Paper Details
Abstract
Preservation of crops relied on early and accurate detection of pests, which significantly reduce both yield and quality. Recent advances in deep learning have provided powerful tools for agricultural pest monitoring. In this study, we proposed YOLOCSP-PEST, an enhanced version of YOLOv7 incorporating the Cross Stage Partial Network (CSPNet) backbone, optimized for pest localization and classification. While YOLOv7 and CSPNet are established techniques, their integration and systematic evaluation across both benchmark (IP102) and real-world local datasets provided new insights into practical deployment for smart agriculture. On the IP102 dataset, YOLOCSP-PEST achieved an mAP of 88.40%, surpassing comparable baselines, while on a locally collected dataset it attained 97.18\% mAP, 97.50\% precision, and 94.89% recall. An ablation study confirmed the benefit of CSPNet integration, and efficiency evaluations demonstrated the model’s applicability to real-world scenarios. These results highlighted YOLOCSP-PEST as a competitive and adaptable framework for pest detection, with potential to improve crop monitoring and management strategies.