Research on Salient Object Detection Based on Attention Nested UNet
Salient object detection has wide applications in image recognition and is a popular area of research in computer vi-sion.Improper fusion of high-level and low-level features can lead to incomplete information extraction;during the downsampling process,the pooling process is prone to issues such as key feature loss and noise feature amplification.This article proposes a nested UNet network incorporating attention mechanism—(AU)2 Net.This method introduces an attention mechanism in the RSU module and constructs an attention mechanism Attention RSU block to ensure that on the basis of rich feature information extraction,it fo-cuses on salient regions,ignores irrelevant information,and suppresses the influence of noise.After comparing the maximum F-mea-sure and MAE of 9 other advanced algorithms on 5 different datasets,it is shown that they outperform other algorithms in most indi-cators.The method proposed in this article can effectively capture target information in complex scenarios and outline the specific details of the target in a more complete manner.
万方数据
维普
