Image Recognition Based on Improved Recurrent Attention Network

For the problem of low recognition rate and low efficiency of single feature in fine-grained domains, an improved recurrent attention convolutional neural network is proposed. In this method, the SepRes network is used to replace the VGG-19 network in the original model, which improves the feature extraction ability while reducing the amount of calculation in the entire process and the required storage space; Joint clustering algorithm is added to APN to generate multiple independent feature regions, so that the entire model makes full use of global information and improves the recognition accuracy. Finally, by defining the loss function to cross-train SepRes and APN, the convergence speed is accelerated. The CUB200-2011 data set is used to test and experiment the method in this paper, and the recognition accuracy can be reached. The operation efficiency and recognition rate have been improved.