Research On Fine-grained Image Retrieval Based On Attention Mechanism And Deep Metric Learning

In recent years,fine-grained images have become a research hotspot of computer vision.Through the comprehensive analysis of the current mainstream fine-grained image retrieval methods,finding the detailed features of the image and improving the feature discrimination are the two focuses of many research methods.This paper proposes solutions to the two focal problems of fine-grained image retrieval.First,to obtain the detailed features of the image,this paper proposes a feature extraction framework based on the attention mechanism.By introducing a selective kernel attention module,the feature collection network can focus on the effective area of the image.At the same time,this paper also designs a penalty-aware memory loss function,through deep metric learning to improve the discriminative power of features.The two main innovations of this article are summarized as follows:1.Aiming at the problem of obtaining detailed features,this paper proposes a fine-grained image retrieval framework based on a selective kernel attention mechanism.The framework first acquires features at multiple scales by fusing the features of different levels of networks,then learns subtle feature representations through the selective kernel attention mechanism,and then uses generalized mean pooling to complete feature aggregation,and uses N elements The group center loss function realizes the learning of the model.2.Aiming at the problem of feature discriminative improvement,this paper designs a method based on the penalty-aware memory loss function.This method introduces a memory bank to store the category center and updates the memory bank according to the accumulation of local information in each batch.Using the category center in the memory bank to construct a triplet,and propose a penalty-aware memory module to complete the construction of a new loss function and metric space.To verify the method’s effectiveness,this article conducts a full range of experiments on four widely used classic fine-grained retrieval data sets and compares them with many current classic fine-grained image retrieval methods.The experimental results show that the method proposed is indeed correct and effective,showing good performance.