Search

Unmanned aerial vehicle (UAV) remote sensing images used for semantic segmentation possess distinct features compared to urban street scene images, including high resolution and a complex background. Spatial information plays a pivotal role in enhancing the performance of semantic segmentation for high-resolution images. The dual-branch architecture for semantic segmentation incorporates supplementary branches to capture spatial information. However, prior research on dual-branch semantic segmentation neglected the interaction between the contextual and spatial branches, leading to suboptimal model performance. In this discourse, the article introduces a dual-branch semantic segmentation framework. This design advances the system's understanding of spatial information while facilitating inter-branch learning through two key modules. Initially, the spatial calibration feature extraction module employs frequency domain processing and learning tactics distinct from the contextual approach to generate image features under varied noise conditions. Calibration is achieved by generating features from diverse angles. Subsequently, the spatially-guided loss function directs the acquisition of spatial information for the spatial branch by condensing the deep image characteristics for the context branch. To assess the generalization capacity of the proposed method, experiments will be conducted on three different datasets. The proposed method's modules will be integrated into three representative dual-branch networks, allowing assessment of the generalization capacity of the key DBCG components. Empirical evidence demonstrates that this approach is highly effective, significantly surpassing the performance of the baseline network.

Deep-learning-based on convolutional neural networks (CNN) has been widely applied in synthetic aperture radar (SAR) target recognition and made significant progress. However, due to the physical effects of the equipment used to collect images, various degrees of speckle noise will be introduced into SAR images. Traditional CNN-based SAR target recognition methods are premised on the same noise intensity in the training and testing set, which is contrary to the target recognition in practice. To alleviate this problem, we propose a novel speckle noise resistant framework for SAR target recognition, called dual-consistency-alignment-based self-supervised learning. First, original SAR images are randomly added to speckle noise with different thresholds through multiplicative noise, after which contrastive pretraining is performed on unlabeled data. During this period, we combine instance pseudolabel consistency alignment and feature consistency alignment to align multiple threshold speckle noise views with original views under the same targets. Finally, the pretrained model is migrated to the downstream SAR speckle noise target recognition task. In this article, speckle noise modeling is conducted based on moving and stationary target capture and recognition data testing set, and experiment results reveal that this method can adapt to different intensities of speckle noise, is robust to modeled SAR image recognition, and maintains a high recognition rate even in small-sample learning.