Search

As military technology continues to evolve and the amount of situational information available on the battlefield continues to increase, data-driven deep learning methods are becoming the primary method for air target intention recognition. Deep learning is based on a large amount of high quality data; however, in the field of intention recognition, it often faces key problems such as low data volume and unbalanced datasets due to insufficient real-world scenarios. To address these problems, we propose a new method called time-series conditional generative adversarial network with improved Hausdorff distance (IH-TCGAN). The innovation of the method is mainly reflected in three aspects: (1) Use of a transverter to map real and synthetic data into the same manifold so that they have the same intrinsic dimension; (2) Addition of a restorer and a classifier in the network structure to ensure that the model can generate high-quality multiclass temporal data; (3) An improved Hausdorff distance is proposed that can measure the time order differences between multivariate time-series data and make the generated results more reasonable. We conduct experiments using two time-series datasets, evaluate the results using various performance metrics, and visualize the results using visualization techniques. The experimental results show that IH-TCGAN is able to generate synthetic data similar to the real data and has significant advantages in the generation of time series data. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have proved to achieve state-of-the-art performance in content-based remote sensing image retrieval (CBRSIR). However, CNNs cannot focus on discriminative features of important objects, resulting in unsatisfactory retrieval performance with complex backgrounds and small objects. We therefore propose a multi-attention fusion network with dilated convolution and label smoothing for CBRSIR. First, a dilated convolutional layer is used to replace the fifth convolutional layer in the network to obtain a large receptive field. Then, a contextual transformer attention (CoT) and an efficient channel attention (ECA) are fused together for spatial-wise and channel-wise discriminative features, respectively. The multi-attention module is embedded between the newly added dilated convolutional layer and the followed average pooling layer. Besides, in order to enhance the differences between the discriminative features of those correct and incorrect classes, label smoothing is used to replace the cross-entropy loss function. Some ablation experiments are conducted on six benchmark datasets. Compared to the baseline AlextNet with the above different modules, the mAP values of the proposed network improve by 14.84% to 24.21%. The results indicate that our network can significantly improve the retrieval performance. In addition, we have also conducted some experiments for network migration and comparison with some recent methods (e.g. a triplet deep metric learning and deep feature learning with latent relationship embedding network). Experimental results illustrate that our network can be effectively migrated to other similar CNN models and can achieve state-of-the-art or competitive results. [ABSTRACT FROM AUTHOR]

Culture plating is worldwide accepted as the gold standard for quantifying viable foodborne pathogens. However, it is time-consuming (1–2 days) and requires specialized laboratory and personnel. This study reported a deep learning enhanced digital microfluidic platform for multiplex detection of viable foodborne pathogens. The new method used a Time-Lapse images driven EfficientNet-Transformer Network (TLENTNet) to type and quantify the bacteria through spatiotemporal features of bacterial growth and digital enumeration of bacterial culture. First, the bacterial sample was prepared with LB medium and injected into a pre-vacuumed microfluidic chip with an array of 800 microwells to encapsulate at most one bacterium in each well. Then, a programmed sliding microscopic platform was used to scan all microwells every 15 min, capturing time-lapse images of bacterial growth within each microwell. Finally, the TLENTNet was used to facilitate bacterial typing and quantification. Under optimal conditions, this platform was able to detect four bacterial species (S. typhimurium, E. coli O157:H7 , S. aureus and B. cereus) with an average accuracy of 97.72% and a detection limit of 63 CFU/mL in 7 h. [ABSTRACT FROM AUTHOR]

Benchmark datasets play an important role in evaluating remote sensing image retrieval (RSIR) methods. The current datasets cover many scene categories, but omit an important scene of color steel sheds, which are widely distributed with a large number on the earth's surface. Therefore, we propose a new benchmark dataset of color steel sheds (CSS) from Google map imagery for RSIR and share it open access in our V-RSIR system. The new dataset has rich intra-class and inter-class diversity, and is composed of blue, red and white color steel sheds with the total number of 2407 remote sensing images. We conduct evaluation experiments on the new dataset by using ten low/mid feature-based and ten deep learning feature-based methods. Experimental results indicate that the dataset is effective for evaluating RSIR methods and using the dataset can construct an effective retrieval model for color steel sheds. Besides, we have experimentally demonstrated that color constancy does affect retrieval performance on our CSS dataset. Furthermore, some experiments of merging the CSS dataset with the PatternNet, VGoogle and NWPU45 datasets are also conducted. Experimental results demonstrate that our dataset can be used as a complement to other retrieval datasets. Furthermore, these experimental results can be used as baseline for future applications on RSIR. [ABSTRACT FROM AUTHOR]

