Marine RAFT Aquaculture Type Identification from Polarimetric SAR Image Based on Dual-Branch Hybrid Encoding Network, Dbhe-Net

Marine raft aquaculture plays a significant role in promoting economic development and ensuring national food security. However, its rapid expansion has led to issues such as marine environmental pollution and illegal use of sea areas. SAR (Synthetic Aperture Radar) imagery, with its all-weather, all-day observation capabilities, provides a new data source for monitoring marine raft aquaculture. However, there are issues of incomplete raft aquaculture type recognition and underutilization of polarized SAR information in current research. Consequently, this paper proposed a segmentation method named dual-branch hybrid encoding network (DBHE-Net) based on GF-3 full-polarization SAR image, which effectively integrated CNN's local perception and Transformer's global expression capabilities for raft aquaculture type recognition. Experiment results demonstrated that DBHE-Net outperforms other classical semantic segmentation models, with a mIoU accuracy of 79.46%. The introduction of the Transformer branch significantly improved the extraction accuracy of fully floating raft aquaculture. This paper holds significant importance for the rational planning of aquaculture areas and the scientific management of coastal ecological environments.