Abstract: Mangrove, being an important part of the ecological environment system, plays an important role in protecting and purifying the environment. The survival environment for mangroves has been seriously devastated due to overexploitation. Therefore, it is very important to monitor mangrove growth. In response to the poor performance of deep learning models in extracting mangrove information from remote sensing images, this paper combined the advantages of Shuffle Transformer and convolutional neural networks, and added ASPP Embedding module to extract feature information and skip attention fusion deep and shallow features to extract mangrove information from remote sensing images. Results show that the extraction precision with the presented model reaches 97.64%, which is 1.38 percentage points higher than that of U-Net network structure. Experimental results show that this method has great advantages in mangrove remote sensing image information extraction.