PestinaNet- A Real-Time Crop Pest Detection System

Crop pests are the major factor that reduce the crop yield and influence the health of crops. However, manual identification of pests is time-consuming, and there is a high probability that some pests would go unnoticeable. Therefore, farmer’s essential task is to identify, maintain and control the pest’s population to ensure crop health and quality yield. In this research, we propose an autonomous pest detection system that uses a deep learning technique to assist farmers in localization of pests. Towards this, we used RetinaNet and modified it for pest detection. The experiments are evaluated on publicly available IP102 dataset comprising 97 classes. The PestinaNet was used for pest detection due to its capability to resolve the issue of class imbalance problem between crop and pest pixels. Our experimental findings reveal a significant performance gain of 51.91% mean average precision(mAP) on the benchmark IP102 dataset in contrast with state-of-the-art methods. The proposed work has practical bearing in the development of an automatic pest detection system.