Paper Details
Abstract
Preservation of crops depended heavily on early and accurate detection of pests, as they caused numerous diseases that reduced both crop yield and quality. Various Deep Learning techniques had been applied to address the challenge of pest detection in crops. In this study, we developed the YOLOCSP-PEST model, a modified version of YOLOv7 incorporating the CSPNet backbone, for effective pest localization and classification. Our proposed model achieved outstanding results compared to other methods, even under challenging conditions such as varying image luminance and orientation. The model was trained and evaluated on two datasets: the IP102 dataset and a local crop dataset. On the IP102 dataset, the model achieved an mAP of 88.40%, a precision of 85.55%, and a recall of 84.25%. On the local dataset, it demonstrated even better performance, achieving a mAP of 97.18%, a precision of 97.50%, and a recall of 94.88%. These results demonstrated that the proposed model was highly effective in real-world pest detection scenarios and could significantly contribute to enhancing crop yield and quality.