Collaborative optimization of channel attention and weighted dice loss for remote sensing image water extraction algorithm

The extraction of river, lake, and reservoir water bodies from high-resolution visible light remote sensing satellite images in complex scenes has always been a major research problem in the field of remote sensing images. Due to factors such as water bodies, terrain, and human activities, rivers in different regions have differences in seasonal runoff and direction, as well as factors such as lighting angles and cloud cover during image generation, making water extraction difficult. This article proposes a network structure SE ResUnet for water extraction tasks to address this issue. The main contributions of this article are threefold: (1) Introducing SE attention mechanism with residual structure in the downsampling process, and learning Unet's skip structure for multi-scale feature extraction and compensation, thereby enhancing the feature extraction ability of water bodies such as rivers, lakes, and reservoirs. (2) Introducing a weighted dice loss function to balance positive and negative samples and enhance the generalization of the model. (3) In the comparative experiments of improving the Unet model with semantic segmentation networks such as Unet, PSPNet, Deeplabv3+on self built dataset and remote sensing interpretation public dataset, Excellent performance and results were achieved on mIoU, OA, and F1score. On the self built dataset, compared with Unet, mIoU, OA, and F1score improved by 0.38%, 0.12%, and 0.08%, respectively; On the publicly available dataset of Kaggle remote sensing interpretation for water extraction, mIoU, OA, and F1score improved by 0.63%, 0.26%, and 0.25%, respectively. The experimental results demonstrate that the strategy combining attention mechanism and weighted loss function has a significant effect on the effectiveness of collaborative improvement of neural network models in water extraction tasks.