关键词: 苹果病害叶片, 图像分类, 迁移学习, SimAM注意力机制, ResNet18

Abstract: Apple disease leaf classification and recognition are significant for disease monitoring and control improvement in apple plantations. Aiming at the problem of apple disease leaf classification and recognition，this paper proposed a ResNet model based on the SimAM attention mechanism. The model combined the SimAM attention module，Swish activation function，and entropy weight Focal Loss function through migration learning and data enhancement operations to improve the accurate recognition of apple disease leaves with uneven sample distribution. Experimental results showed that the improved SimAM ResNet18 model achieved an accuracy of 94. 68% on the test set，which was a 2.89% improvement compared to the benchmark network ResNet18. Compared to other classical convolutional classification models， AlexNet， VGG16， and GoogLeNet， the model accuracy improved by 7. 02%，5. 25%，and 4.31%. The results show that the ResNet model based on the SimAM attention mechanism has a high potential for apple disease leaf classification and recognition with uneven sample distribution.

Key words: apple disease leaf, image classification, transfer learning, SimAM attention mechanism, ResNet18