DCH-YOLO Detection Model for Corn Pests in Complex Backgrounds

Hong-Jun Zhang,Zhen-Dong Yin,Da-Sen Li,Yan-Long Zhao

DOI: https://doi.org/10.1109/icispc63824.2024.00012

2024-01-01

Abstract:Corn cultivation holds significant importance for global food security, economic development, energy sustainability, and ecological environmental protection. Enhancing the detection accuracy of corn pests in complex natural environments is crucial. Addressing challenges such as small target sizes, complex backgrounds, and low detection accuracy in corn pest detection, this study proposes a DCH-YOLO network designed for detecting common corn pests in complex environments. Built upon the YOLOv8 object detection model, this network integrates deformable convolutional network, CBAM, and HarvedIntersection over Union (HIoU) modules. The DCN module dynamically adjusts the sampling position and size of convolutional kernels based on the features of small targets, accurately capturing the position information of pests. The CBAM module adaptively learns the relationships between different positions and channels in the feature map, enhancing focus on small targets and better distinguishing features between pests and backgrounds. HIoU improves the traditional Intersection over Union (IoU) by considering multiple factors such as IoU, position, size, and shape between predicted and ground truth boxes more effectively. Compared to other detection networks, the DCHYOLO network achieves the highest mAP value of $77.2 \%$ in detecting corn pests. In comparative experiments with the original model, the precision, recall, and mAP@0.5 evaluation indicators were improved by $9.1 \%, 9.8 \%$, and $5.5 \%$, respectively, demonstrating outstanding performance. The proposed DCHYOLO model effectively enhances detection accuracy and recognition rates, providing a technical foundation for detecting common corn pests in complex backgrounds.