Underwater image segmentation based on attention mechanism and the U-net model
To improve the effect of underwater image segmentation,this paper proposes an underwater image seg-mentation model that uses a three-branch attention module to enhance the U-net model.Specifically,the atten-tion mechanism achieves cross-channel interactive information,which realized multi-dimensional interaction with-out reducing the dimensionality.The experiments on the VOC2007 and SUIM datasets show that the mIOU value of the proposed method is 72.05 and the mPA value 81.3 on the VOC2007 dataset,which are better than the traditional U-net network with mIOU value 58.74 and mPA value 71.13.The mIOU value is 70.374 and the mPA value 82.838 of the proposed method on SUIM dataset,which is better than the traditional U-net network's mIOU value 68.89 and mPA value 82.51.The proposed method can better perform underwater image segmen-tation.
万方数据
维普
