Research on land and sea segmentation of remote sensing images based on improved UNET

This paper proposes a semantic segmentation model based on UNet, which aims to solve the problem of land and sea segmentation in remote sensing images under complex environments. Among the most advanced segmentation methods, semantic segmentation networks show great potential in this task, but there is still room for improvement. This paper proposes a sea-land segmentation method based on the UNet model. On the basis of the Unet model, the FPN structure is introduced to better fuse features at different scales; multi-parallel void spatial pyramid pooling is introduced to improve the ability of the model to comprehensively capture contextual information at different scales under the complex background of sea-land segmentation; a composite loss function is designed to calculate the loss in the model to improve the performance and accuracy of the sea-land segmentation model. This paper tests the network model on the collected sea-land dataset. The results show that the method in this paper has achieved 87.62%, 88.72%, 88.06%, and 86.24%inthe evaluation indicators of Precision, Recall, F1 score, and Intersection-over-Union ratio, respectively, which are 7.27%, 8.04%, 7.01%, and 4.62% higher than the initial Unet model. The segmentation performance is better than Unet, FCN, Deeplab and other classic semantic segmentation networks.