Multiple Defect Classification Method for Green Plum Surfaces Based on Vision Transformer

Green plums have produced significant economic benefits because of their nutritional and medicinal value. However, green are affected by factors such as plant diseases insect pests during growth, picking, transportation, storage, which seriously affect the quality products, reducing At present, in detection plum defects, some researchers applied deep learning to identify surface defects. recognition rate is not high, types defects identified singular, classification detailed enough. In actual production process, often more than one defect, existing methods ignore minor Therefore, this study used vision transformer network model all on surfaces plums. The dataset was classified into multiple based four (scars, flaws, rain spots, rot) type feature (stem). After permutation combination these a total 18 categories were obtained after screening, combined with situation. Based VIT model, fine-grained defect link added for analysis layer major hazard level secondary improved has an average accuracy 96.21% plums, better that VGG16 network, Desnet121 Resnet18 WideResNet50 network.