Research And Application Of Co-salient Object Detection Algorithm

The purpose of synergy and significant target detection is to detect significant objects that appear in a set of images at the same time.In the past,most methods will first summarize the consensus knowledge of the entire group,and then search for the corresponding objects in each image.To sum up the consensus knowledge of the entire group,the lack of robustness,scalability or stability,and simply integrating consensus characteristics and image characteristics in the search process of the entire image will lead to The synergy characteristics are not accurate enough.Therefore,in order to improve the consensus robustness,scalability and stability of the entire image group,this article proposes a significant and significant target detection method based on the semantic perception of semantics between images.It can explore the common semantics of the entire image group in an efficient way Essence In addition,for the collaborative features of the collaborative objects and the complex scenes,the synergistic features of images cannot be effectively expressed.This article is designed with a consensus cross-paying attention network.The network can better cope with challenges and detect more comprehensive and more accurate synergy and significant goals.For the problems of synergy and significant target detection,the specific work of this article is as follows:(1)Semantic-aware co-salient object detection method based on imagesA deep network framework is proposed to tap the semantic correlation between different image groups.The adopted network is mainly composed of two parts: semantic perception synergy module and auxiliary classification fusion module.First use the expansion convolution to extract and enhance the features from the image.Subsequently,the semantic synergy module was used to synergistic object recognition.The final auxiliary classification fusion module is applied to the integration consensus characteristics and multi-scale features.The main network of the module extracts outstanding features in a top-down manner,aiming to comprehensively search for the consistency of the image group.Compared with multiple classic deep learning network models on three data sets on the three data sets,the algorithm shows superior performance.(2)Consensus cross attention network for co-salient object detectionIn order to more accurately detect the obscure objects,a novel consensus cross-paying attention has been proposed.It effectively uses global information to judge consistency and introduce three modules.Specifically,the multi-scale feature aggregation module can effectively extract and enhance the features from the image group to prepare for subsequent processing.Consensus cross-attention module is used to enhance the prominent area in training,and it pays more attention to the entire image dependence.Finally,the similarity aggregation module is applied to the consensus features and multi-scale features.A large number of experiments were performed on the three challenging datasets,and this method has achieved significant results.