Research On Fine-Grained Image Retrievalalgorithm Based On Deep Learning Feature Extraction

Image retrieval is an important research direction in the field of computer vision.With the introduction of deep learning,the way of using convolutional neural networks to extract features to represent images has received widespread attention.For fine-grained images,the fact that intra-class differences are greater than the inter-class differences is the core problem of retrieval,and obtain more discriminative features is a more challenging content.Based on deep learning and feature extraction,this paper conducts research on fine-grained image retrieval.The main work is as follows.Firstly,to address the problem that the semantic information of the output feature maps of convolutional neural networks is usually high-dimensional and weakly descriptive,this paper proposes a weighted local region hashing algorithm for fine-grained image retrieval.The output feature map of Res Net18 is cross-dimensionally weighted to enhance its descriptiveness,and the weighted feature map is divided into local regions,and features are calculated based on the activation points within it.The hash network is utilized to learn low-dimensional binary centers and hash codes that are compact intra-class and separated inter-class,and to improve retrieval efficiency by minimising the overall loss.The experiments validated the effectiveness of the method on four fine-grained benchmark datasets,such as CUB-200-2011,and the results showed an improvement in retrieval accuracy over all eight baseline methods such as Exch Net.Secondly,the regions containing key information can be judged by the gradients of the convolution kernel parameters and weights.Therefore,a fine-grained image retrieval method based on the aggregation of convolutional kernel parameter gradients feature is designed.Using the output feature map of Res Net50 to design a loss function and perform back propagation,calculate the aggregated features by the weights of the convolution kernels of a specific convolution layer,and fuse the multilayer vectors into a final representation of the image for retrieval.The experimental results show that this method has better retrieval performance.Finally,in order to improve the effectiveness of unsupervised fine-grained image retrieval,a fine-grained image retrieval method is studied based on weighted-map target localization feature method.Cross dimensional weighting is applied to the VGG-16 output feature map,then use the weighted map to calculate the highly activated pixels points to obtain the maximum target localization area,and convert and fuse it into the most robust feature representation of querying images.The loss function is used to calculate the channel weight of the weighted map and the active map,and the visualization saliency maps is obtained through linear combination.The experimental results demonstrate that this method can improve the accuracy of fine-grained image retrieval.