Pest Net-X: Vision Transformer AI For Real-Time Multispectral Pest Detection

Authors: Dr. C Nandini, Professor Rajesh M, Shashikala B Thakur, Srikumar L, Varshitha G C, Yashaswini M A

Abstract: Pest Net-X presents an innovative AI-driven solution for real-time, multispectral pest detection in agriculture, leveraging a hybrid Vision Transformer (ViT) architecture optimized for edge deployment. Unlike conventional CNN-based approaches, Pest Net-X integrates RGB and near-infrared (NIR) spectral analysis to identify pests at early developmental stages (egg/nymph phases) with 95.1% accuracy—surpassing existing tools like Plantix by 8.8%. The system features a farmer-centric mobile app with bilingual (Kannada/English) support, offline functionality, and explainable AI (Grad-CAM++ heatmaps) to deliver actionable pest advisories. Field trials with 50 South Indian farmers demonstrated a 27% reduction in crop losses and ₹5,800/acre cost savings through precision pesticide use. Pest Net-X’s lightweight TensorFlow Lite implementation achieves 47ms inference latency, making it viable for low-end smartphones in resource-limited settings. This work bridges critical gaps in agricultural AI by combining multispectral ViT technology, edge computing, and vernacular accessibility to empower sustainable farming practices.