A tomato disease identification method based on leaf image automatic labeling algorithm and improved YOLOv5 model

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Tomato production is globally significant, requiring accurate disease identification for yield and quality.
• Convolutional Neural Networks (CNNs) are vital for disease identification but demand extensive manual image annotation.
• Manual annotation is time-consuming and costly, hindering research efficiency.

Purpose of the Study:

• To develop an automated tomato leaf image labeling algorithm to streamline disease identification.
• To enhance the accuracy and balance of recognition effects for various tomato diseases.
• To improve upon existing methods for tomato disease detection using deep learning.

Main Methods:

• Proposed BC-YOLOv5 method for tomato disease recognition.
• Incorporated an automatic tomato leaf image labeling algorithm.
• Modified the YOLOv5 Neck structure with a weighted bi-directional feature pyramid network.
• Integrated a convolution block attention module and adjusted the detection layer input channels.

Main Results:

• BC-YOLOv5 achieved an excellent image annotation pass rate exceeding 95% for tomato leaves.
• Demonstrated superior performance indices compared to existing models for tomato disease identification.
• Successfully identified healthy growth and nine types of diseased tomato leaves.

Conclusions:

• BC-YOLOv5 enables automatic labeling of tomato leaf images prior to training.
• The method accurately identifies nine common tomato diseases with improved and balanced recognition.
• Provides a reliable and efficient solution for tomato disease identification in agriculture.