Agriculture Pest Detection using YOLOv9 with a Novel Layer Addition

The early detection of agricultural pests is vital for crop protection and improving productivity on farms. In this work, we present an improved object detection model based on the YOLOv9 platform, where we have implemented a custom layer to increase the detection of small and non-uniformly shaped pests, a common issue faced in pest detection. The dataset used is a quality-recommended and well-labeled dataset of pest images of different crops and different conditions. The findings indicate that the improved model is more accurate in precision, recall, and mean average precision (mAP) than the baseline YOLOv9. The study also assesses the computational efficiency of the model to guarantee its application for real-time pest detection. The results indicate that this method has the potential to curb pesticide overuse and promote more sustainable pest control practices in agriculture.