A lightweight rice pest detection algorithm based on improved YOLOv8.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Accurate rice pest detection is vital for effective pest control, crop yield, and quality.
• Challenges in rice pest identification include high similarity between pest classes, age variations within a class, and complex backgrounds.
• Existing deep neural network models struggle with rapid and accurate identification of diverse rice pests.

Purpose of the Study:

• To develop a fast and accurate deep learning model for rice pest detection and identification.
• To address the limitations of current methods in handling pest visual complexities.

Main Methods:

• Introduction of Rice-YOLO, a novel object detection model based on YOLOv8-N.
• Incorporation of an efficient detection head tailored for pest characteristics.
• Integration and enhancement of deep supervision layers and a dynamic upsampling module.

Main Results:

• Rice-YOLO demonstrated superior performance compared to existing object detection algorithms on benchmark datasets (IP102 and R2000).
• Achieved high performance metrics: 78.1% mAP@0.5, 62.9% mAP@0.5:0.95, and 74.3% F1 score.
• The model effectively handles the challenges of interclass similarity, intraclass age differences, and complex backgrounds.

Conclusions:

• Rice-YOLO offers a significant advancement in automated rice pest detection and identification.
• The proposed model provides a robust and efficient solution for agricultural applications, contributing to improved pest management strategies.
• This research highlights the potential of tailored deep learning architectures for specialized agricultural challenges.