The research and development of high-precision and low-cost apple intelligent grading technology is the core issue to extend the apple industrial chain and improve the quality and efficiency of the fruit industry. In order to solve the problems of low accuracy and weak robustness of traditional computer vision technology in apple external quality classification, an apple appearance classification method based on deep learning (multiple convolutional neural network DXNet model) was proposed. Firstly, totally 15000 apple images covering different appearance levels were taken in Yan’an supermarkets, orchards and other places, and then labeled manually. A database of apple images with extensive coverage of external quality information and large sample size was established. Then, on the basis of comparing and analyzing the classical convolution network model, the classical model was optimized and improved by the method of model fusion, and the convolution part of the classical model was extracted and fused to be the feature extractor, and the fully connected layer was shared to be the classifier, batch normalization and regularization techniques were used to prevent the model from over fitting. Totally 15000 images were used for training and 4500 images were used for testing. The results showed that the classification accuracy of the improved DXNet model was higher than that of the classical model, and the classification accuracy reached 97.84%, the validity of the method applied to apple external quality classification was verified.