Journal of Agricultural Machinery Engineering ISSN:2799-8673(Print) 2799-8819(Online) 농업기계공학

Pests and diseases cause significant agricultural losses each year, posing a serious threat to global food security. According to the Food and Agriculture Organization, while approximately 50% more food production will be required by 2050, nearly 40% of current global crop yields are already lost to pest damage. As climate change alters the ecology and occurrence patterns of insect pests, conventional manual monitoring methods suffer from low accuracy and efficiency, limiting timely responses. In addition, earlier attempts to automate trap monitoring relied mainly on conventional image-processing techniques. Although such approaches showed the possibility of automated pest detection, their performance was often sensitive to pest posture, illumination changes, and complex field backgrounds, which limited robust application in real-world environments. To address these challenges, this research proposes a deep learning–based object detection model for species-level automatic detection and identification of two major pests— Spodoptera litura (tobacco cutworm) and Spodoptera exigua (beet armyworm)—within automatic trap environments. Model performance was evaluated using Precision, Recall, mAP@50, and mAP@75 metrics, based on training and validation with field-collected trap image data. Preliminary results demonstrate that the proposed system achieves reliable multi-species detection under varying environmental conditions, reduces dependence on on-site inspection through remote monitoring, and shows potential for supporting pest management decisions such as early warning and density estimation.