Multiscale Parallel Algorithm for Early Detection of Tomato Gray Mold in a Complex Natural Environment.

Area of Science:

• Agricultural technology
• Computer vision
• Plant pathology

Background:

• Intelligent agricultural Internet of Things (IoT) systems require accurate and real-time plant disease detection.
• Existing lightweight models like MobileNetv2-YOLOv3 offer real-time performance but lack sufficient accuracy for natural environments.
• Tomato gray mold poses a significant threat to crop yield, necessitating advanced detection methods.

Purpose of the Study:

• To develop an accurate and real-time plant disease detection algorithm for intelligent agriculture.
• To improve the detection accuracy of multiscale tomato gray mold lesions.
• To enhance the MobileNetv2-YOLOv3 model for practical application in natural environments.

Main Methods:

• Proposed a multiscale parallel algorithm, MP-YOLOv3, based on the MobileNetv2-YOLOv3 architecture.
• Implemented a multiscale feature fusion method for enhanced detection capabilities.
• Integrated an efficient channel attention mechanism into the detection layer for feature enhancement.
• Utilized a parallel detection algorithm to optimize performance.

Main Results:

• The MP-YOLOv3 algorithm accurately and in real-time detects multiscale tomato gray mold lesions.
• Achieved an F1 score of 95.6% and an average precision of 93.4% on a custom dataset.
• The model boasts a small size (16.9 MB) and rapid detection time (0.022 s per image).

Conclusions:

• The MP-YOLOv3 algorithm effectively addresses the limitations of existing models for plant disease detection.
• It provides a viable solution for real-time, accurate detection of tomato gray mold in complex natural settings.
• The developed model contributes to the advancement of intelligent agricultural monitoring systems.