Image classification method based on multi-granularity Transform

The invention relates to the technical field of computer vision, in particular to an image classification method based on multi-granularity Transform. The method comprises the following steps: step 1, sending a picture into a progressive marking module, and marking the picture as a picture vector through stacked convolutional layers of various sizes; 2, sending the picture vectors to a token aggregation module of a first stage, and adjusting the number and dimension of the picture vectors by using convolution operation; step 3, sending the adjusted picture vector into a multi-granularity Transform module of the stage, and extracting a feature vector of the picture; 4, repeating the step 2 and the step 3, and obtaining a final output vector from the multi-granularity Transform module of the last stage; and step 5, sending the final output vector into a classification unit to obtain the category of the picture, and completing image classification. According to the method, the calculation complexity of the Trans