融合多尺度与边缘特征的水葫芦语义分割方法OA

To address the prevalent issues of multi-scale object recognition confusion and loss of edge details in semantic segmentation tasks for floating objects such as water hyacinth,this paper proposed a semantic segmentation method for water hyacinth based on the fusion of multi-scale and edge features,termed the WH-FFSSM model.A dual-backbone network integrating convolution and attention mechanisms was designed to synchronously capture global multi-scale features and local detailed information of water hyacinth.A dual-branch guided aggregation layer was constructed to achieve effective alignment and synergistic enhancement of heterogeneous features.An edge detail detection layer and dense skip connections within the decoder were introduced to significantly improve the model's perception of target edges and its capability for multi-level feature representation.Comprehensive experimental evaluations on a self-constructed dataset demonstrated that the proposed method achieved a mean intersection over union of 91.5％and a mean accuracy of 95.4％on the test set.These results significantly outperform existing baseline models,verifying the effectiveness and superiority of the proposed method for accurate water hyacinth segmentation in complex water environments.