The U-Net-like model has been widely studied in the field of building extraction. However, most of these models are based on locally sensed Convolutional Neural Networks(CNNs) designed with symmetric structure and single feature processing, which cannot accurately identify buildings with different sizes, shapes, and colors in remote sensing images. To overcome these problems, we propose the asymmetric cascade fusion network(ACFN), based on the Vision Transformer(ViT), to design a novel asymmetric architecture to recognize buildings of different sizes and shapes by processing multi-granularity features by different means. First, the asymmetric architecture obtains multi-granularity features with global contextual information by embedding different types of attention in encoder-decoders of different sizes. This architecture can identify densely distributed and occluded buildings by semantic reasoning in remote sensing images with complex information. Second, we design a multi-branch weighted pyramid pooling module, which sets different branch weights to offset the background noise introduced in introducing global contextual information. Our ACFN significantly improves the Beijing buildings, ISPRS-Vaihingen, and LoveDA datasets.
Sleep posture is closely related to sleep quality. Moreover, several studies reveal that an incorrect sleep position can result in physical pain. A non-invasive image-based method was proposed for identifying ten sleep postures with high accuracy. The positions of the legs and arms was considered and more complex but common sleep postures was classified, such as fatal left, yearner left, log left, fatal right, yearner right, log right, soldier down, faller down, soldier up, faller up. Input of depth images were preprocessed and a deep multi-stream convolutional neural network was adopted for classification. The work is available for natural scenarios in which people sleep with blanket or quilt covering. Finally, 22 subjects were participated for recording depth images of 10 types of sleep postures, and efficiency of the network was also evaluated.
RGB-IR cross-modality person re-identification (ReID) can be seen as a multicamera retrieval problem that aims to match pedestrian images captured by visible and infrared cameras. Most of the existing methods focus on reducing modality differences through feature representation learning. However, they ignore the huge difference in pixel space between the two modalities. Unlike these methods, we utilize the pixel and feature alignment network (PFANet) to reduce modal differences in pixel space while aligning features in feature space in this paper. Our model contains three components, including a feature extractor, a generator, and a joint discriminator. Like previous methods, the generator and the joint discriminator are used to generate high-quality cross-modality images; however, we make substantial improvements to the feature extraction module. Firstly, we fuse batch normalization and global attention (BNG) which can pay attention to channel information while conducting information interaction between channels and spaces. Secondly, to alleviate the modal difference in feature space, we propose the modal mitigation module (MMM). Then, by jointly training the entire model, our model is able to not only mitigate the cross-modality and intramodality variations but also learn identity-consistent features. Finally, extensive experimental results show that our model outperforms other methods. On the SYSU-MM01 dataset, our model achieves a rank-1 accuracy of and an mAP of .
8 September 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Change detection (CD) in optical remote sensing images has advanced significantly with the adoption of deep learning. However, CD inherently faces two challenges: (1) varying sizes and shapes of change regions arising from spatial complexity, and (2) pseudo-changes caused by temporal migrations, such as seasonal variations and lighting differences. Existing methods typically focus on single-grained information during the feature interaction stage, which limits their ability to perceive changes in features of varying sizes. Additionally, these methods inadequately address ambiguous regions during the difference capture stage, making them highly susceptible to pseudo-change interference. To address these challenges, we propose a novel network called the Spatio-temporal Multi-granularity Intermingling Network (STMINet). First, we introduce the Spatio-temporal Multi-granularity Interleaving Module (STMIM) to capture multi-granularity information across both time and space, greatly enhancing the detection of changes in features of varying sizes and shapes. Second, we propose the Multi-branch Differential Acquisition (MDA), which incorporates information from inconspicuous regions and mitigates pseudo-change interference through a three-branch design. Experimental results on four publicly available datasets (LEVIR-CD, GZ-CD, WHU-CD, and SYSU-CD) demonstrate that STMINet significantly outperforms state-of-the-art methods in performance metrics and visualization, achieving F1 score improvements of 0.22–3.42% over existing approaches while employing a simple ResNet-18 backbone. The source code will be published at https://github.com/heikeyuhuajia/STMINet.
With the increasing urbanization and the popularity of traffic video surveillance, and the rapid development of object detection algorithms, object detection of traffic events through video analysis has become possible. This paper proposes the design and implementation of a traffic event detection system based on video analysis. Firstly, the video stream is processed, mainly for video access and display, and the compression of neural networks achieves the network acceleration. Secondly, the structured information of vehicles is obtained by nighttime vehicle detection and joint detection and tracking. Finally, the car's driving behavior is analyzed through video calibration and video analysis. The system has been used online in many places in China and has achieved remarkable results.