The invention provides a red date quality sorting method based on deep learning. The method comprises the steps that a large quantity of red date samples with different quality are collected, the samples are divided into four types, namely plump dates, dry bar dates, cracked dates and blemished dates, and the samples are divided into a training set and a test set; collected red date sample imagesare preprocessed, wherein image processing operation, such as graying, binaryzation, median filtering and ROI area-of-interest extraction, is performed on the red date images on a white background, the red dates in the large white background are picked out, and the sizes of the sample images need to be normalized; and a network with an improved structure is used to train the sample training set toobtain a model, and the model generated through training is utilized to classify the test sample images. Compared with existing red date quality sorting technologies, through the method, the accuracyand speed of red date sorting are improved, and the problems that manual sorting of red dates is low in efficiency and accuracy and consumes excessive manpower resources are successfully solved.