Automatic Extraction Method of Aquaculture Sea Based on Improved SegNet Model

Abstract

Timely, accurate, and efficient extraction of aquaculture sea is important for the scientific and rational utilization of marine resources and protection of the marine environment. To improve the classification accuracy of remote sensing of aquaculture seas, this study proposes an automatic extraction method for aquaculture seas based on the improved SegNet model. This method adds a pyramid convolution module and a convolutional block attention module based on the SegNet network model, which can effectively increase the utilization ability of features and capture more global image information. Taking the Gaofen-1D image as an example, the effectiveness of the improved method was proven through ablation experiments on the two modules. The prediction results of the proposed method were compared with those of the U-Net, SegNet, and DenseNet models, as well as with those of the traditional support vector machine and random forest methods. The results showed that the improved model has a stronger generalization ability and higher extraction accuracy. The overall accuracy, mean intersection over union, and F1 score of the three test areas were 94.86%, 87.23%, and 96.59%, respectively. The accuracy of the method is significantly higher than those of the other methods, which proves the effectiveness of the method for the extraction of aquaculture seas and provides new technical support for automatic extraction of such areas.