A Lightweight Leaf Disease Recognition Model for MobileNetV2 Based on Transfer Learning and Knowledge Distillation

Accurate identification of plant diseases is critical for the sustainability and efficiency of agricultural production. Traditional image recognition models have limitations in fine-grained disease feature extraction, which may affect the accuracy and usefulness of recognition. To address this issue, this paper presents a MobileNetV2 model optimized by knowledge distillation techniques that significantly improves plant disease recognition on mobile devices. By using a lightweight but efficient student model and combining it with a transfer learning technique, we successfully approach the performance of the complex teacher model DenseNet201, achieving high accuracy and utility of the model. The model is particularly suitable for resource-constrained devices such as mobile and embedded systems.