Accurate and efficient predictions of pollutants in the atmosphere provide a reliable basis for the scientific management of atmospheric pollution. This study develops a model that combines an attention mechanism, convolutional neural network (CNN), and long short-term memory (LSTM) unit to predict the O 3 and PM 2.5 levels in the atmosphere, as well as an air quality index (AQI). The prediction results given by the proposed model are compared with those from CNN-LSTM and LSTM models as well as random forest and support vector regression models. The proposed model achieves a correlation coefficient between the predicted and observed values of more than 0.90, outperforming the other four models. The model errors are also consistently lower when using the proposed approach. Sobol-based sensitivity analysis is applied to identify the variables that make the greatest contribution to the model prediction results. Taking the COVID-19 outbreak as the time boundary, we find some homology in the interactions among the pollutants and meteorological factors in the atmosphere during different periods. Solar irradiance is the most important factor for O 3 , CO is the most important factor for PM 2.5 , and particulate matter has the most significant effect on AQI. The key influencing factors are the same over the whole phase and before the COVID-19 outbreak, indicating that the impact of COVID-19 restrictions on AQI gradually stabilized. Removing variables that contribute the least to the prediction results without affecting the model prediction performance improves the modeling efficiency and reduces the computational costs. [Display omitted] • CNN coupled with LSTM and attention mechanism for atmospheric predictions. • Proposal outperforms other models, with R2 > 0.9 between predicted and observed data. • Sensitivity analysis identifies factors contributing most to pollutant levels. • Removing variables that contribute little to output improves modeling efficiency. • Key factors remain similar during the whole period and before COVID-19 outbreak. [ABSTRACT FROM AUTHOR]

Parameter-server frameworks play an important role in scaling-up distributed deep learning algorithms. However, the constant growth of neural network size has led to a serious bottleneck on exchanging parameters across machines. Recent efforts rely on manually setting a parameter-exchanging interval to reduce communication overhead, regardless of the parameter-server's resource availability as well. It may face poor performance or inaccurate results for inappropriate interval. Meanwhile, request burst may occur, exacerbating the bottleneck. In this paper, we propose an approach to automatically set the optimal exchanging interval, aiming to remove the parameter-exchanging bottleneck and to evenly utilize resources without losing training accuracy. The key idea is to increase the interval on different training nodes on the basis of the knowledge of available resources and choose different intervals for each slave node to avoid request bursts. We adopted this method to optimize the parallel Stochastic Gradient Descent algorithm, through which we successfully sped up parameter-exchanging process by eight times. [ABSTRACT FROM AUTHOR]

Background: Digital histopathology provides valuable information for clinical decision-making. We hypothesized that a deep risk network (DeepRisk) based on digital pathology signature (DPS) derived from whole-slide images could improve the prognostic value of the tumor, node, and metastasis (TNM) staging system and offer chemotherapeutic benefits for gastric cancer (GC). Methods: DeepRisk is a multi-scale, attention-based learning model developed on 1120 GCs in the Zhongshan dataset and validated with two external datasets. Then, we assessed its association with prognosis and treatment response. The multi-omics analysis and multiplex Immunohistochemistry were conducted to evaluate the potential pathogenesis and spatial immune contexture underlying DPS. Results: Multivariate analysis indicated that the DPS was an independent prognosticator with a better C-index (0.84 for overall survival and 0.71 for disease-free survival). Patients with low-DPS after neoadjuvant chemotherapy responded favorably to treatment. Spatial analysis indicated that exhausted immune clusters and increased infiltration of CD11b+CD11c+ immune cells were present at the invasive margin of high-DPS group. Multi-omics data from the Cancer Genome Atlas-Stomach adenocarcinoma (TCGA-STAD) hint at the relevance of DPS to myeloid derived suppressor cells infiltration and immune suppression. Conclusion: DeepRisk network is a reliable tool that enhances prognostic value of TNM staging and aid in precise treatment, providing insights into the underlying pathogenic mechanisms. [ABSTRACT FROM AUTHOR]

