Vein contraction and venous compression typically caused by low temperature and excessive placement pressure can blur the captured finger vein images and severely impair the quality of extracted features. To improve the quality of captured finger vein image, this paper proposes a 26-layer generator network constrained by Neighbors-based Binary Patterns (NBP) texture loss to recover the clear image (guessing the original clear image). Firstly, by analyzing various types and degrees of blurred finger vein images captured in real application scenarios, a method to mathematically model the local and global blurriness using a pair of defocused and mean blur kernels is proposed. By iteratively and alternatively convoluting clear images with both kernels in a multi-scale window, a polymorphic blur training set is constructed for network training. Then, NBP texture loss is used for training the generator to enhance the deblurring ability of the network on images. Lastly, a novel network structure is proposed to retain more vein texture feature information, and two residual connections are added on both sides of the residual module of the 26-layer generator network to prevent degradation and overfitting. Theoretical analysis and simulation results show that the proposed neighbors-based binary-GAN (NB-GAN) can achieve better deblurring performance than the the-state-of-the-art approaches.

Remote sensing change detection (RSCD) models must be lightweight to be suitable for deployment on edge devices. However, existing methods face challenges in simultaneously achieving high accuracy and fast computation. To address these issues, we propose the Adaptive Separation Complementary Fusion Network (ASCF-Net) based on the MobileNetV2 backbone, incorporating three primary modules. First, we develop an Asymmetric Feature Enhancement Module (AFEM) based on the strong-weak separation principle. Specifically, standard convolutions are applied to information-rich strong components derived from aggregated features from one upper and two lower layers, while depthwise separable convolutions process weak components with lower information content. This design enhances both feature representation and computational efficiency. Second, we present a Multi-Complementary Fusion Module (MCFM) that integrates complementary information from multiple domains, including scale, spatial structure, and semantic levels. This module employs cascaded convolutions with varying dilation rates to expand receptive fields, thereby facilitating the capture of multi-scale change regions. Spatial information is further aggregated, local neighborhoods within the difference map are enlarged through a downsampling–convolution–upsampling operation, and temporal feature channels are randomly exchanged to enhance semantic change perception. All these strategies are adaptively fused using a channel-wise average intensity attention mechanism. Finally, we introduce a Boundary Enhancement Decoder (BED) that incorporates a boundary-aware enhancement module and integrates the Laplacian operator into the original multilevel feature aggregation decoder. This approach significantly improves change boundary localization accuracy by enhancing edge features of change regions. Experiments conducted on five challenging RSCD datasets demonstrate that ASCF-Net achieves state-of-the-art performance among lightweight models, with only 3.89M parameters and 2.34G floating-point operations (FLOPs). ASCF-Net achieves Intersection over Union (IoU) scores of 94.78% on CDD-CD, 88.52% on WHU-CD, 79.74% on GVLM-CD, 73.03% on SYSU-CD, and 64.34% on UAV-CD, respectively. Compared with the best-performing prior lightweight method on each respective dataset, our model achieves an average improvement of 2.23% across all five datasets, with a maximum gain of 3.89%. The code is available at https://github.com/vccccx/ASCF-Net.