Picture classification method based on improved SwinTransform

The invention discloses a picture classification method based on an improved SwinTransform, and relates to the technical field of picture classification. The method at least comprises the following steps that S1, a SwinTransform picture classification framework is built, a Transform is built by adopting a sliding window operation and hierarchical construction mode, and then a Swinin Transform model composed of a window multi-head self-attention layer, a sliding window multi-head self-attention layer, a standardization layer and a multi-layer perceptron is formed; the deep neural network SwinTransform model based on the self-attention mechanism has a multi-head self-attention mechanism, feature extraction can be performed through the multi-head self-attention mechanism, global features can be learned by using the self-attention mechanism compared with a CNN model, dependence on external information can be reduced, data or internal correlation of the features can be better captured, and the feature extraction e