Performance Analysis of the YOLOv5 Model in Planthopper Pest Detection: From Nano to Medium

The problem of pests, especially planthoppers, has become a big challenge in the agricultural sector, especially brown planthoppers which affect the quality and quantity of rice production. This research uses Artificial Intelligence (AI) technology, specifically the Convolutional Neural Network (CNN) method with the You Only Look Once (YOLO) v5 algorithm, to detect and classify planthopper pests. The focus of the research is to compare the performance of YOLO v5 in three variants: Nano, Small, and Medium. The research results show that YOLOv5 Medium provides the best performance with the highest mAP50-95 value of 0.332. The YOLOv5 Nano outperforms the Small in terms of precision and mAP50, while the YOLOv5 Small has slightly higher recall. Confusion Matrix results for YOLOv5 Medium show Precision (P) = 0.539, Recall (R) $=0.575, \text{mAP}50=0.530$, and mAP50-95 = 0.332. This research emphasizes the importance of adapting the YOLO method to suit field conditions, taking into account factors such as agricultural needs, scale, and available resources to ensure effective pest detection.