Image Retrieval Algorithm Based on Deep Learning

The traditional hashing method of manual feature extraction uses image tags as the supervision information to obtain the loss function, and the retrieval accuracy is low and the effect is not good. This paper proposes a new deep learning image retrieval algorithm based on the traditional supervised hash algorithm. The algorithm integrates feature learning and hash code learning in an end-to-end framework, and converts multi-labels of images into binary paired labels. Based on the AlexNet framework, a feature learning module is established, and a pair of loss function and a balanced hash code loss function are combined to generate a loss function for network training. After the experimental test of the CIFAR-10 data set, the method of this paper greatly improves the average accuracy of image retrieval.