The extraction of small water bodies in the Yellow River Basin has always been a key issue of concern in the fields of remote sensing technology application, water resource management, environmental science, and geographic information systems. Due to factors such as water bodies, human activities, and cloud cover, water body extraction becomes difficult. In addition, convolutional neural networks are prone to losing small water body feature information during the process of extracting local features, which can cause more imbalance between positive and negative samples of water bodies and non-water bodies. In response to these issues, this study focused on a specific research area—the middle and lower reaches of the Yellow River. We processed and analyzed high-resolution optical satellite images collected from the Yellow River Basin and other areas, with a particular emphasis on precise identification of small water bodies, and proposed a network structure, the SE-Attention-Residual-Unet (SE-ResUnet), for water extraction tasks.The main contributions of this article are threefold: (1) Introducing a channel attention mechanism with residual structure in the down-sampling process, and learning Unet’s skipping structure for multi-scale feature extraction and compensation, thereby enhancing the feature extraction ability of small water bodies, including rivers, lakes, and reservoirs. (2) Introducing a weighted-Dice (W-Dice) loss function to balance positive and negative samples and enhance the generalization of the model. (3) In comparative experiments on improving the Unet model with semantic segmentation networks such as Unet, PSPNet, Deeplabv3+ on a self-built dataset and remote sensing interpretation public dataset, excellent performance and results were achieved on the mIoU, OA, and F1-score metrics. On the self-built dataset, compared with Unet, the mIoU, OA, and F1-score improved by 0.38%, 0.12%, and 0.08%, respectively. On the publicly available dataset, for remote sensing interpretation of water extraction, the mIoU, OA, and F1-score improved by 0.63%, 0.26%, and 0.25%, respectively. The experimental results demonstrate that a strategy combining an attention mechanism and a weighted loss function has a significant effect on the effectiveness of the collaborative improvement of neural network models in water extraction tasks.

Affected by complex backgrounds and multi-scale object characteristics, object detection in remote sensing images faces significant challenges in accuracy. Despite advancements in the methods utilizing convolutional neural networks (CNN) and self-attention, they encounter two fundamental challenges: CNNs are restricted by their limited receptive fields, giving rise to inadequate global feature representation, whereas self-attention mechanisms, while adept at capturing long-range dependencies, suffer from heightened computational complexity that hampers practical application efficiency and may diminish the representation of local detail features. To resolve these challenges, this article proposed an innovative CNN-Mamba fusion-based detection model —MambaRetinanet— which uses a well-designed synergistic perception module (SPM) to efficiently model the global information and enhance the extraction of local features. In addition, for improving the feature pyramid network (FPN), we introduced a differentiated feature processing strategy and designed an asymmetric feature pyramid—MambFPN—based on this strategy to balance detection accuracy and computational efficiency. The experimental results indicate that MambaRetinanet has significant advantages on four mainstream remote sensing datasets: the mean Average precision (mAP) on DOTA-v1.0, DOTA-v1.5, DOTA-v2.0 and DIOR-R datasets reached 77.50, 70.21, 57.17 and 71.50 respectively, which is an average increase of 11% in comparison to that of the baseline. Notably, on the DOTA-v2.0 dataset, MambaRetinanet demonstrates advantages over the current one stage SOTA model, enhancing mAP scores by approximately 2 percentage points, thereby validating the efficacy and generalizability of the MambaRetinaNet in complex remote sensing scenarios.