An Improved Lightweight Pest Detection Method Based on YOLOv8

Abstract

This study systematically addresses the limitations of traditional pest detection methods and proposes an optimized version of the YOLOv8 object detection model. By integrating the GhostConv convolution module and the C3Ghost module, the Polarized Self-Attention (PSA) mechanism is incorporated to enhance the model’s capacity for extracting pest features. Experimental results demonstrate that the improved YOLOv8 + Ghost + PSA model achieves outstanding performance in critical metrics such as precision, recall, and mean Average Precision (mAP), with a computational cost of only 5.3 GFLOPs, making it highly suitable for deployment in resource-constrained agricultural environments.