A Fish Image Segmentation Method Based on Image Enhancement and U2-Net
Mingjie Zhu, Wei Liao, Zhen Cheng Xu · 2022
Underwater target monitoring is of great importance for the aquaculture industry, and using deep learning algorithms to assist in monitoring is a feasible and efficient approach. This paper implements a fish segmentation method in low-contrast, color-biased underwater environments using Auto-MSRCR image enhancement and U2-Net semantic segmentation algorithm. The experimental results show that this method can accurately segment the mask information of underwater fish and reduce the impact caused by the underwater environment. Compared with the model without image enhancement, the accuracy of the model trained by this method is improved by about 5%.