Dangerous Farm Insect Detection using YOLOv8

This project focuses on developing an automated detection system for identifying dangerous farm insects using computer vision techniques. The dataset, sourced from Kaggle, includes 15 insect classes with around 120 samples each. Images were preprocessed and annotated using Roboflow to ensure accurate labeling for object detection.

The YOLOv8 (You Only Look Once, version 8) model was selected for its high speed, precision, and flexibility in handling detection and segmentation tasks. Through training and evaluation, the model achieved strong performance on clearly labeled insect classes but faced challenges with multi-labeled images due to overlapping bounding boxes. The overall accuracy ranged between 57–60%, with single-object classes performing significantly better.

This system can assist in automated pest monitoring and management, potentially improving agricultural productivity and minimizing crop damage. Future improvements may involve refining labeling quality and enhancing multi-object detection capabilities.

Project Report