A DeepLabv3+architecture improved method for identifying mangrove boundaries in high-resolution remote sensing images
In this study,aiming at the monitoring and protection of mangroves,an improved scheme based on the DeepLabv3+model is proposed to enhance the recognition accuracy of mangroves in high-resolution remote sensing images.The improvement mainly involves introducing depthwise sepa-rable convolution and SE(Squeeze and Excitation)attention mechanism into the ASPP(Atrous Spa-tial Pyramid Pooling)structure of DeepLabv3+,as well as incorporating a CBAM(Convolutional Block Attention Module)attention mechanism and multi-scale fusion technique into the decoding end.These innovative structural designs strengthen the model's ability to capture and represent key features of mangroves,thereby reducing missed detections and false positives.After rigorous accuracy evalua-tion,the improved DeepLabv3+model achieved an overall accuracy of 99.60％,with recall,man-grove Intersection over Union(Mangrove-IoU),and mangrove Fl-score reaching 96.05％,95.31％,and 97.60％,respectively.Compared with the original DeepLabv3+model and other popular models such as HRNet and PSPNet,the improved model demonstrated superior performance in all major eval-uation metrics,significantly improving the recognition accuracy and boundary extraction capability of mangroves.Furthermore,the model's generalization ability and application potential were also verified in the application analysis.The application analysis further verified the generalization ability and ap-plication potential of the model.The research results can optimize the real-time monitoring technology of mangroves.
NETL
NSTL
万方数据
