Tobacco leaf baking state identification method based on fine granularity classification

The invention discloses a tobacco leaf curing state identification method based on fine-grained classification, and the method comprises the steps: collecting tobacco leaf curing images, and building a tobacco leaf curing image data set; preprocessing the tobacco leaf baking image data set; performing feature extraction on the preprocessed tobacco leaf baking image to generate a feature map; fusing the feature maps; performing linear mapping processing on the fused feature map to obtain a feature sequence; the feature sequences are sent to N encoder layers, attention weights of the first N-1 layers and output of the N-1 layers are recorded, and the feature sequences input in the Nth layer are screened; inputting the screened features into the Nth encoder layer, and inputting the output into a classifier to classify the input tobacco leaf baking images; completing the training of a tobacco leaf curing state recognition model by comparing a loss function; and inputting a to-be-classified tobacco leaf baking ima