Existing methods fail to produce region-complete and boundary-clear prediction im-age in camouflaged object detection due to subtle texture differences and low boundary distinction be-tween camouflaged object and backgrounds.To address this issue,a hierarchical feature fusion-guided camouflaged object detection network is proposed.Firstly,the original features of camouflaged object are fully extracted using a residual network to avoid feature missing.Then,high-level features with rich semantic information guide the fusion of low-level features in a top-down manner,and an atten-tion mechanism is introduced to give more weight to critical features.Finally,region-complete and boundary-clear prediction images are generated by using discriminative features of camouflaged ob-jects.Experimental performance on three benchmark datasets surpasses 8 existing methods,and the generated prediction images are closer to the ground truth.
万方数据
维普
