New Method for Tomato Disease Detection Based on Image Segmentation and Cycle-GAN Enhancement

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Limited high-quality datasets hinder the development of intelligent agriculture and deep learning (DL) models.
• Existing image enhancement methods often produce correlated samples and fail to remove environmental noise effectively.
• Accurate plant disease detection is crucial for food security and agricultural productivity.

Purpose of the Study:

• To develop an automatic leaf segmentation algorithm (AISG) for enhancing disease feature extraction in limited datasets.
• To improve the performance of DL models for plant disease detection by addressing data scarcity and noise.
• To enable cross-category image transformation for data augmentation using generative adversarial networks.

Main Methods:

• Designed an automatic leaf segmentation algorithm (AISG) based on the EISeg method to isolate leaf disease features from background noise.
• Utilized the Cycle-GAN network for cross-category image transformation to augment limited sample data.
• Employed transfer learning to train a MobileNet model on the enhanced dataset for tomato disease classification.

Main Results:

• The proposed method achieved a classification accuracy of 98.61% for ten types of tomato diseases.
• The AISG effectively separated disease characteristics from background noise, enhancing feature extraction.
• The Cycle-GAN data enhancement and MobileNet transfer learning significantly improved model performance compared to existing methods.

Conclusions:

• The developed AISG and data enhancement techniques effectively address the challenges of low accuracy and insufficient training data in tomato disease detection.
• This approach provides a valuable reference for detecting various plant diseases using deep learning.
• The study highlights the potential of image enhancement and generative models in advancing intelligent agriculture.