MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202541124598 A) filed by Angel Sini S. A; Harininiv Atha S; and Sam Varghese George, Chennai, Tamil Nadu, on Dec. 10, 2025, for 'light weight deep learning model for pest detection and mitigation.'

Inventor(s) include Sam Varghese George.

The application for the patent was published on Feb. 13, under issue no. 07/2026.

According to the abstract released by the Intellectual Property India: "The study relates to an Al-based computer vision system for automatic detection and monitoring of agricultural pests from crop images. The system is designed for use by fanners, agricultural researchers, fanner-aggregators, and smart-farming automation platforms, and provides real-time identification of pest instances directly from captured field images. The invention employs multiple lightweight versions of the YOLO object-detection architecture, including YOLOv5-Nario, YOLOv8-Nano, and YOLOvl l-Nano, trained on the IP 102 pest dataset comprising diverse pest categories with detailed bounding-box annotations. The dataset, originally in VOC2007 XML format, is pre-processed into YOLO-compatible label files and organized into training and validation sets, with additional augmentation applied to address farm-level variations such as illumination changes, background noise, and image distortions. A unified training and evaluation pipeline is used for all YOLO variants to ensure consistent comparison, incorporating dataset preparation, model initialization, training. validation on unseen images, and storage of trained model weights. System performance is CD O re Q. CD assessed using precision, recall, accuracy, mean Average Precision (niAP), inference speed, model size, and computational usage. The real-time inference engine outputs annotated detections including bounding boxes, pest class labels, and confidence scores, enabling CM E o deployment in low-computation agricultural environments. The invention further supports future extensions including additional pest categories, adaptation to seasonal and oo ion Tf CM environmental conditions, integration with automated crop-monitoring and alert systems, and export of trained models to lightweight formats such as TensorFlow Lite and ONNX for in CM o CM CM roo oo CM deployment on mobile, edge, and loT devices."

Disclaimer: Curated by HT Syndication.