Modeling-based prediction methods enable rapid, reagent-free air pollution detection based on inexpensive multi-source data than traditional chemical reaction-based detection methods in order to quickly understand the air pollution situation. In this study, a convolutional neural network (CNN) and long and short-term memory (LSTM) neural networks are integrated to create a CNN-LSTM time series prediction model to predict the concentration of PM 2.5 and its chemical components (i.e., heavy metals, carbon component, and water-soluble ions) using meteorological data and air pollutants (PM 2.5 , SO 2 , NO 2 , CO, and O 3). In the integrated CNN-LSTM model, the CNN uses convolutional and pooling layers to extract features from the data, whereas the powerful nonlinear mapping and learning capabilities of LSTM enable the time series prediction of air pollution. The experimental results showed that the CNN-LSTM exhibited good generalization ability in the prediction of As, Cd, Cr, Cu, Ni, and Zn, with a mean R2 above 0.9. Mean R2 predicted for PM 2.5 , Pb, Ti, EC, OC, SO 4 2−, and NO 3 − ranged from 0.85 to 0.9. Shapley value showed that PM 2.5 , NO 2 , SO 2 , and CO had a greater influence on the predicted heavy metal results of the model. Regarding water-soluble ions, the predicted results were dominantly influenced by PM 2.5 , CO, and humidity. The prediction of the carbon fraction was affected mainly by the PM 2.5 concentration. Additionally, several input variables for various components were eliminated without affecting the prediction accuracy of the model, with R2 between 0.70 and 0.84, thereby maximizing modeling efficiency and lowering operational costs. The fully trained model prediction results showed that most predicted components of PM 2.5 were lower during January to March 2020 than those in 2018 and 2019. This study provides insight into improving the accuracy of modeling-based detection methods and promotes the development of integrated air pollution monitoring toward a more sustainable direction. [Display omitted] • CNN coupled with LSTM for atmospheric time series predictions. • Many elements had better simulation effects when atmospheric pollutants as inputs. • CNN-LSTM has superior prediction performance with more than 90% accuracy. • The Shapely value method identifies the key indicators for predicting pollutant. • Predicted pollutant contents were lower during COVID-19 outbreak than in 2019. [ABSTRACT FROM AUTHOR]

Aiming at the three main problems of power system—leakage, high temperature and physical damage, a new image fusion of fault detection method in power system based on deep learning is proposed in this paper. The core of deep learning is achieved by capsule network model. The model is trained and tested by self-built image dataset of power system. There are three types of dataset: visible images,infrared images and ultraviolet images. After being preprocessed and feature-extracted, the visible image is used as the fusion image background, the infrared image provides the thermal information of power equipment, and the ultraviolet image provides the electric field information on the exterior of power equipment. The collected images are decomposed into corresponding high frequency component image and low frequency component image respectively, which reconstructed into fused images by the capsule network model. With the registration of the three types of images, the faults in the power system can be detected and displayed accurately in the fused image. [ABSTRACT FROM AUTHOR]

Image restoration is essentially recognized as an ill-posed problem. A promising solution in recent years is incorporating deep network-driven priors into the iterative restoration procedure as constrained conditions. Among them, deep mean-shift prior utilizes the denoising autoencoder to play the role of prior updating. In this study, we present multiple wavelets guided deep mean-shift prior, which integrates the advantages of structural representation in the wavelet transform and learning ability in deep network. Specifically, by re-arranging the multi-view and multi-resolution features generated by multiple wavelet transforms as the input of denoising autoencoder, a more powerful prior information is learned. It benefits from the recurrent structure-preserving and multi-view complementary aggregation properties. We embed the learned prior information into the iterative recovery process and adopt proximal gradient descent to tackle it. Extensive experiments on image deblurring and compressed sensing tasks demonstrated significantly improved performances both visually and quantitatively. [Display omitted] • The model inherits structural representation in wavelet and learning ability in CNN. • Multi-channel features generated by multi-wavelets is rearranged as the network input. • It involves the recurrent structure-preserving and multi-view aggregation properties. [ABSTRACT FROM AUTHOR]

• A light learning framework with fewer training parameters is proposed. • LiNet can work directly with raw temporal signals. • LiNet can realize quick, accurate, efficient rotating machinery fault diagnosis. • Domain adaptation ability is improved by combining LiNet with AdaBN. • Structural advantages of LiNet is verified in Discussion. Many recent studies on deep learning models have focused on increasing accuracy for mechanical fault data sets, while disregarding the influences of model complexity on calculation efficiency and training time, resulting in the late detection of fault types. To recognize fault types quickly and accurately, and improve the efficiency of fault diagnosis, this study proposes an improved fault diagnosis method based on CNN for rotating machineries, called the light neural network (LiNet), with fewer parameters. The method uses raw vibration signals as input. Feature extraction of LiNet is realized by one convolutional block with 3 × 1 convolution kernels, two light_modules consisting of uneven-sized convolution blocks, residual structures, and one convolutional block with 1 × 1 convolution kernels. The classification stage identifies fault types through global average pooling. Finally, adaptive batch normalization (AdaBN) is combined with LiNet to explore the domain adaptation ability of the method. Experiments were carried out using two mechanical datasets, the Southeast University (SEU) gearbox dataset and Case Western Reserve University (CWRU) bearing dataset, to verify the effectiveness of the LiNet. Compared with existing methods, the proposed method is more accurate and achieves nearly 100% accuracy with normal signals while maintaining good performance under different working loads. [ABSTRACT FROM AUTHOR]