A Recognition Method of Soybean Leaf Diseases Based on an Improved Deep Learning Model.

Area of Science:

• Agricultural Science
• Plant Pathology
• Computer Vision

Background:

• Accurate detection of soybean diseases is crucial for crop yield and quality, but traditional methods are slow, inaccurate, or subjective.
• Existing deep learning models struggle with recognition accuracy, and chemical analyses are time-consuming.
• Manual assessment of soybean diseases is prone to subjective bias, hindering reliable diagnosis.

Purpose of the Study:

• To develop a rapid and accurate method for identifying soybean leaf diseases using a novel deep learning model.
• To improve upon the limitations of traditional disease detection methods in terms of speed, accuracy, and objectivity.
• To establish a robust soybean disease recognition system for enhanced agricultural management.

Main Methods:

• A Residual Attention Network (RANet) model was developed by integrating attention mechanisms and shortcut connections into the ResNet18 architecture.
• The OTSU algorithm was employed for background removal in soybean leaf images.
• Image enhancement techniques were utilized to expand the dataset of soybean disease images.

Main Results:

• The proposed RANet model achieved an average soybean leaf disease recognition accuracy of 98.49%.
• The F1-score for disease recognition reached 98.52%.
• The model demonstrated a rapid recognition time of 0.0514 seconds per image.

Conclusions:

• The RANet model provides an accurate, fast, and efficient solution for soybean leaf disease recognition.
• This deep learning approach overcomes the limitations of traditional methods, offering significant potential for agricultural applications.
• The developed model supports precise soybean breeding, cultivation, and management through reliable disease identification.