Abstract: In view of the problem that traditional citrus classification methods mainly rely on manual extraction of features,which are complex and inefficient,and difficult to achieve efficient recognition in the food industry environment,an improved residual network citrus classification method based on attention mechanism is proposed.In this study, attention mechanism is added on the basis of residual network（ResNet34）,which increases the weight of useful information, weakens the influence of irrelevant information,and improves the expression ability of network model and the classification ability of model.The experimental results show that the classification accuracy of the residual network based on attention mechanism for healthy and defective citrus reaches 99.02%,which is relatively improved compared with the original ResNet34 model.The residual network with attention mechanism has better feature extraction ability for citrus surface defects, and can extract more citrus defect feature information.This study helps to improve the productivity of citrus industry and provides a reference for citrus defect identification.