Apple Leaf Disease Diagnosis Based on Knowledge Distillation and Attention Mechanism

Accurately diagnosing apple leaf diseases can reduce the use of pesticides and improve the quality of fruits, which is of significance to smart agriculture. Convolutional neural network as a deep learning model is widely used in the field of intelligent diagnosis of apple leaf diseases. Deploying a deep neural network for apple disease diagnosis to mobile devices allows for smarter, more efficient, and more accurate disease identification. However, classical convolutional neural networks have some limitations on agricultural disease diagnosis, such as a huge number of parameters, heavy computation, and long inference time. Thus, such a complex deep learning model is not easily deployed to mobile devices. To address the above problems, we propose the ECA-KDNet, an improved lightweight model based on the ECA attention mechanism and knowledge distillation, which shows superiority in accuracy, robustness, and lightweight. The experimental results show that compared with the classical convolutional neural network models, ECA-KDNet improves accuracy (98.28%) while ensuring lightweight (3.38 M).