This is the authors' abstract. We don't add key points for this paper.
Remote sensing images of coastal aquaculture areas present complex challenges due to intricate textures and spectral characteristics, often resulting in classification confusion and omission errors. Additionally, variations in water coloration within aquaculture ponds, caused by sediment, algal blooms, and changes in water quality, further complicate the ability of segmentation algorithms to accurately distinguish aquaculture areas from surrounding water bodies. To address these challenges, this paper presents an enhanced SegFormer network that incorporates feature enhancement modules and integrates atrous spatial pyramid pooling with advanced feature upsampling mechanisms. This integration allows for the extraction of deep semantic features with global context, while simultaneously restoring fine details. Furthermore, we introduce a Multi-Dimensional Feature Fusion Module (MFFM) equipped with a Channel Attention Mechanism. The MFFM effectively preserves image features with large receptive fields and high-level semantics through hierarchical feature fusion, combining these with local detail information. Experimental results show that our method achieved 93.60% Acc, 78.3% MIoU, and 80.9% Kappa, outperforming mainstream networks, such as FCN, DeepLabv3+, and HRNet in seawater aquaculture area extraction.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Li et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: