Materials, Methods & Technologies, Volume 19, 2025

Abstract: Insect pests that affect corn leaves are significant contributors to agricultural losses. Despite corn being a globally consumed food crop, there is lack of research on the automated identification of these pest’s damage, which limits farmers` ability to implement effective control measures. This research aims to fill these gaps by implementing various YOLOv10 architectures (10n, 10s, 10m, 10l, 10x) for the automated identification and localization of insect pests on corn leaves. A total of 1,395 images of highly affected corn leaves from Pokhara, Nepal, were collected, showing damage from grasshoppers and fall armyworms. The collected images underwent5-fold cross validation and split dataset into Train, Valid, Test in a ratio of 3:1:1 being fed into various architectures. Among the pest detection models tested, YOLOv10x outperformed the others in terms of mean average precision (mAP50 = 0.799) at an IoU threshold of 50%. Regarding the Precision-Recall (PR) relationship YOLOv10m has balancing nature of highest precision (Precision= 0.808) with high recall value (Recall= 0.807), this trend is not followed by YOLOv10l and YOLOv10x. In terms of latency, YOLOv10n required the least average time to detect each image during testing, making it the most lightweight model among those tested. The central tendency of YOLOv10m is highest among all YOLOv10 variants with decent spread box over interquartile range. Overall, with sufficient resource availability YOLOv10x is beneficial while with resource constraints condition YOLOv10m is efficient choice for insect damage detection in corn leaves. This research is valuable for farmers in detecting insect pests on corn leaves, enabling timely implementation of measures to reduce agricultural damage caused by pests. We also recommend optimizing the architectures to enhance the precision of pest detection.