Detection of cotton pests using an enhanced deep learning model

Abstract

Pests in cotton fields significantly impact the normal growth and development of cotton plant, resulting in a decline in both quality and yield, subsequently affecting the productivity of farmers. In addressing the prevalent elongated structures found in insects, this study extends the YOLOv8n algorithm by introducing dynamic snake convolution. This addition facilitates efficient learning of the elongated features of cotton insects. Our algorithm achieved a F1-score value of 92.71 %, an mAP50 value of 97.50 %,an mAP50-95 value of 80.13 %.Additionally, we conducted comparative experiments with well-known object detection algorithms, including Efficientdet, Retinanet, SSD, YOLOv5, YOLOv8n, and YOLOv8s. The results demonstrate that our algorithm exhibits higher accuracy and precision.Furthermore, we evaluated our approach on additional publicly available insect datasets, revealing that our Snake-YOLO algorithm outperforms in detecting insects with elongated features.