Landsat imagery and the Google Earth Engine and quantify its spatiotemporal dynamics that occur as a consequence of the development of protected agriculture in the study area. The multi-temporal greenhouse maps were produced using random forest supervised classification at seven-time intervals, and the overall accuracy of the results greater than 90%. The total area of greenhouses in the study area expanded by 1061.94 km 2 from 1990 to 2018, with the largest growth occurring in 1995–2010. And a large number of increased greenhouses occurred in 10–35 km northwest and 0–5 km primary roads buffer zones. Differential change trajectories between the total area and number of patches of greenhouses were revealed using global change metrics. Results of five landscape metrics showed that various landscape patterns occurred in both spatial and temporal aspects. According to the value of landscape expansion index in each period, the growth mode of greenhouses was from outlying to edge-expansion and then gradually changed to infilling. Spatial heterogeneity, which measured by Shannon's entropy, of the increased greenhouses was different between the global and local levels. These results demonstrated the advantage of utilizing Landsat imagery and Google Earth Engine for monitoring the development of greenhouses in a long-term period and provided a more intuitive perspective to understand the process of this special agricultural production approach than relevant social science studies. [ABSTRACT FROM AUTHOR]

Large-scale crop mapping provides important information in agricultural applications. However, it is a challenging task due to the inconsistent availability of remote sensing data caused by the irregular time series and limited coverage of the images, together with the low spatial resolution of the classification results. In this study, we proposed a new efficient method based on grids to address the inconsistent availability of the high-medium resolution images for large-scale crop classification. First, we proposed a method to block the remote sensing data into grids to solve the problem of temporal inconsistency. Then, a parallel computing technique was introduced to improve the calculation efficiency on the grid scale. Experiments were designed to evaluate the applicability of this method for different high-medium spatial resolution remote sensing images and different machine learning algorithms and to compare the results with the widely used nonparallel method. The computational experiments showed that the proposed method was successful at identifying large-scale crop distribution using common high-medium resolution remote sensing images (GF-1 WFV images and Sentinel-2) and common machine learning classifiers (the random forest algorithm and support vector machine). Finally, we mapped the croplands in Heilongjiang Province in 2015, 2016, 2017, which used a random forest classifier with the time series GF-1 WFV images spectral features, the enhanced vegetation index (EVI) and normalized difference water index (NDWI). Ultimately, the accuracy was assessed using a confusion matrix. The results showed that the classification accuracy reached 88%, 82%, and 85% in 2015, 2016, and 2017, respectively. In addition, with the help of parallel computing, the calculation speed was significantly improved by at least seven-fold. This indicates that using the grid framework to block the data for classification is feasible for crop mapping in large areas and has great application potential in the future. [ABSTRACT FROM AUTHOR]

Coastal land cover classification is a significant yet challenging task in remote sensing because of the complex and fragmented nature of coastal landscapes. However, availability of multitemporal and multisensor remote sensing data provides opportunities to improve classification accuracy. Meanwhile, rapid development of deep learning has achieved astonishing results in computer vision tasks and has also been a popular topic in the field of remote sensing. Nevertheless, designing an effective and concise deep learning model for coastal land cover classification remains problematic. To tackle this issue, we propose a multibranch convolutional neural network (MBCNN) for the fusion of multitemporal and multisensor Sentinel data to improve coastal land cover classification accuracy. The proposed model leverages a series of deformable convolutional neural networks to extract representative features from a single-source dataset. Extracted features are aggregated through an adaptive feature fusion module to predict final land cover categories. Experimental results indicate that the proposed MBCNN shows good performance, with an overall accuracy of 93.78% and a Kappa coefficient of 0.9297. Inclusion of multitemporal data improves accuracy by an average of 6.85%, while multisensor data contributes to 3.24% of accuracy increase. Additionally, the featured fusion module in this study also increases accuracy by about 2% when compared with the feature-stacking method. Results demonstrate that the proposed method can effectively mine and fuse multitemporal and multisource Sentinel data, which improves coastal land cover classification accuracy. [ABSTRACT FROM AUTHOR]

Accurate urban land-use mapping is a challenging task in the remote-sensing field. With the availability of diverse remote sensors, synthetic use and integration of multisource data provides an opportunity for improving urban land-use classification accuracy. Neural networks for Deep Learning have achieved very promising results in computer-vision tasks, such as image classification and object detection. However, the problem of designing an effective deep-learning model for the fusion of multisource remote-sensing data still remains. To tackle this issue, this paper proposes a modified two-branch convolutional neural network for the adaptive fusion of hyperspectral imagery (HSI) and Light Detection and Ranging (LiDAR) data. Specifically, the proposed model consists of a HSI branch and a LiDAR branch, sharing the same network structure to reduce the time cost of network design. A residual block is utilized in each branch to extract hierarchical, parallel, and multiscale features. An adaptive-feature fusion module is proposed to integrate HSI and LiDAR features in a more reasonable and natural way (based on "Squeeze-and-Excitation Networks"). Experiments indicate that the proposed two-branch network shows good performance, with an overall accuracy of almost 92%. Compared with single-source data, the introduction of multisource data improves accuracy by at least 8%. The adaptive fusion model can also increase classification accuracy by more than 3% when compared with the feature-stacking method (simple concatenation). The results demonstrate that the proposed network can effectively extract and fuse features for a better urban land-use mapping accuracy. [ABSTRACT FROM AUTHOR]