A lightweight CNN model for pepper leaf disease recognition in a human palm background

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Manual diagnosis of pepper leaf diseases is time-consuming and prone to errors.
• Accurate identification of pepper diseases is vital for crop yield and quality.
• Existing automated methods may lack efficiency and accuracy in field conditions.

Purpose of the Study:

• To develop a lightweight convolutional neural network (CNN) for accurate pepper leaf disease recognition.
• To create a practical mobile application for real-time disease diagnosis.
• To address the limitations of manual disease identification in agricultural settings.

Main Methods:

• Acquired a dataset of healthy and diseased pepper leaf images, including common conditions like viral diseases, brown spots, and leaf mold.
• Designed and implemented a novel CNN model based on GGM-VGG16 architecture with Ghost modules, global average pooling, and multi-scale convolution.
• Developed an Android application integrating the trained CNN model for deployment on mobile terminals.

Main Results:

• The proposed CNN model achieved 100% accuracy on images with a human palm background, common in field settings.
• The model demonstrated strong performance on other backgrounds, with an accuracy of 87.38%.
• The resulting mobile application is compact (12.84 MB) and provides robust disease recognition capabilities.

Conclusions:

• The developed lightweight CNN model offers a highly accurate and efficient solution for pepper leaf disease identification.
• The mobile application provides a practical tool for farmers and agricultural professionals to diagnose diseases in the field.
• This AI-driven approach enhances crop management and contributes to food safety and quality.