CNN-Based Insect Detection Using YOLO for Resilient Agricultural Systems

Recently, convolutional neural networks (CNNs) have emerged as powerful tools across various domains, such as computer vision, audio processing, and text analysis, thanks to their outstanding performance in cutting-edge applications. In this study, we propose a CNN-based system for detecting harmful insects in agricultural fields, leveraging specialized datasets. Our approach utilizes the capabilities of three YOLO architectures, incorporating advanced techniques in deep learning and computer vision. In our work, we focus on four selected classes from the standard IP102 benchmark dataset for pest recognition, Black Cutworm, Red Spider, Aphids, and Flea Beetle, due to their significant impact on crop health and productivity. We propose a convolutional neural network (CNN)-based architecture using "You Only Look Once" (YOLO), specifically YOLOv5, YOLOv10, to process and evaluate our model. During training, the models achieved mAP scores of 83% for YOLOv5 and 86% for YOLOv10. Our experiments yielded high test accuracies, exceeding 92% for YOLOv5 and YOLOv10. The goal is to reduce pesticide usage, enable timely preventive actions, and mitigate economic losses by predicting infestations, supporting rapid interventions, and promoting sustainable agricultural practices through smart farming technologies. However, the application of CNNs to pest detection in agricultural contexts remains underexplored, particularly within the broader framework of climate action, sustainable land use, and resilient agricultural systems.