Apple Leaf Disease Identification with a Small and Imbalanced Dataset Based on Lightweight Convolutional Networks

Area of Science:

• Agricultural technology
• Plant pathology
• Computer vision

Background:

• Accurate plant disease identification is crucial for agriculture.
• Existing methods for apple leaf disease detection have limitations.

Purpose of the Study:

• To develop a rapid and accurate method for identifying apple leaf diseases.
• To propose a new lightweight convolutional neural network (CNN) model named RegNet.

Main Methods:

• Collected a dataset of 2141 images of 5 apple leaf diseases and healthy leaves.
• Developed and trained a novel lightweight CNN, RegNet.
• Compared RegNet's performance against state-of-the-art models like ShuffleNet, EfficientNet-B0, MobileNetV3, and Vision Transformer.

Main Results:

• RegNet-Adam achieved an average accuracy of 99.8% on the validation set.
• RegNet obtained an overall accuracy of 99.23% on the test set.
• RegNet outperformed all compared pre-trained models in identifying apple leaf diseases.

Conclusions:

• The proposed RegNet model enables rapid and accurate identification of apple leaf diseases.
• Transfer learning with RegNet is effective for agricultural disease detection.
• This research contributes to intelligent agriculture through advanced image recognition.