MUMBAI, India, July 13 -- Intellectual Property India has published a patent application (202641082573 A) filed by Dr. Md. Farukh Hashmi, Assistant Professor Department Of Ece, National Institute Of Technology, Warangal. on July 04, 2026, for Learning Novel Rice Pathogen Variations Using Semi-Supervised And Zero-Shot Techniques.
Inventors include Dr. Md. Farukh Hashmi, Assistant Professor Department Of Ece, National Institute Of Technology, Warangal.; Dr. Mettu Srinivas, Associate Professor Department Of; Dr. Sanjaya Kumar Panda, Associate Professor; T. Paramesh, Assistant Professor Department Of Cse; Vishnuvatdhan Jupelly, Assistant Professor Department; and Dr. Cheva Mahender, Assistant Professor Department Of H&s, Institute Of Aeronautical Engineering..
The application for the patent was published on July 10, 2026, under issue no. 28/2026.
Abstract: Abstract The present invention discloses a unified deep-learning framework for detecting and classifying known and previously unseen rice pathogen variations from leaf imagery under severe label scarcity. A Vision Transformer (ViT-B/16) backbone is pre trained on unlabelled rice leaf images using a DINO self-supervised objective, fine-tuned semi-supervised using FixMatch consistency regularization and a MeanTeacher network with only about 2% labelled data, and aligned with CLIP text embeddings of natural- language pathogen attribute descriptions to enable zero-shot classification of novel pathogen strains. Monte Carlo Dropout provides calibrated uncertainty estimates that flag ambiguous predictions for expert review, and the model is quantized to INT8 precision for real-time edge deployment. The framework achieved 96.4% semi-supervised accuracy, an 89.1% average zero-shot detection rate on novel pathogen classes, a 14 ms inference latency on embedded hardware, and 84.7% accuracy on a cross-domain clinical mycology task, providing a scalable, label-efficient, and deployment-ready solution for rice disease surveillance.
Disclaimer: Curated by HT Syndication.
