Leveraging YOLO deep learning models to enhance plant disease identification

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Plant diseases cause significant crop losses, impacting food security and economies, especially in developing nations.
• Early and accurate disease identification is challenging due to symptom variability and similarity.
• Artificial intelligence (AI) offers potential for automated plant disease classification, but faces challenges like data imbalance and annotation costs.

Purpose of the Study:

• To evaluate the performance of You Only Look Once (YOLO) deep learning models, specifically YOLOv3 and YOLOv4, for the early identification of fruit plant diseases.
• To compare the accuracy, speed, and efficiency of YOLOv3 and YOLOv4 in detecting bacterial spots on peach leaves and scorch disease on strawberry leaves.

Main Methods:

• Utilized the publicly available Plant Village dataset for training and validation.
• Implemented and trained YOLOv3 and YOLOv4 models for image-based classification of healthy and diseased peach and strawberry leaves.
• Assessed model performance based on accuracy, Mean Average Precision (mAP), and detection time.

Main Results:

• YOLOv3 achieved 97% accuracy and 92% mAP, with a detection time of 105 seconds.
• YOLOv4 demonstrated superior performance with 98% accuracy and 98% mAP, completing detection in only 29 seconds.
• YOLOv4 exhibited faster, more precise, and less complex performance, particularly for multi-object detection.

Conclusions:

• YOLOv4 is a highly effective and efficient deep learning model for real-time plant disease detection.
• The proposed AI approach holds significant potential to enhance agricultural productivity and economic outcomes, especially in developing countries.
• Automated plant disease diagnosis using advanced AI can revolutionize crop protection strategies.