KoiU-Net:基于多特征增强融合的纺锤形鱼类图像分割方法OA北大核心CSTPCD

In view of the blurred edges and vague features in the fusiform fish image segmentation task,a multi-feature enhancement and fusion based fusiform fish image segmentation network KoiU-Net is proposed.On the basis of the classical U-Net model,a multi-scale feature cross perception module and a multi-scale feature fusion module are designed to enhance the processing capability of fusiform fish features,so as to cope with the problems of blurred edges and complex features in the segmentation of fusiform fish image.The multi-scale upsampling module is designed to extract finer feature information.Experiments on the fusiform fish image dataset show that the KoiU-Net achieves significant improvement in segmentation accuracy,averaging 98.63%,in comparison with the other segmentation networks such as the original U-Net.The effectiveness of each design module is also verified by ablation experiments,and the multi-scale feature cross perception module plays a key role in improving the segmentation performance.This study provides effective technical support for further implementation of fusiform fish growth state monitoring and lays the foundation for the further development in this field.