Abstract

Pest diseases are serious global agricultural issues that lead to lower crop yields, increased cost of production and excessive pesticide use. Traditional methods of identifying pest infestation (e.g., using field scouting methods) rely on intensive labor, human time, and human errors, thus making them impractical for large-scale and sustainable farming. This project is a structured deep learning-based system for automatically identifying pest diseases and pests through image identification. The system is developed using the YOLOv11 state-of-the-art model for object identification and has been trained on a custom-dataset from the objects of three pest species - bagworms, aphids, and whiteflies. The images representing each pest were pre-processed and augmented in order to equalize data and optimal modeling performance. The experimental evaluation of the trained model archieved a precision of 0.88, recall of 0.80, and mAP@0.5 of 0.85, outperforming conventional detection methods and demonstrating strong reliability even with imbalanced classes., thus demonstrating the proposed system is viable for use in real-world agricultural environment. The proposed system can provide an intervention to pest infestation enabling early and timely diagnoses of pest infestation, which in turn may help reduce over-use of pesticides, and contribute to more targeted use of pesticides and sustainable farming practices.

This work is licensed under a Creative Commons Attribution 4.0 License.