Application of an Improved YOLOv7 Model in Disease and Pest Detection for Lychee Trees

Disease and pest detection grounded in the principles of deep neural network architectures has become a prominent research direction in the field of agricultural technology, attracting widespread attention and emphasis from researchers globally in recent years. This paper investigates the application of an improved YOLOv7 algorithm in the detection of diseases and pests in lychee trees. It introduces a new Efficient Complete Intersection over Union (ECIoU) regression loss function, which, compared to the original CIoU loss function, enhances detection accuracy and convergence speed. By comparing with six other advanced models, the improved algorithm introduced in this manuscript exhibits superior performance compared to existing leading-edge models, in terms of mean average precision (mAP) curves, loss curves, and Precision-Recall curves. Finally, the improved model delineated in this document holds the potential to aid lychee growers in effectively managing pests and reducing the use of pesticides, providing practical benefits for sustainable agriculture.