Cocoguard: A Comparative Study of YOLO-Family Models and Faster R-CNN for Coconut Pest Identification With Rule-Based Decision Support

Abstract

Coconut farming is a vital component of the Philippine agricultural sector; however, destructive pests such as the Asiatic Palm Weevil, Rhinoceros Beetle, Brontispa beetle, and Slug Caterpillar can significantly reduce crop yields when infestations are not detected early. Conventional pest monitoring relies on manual inspection and delayed expert consultation, making pest identification slow and often inaccessible for smallholder farmers. This study proposes CocoGuard, a mobile-accessible artificial intelligence-assisted system for coconut pest detection using deep learning object detection models. A dataset of 2,076 coconut pest images, expanded to 5,398 images through augmentation, was prepared and annotated into seven pest classes. Several object detection models were benchmarked using Precision, Recall, F1-score, and mean Average Precision. Results show that YOLO26s achieved the best performance, obtaining 92.72% mAP@0.5, 92.51% precision, and 89.49% recall while maintaining computational efficiency suitable for mobile deployment.