MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202631071264 A) filed by Guru Nank Institute Of Technology on June 08, 2026, for “agrixai-Net: A Hybrid Explainable Deep Learning Framework For Intelligent Crop Disease Detection And Yield Prediction”.
Inventors include Dr. Suman Bhattacharya; Dr. Mahamuda Sultana; Mr. Arghadip Chakraborty; and Ms. Paramita Sarkar.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: Abstract A hybrid explainable deep learning framework designated as AgriXAI-Net is disclosed for the simultaneous execution of crop disease detection and predictive yield analytics. The system architecture comprises a dual-stream feature extractor wherein a Vision Transformer (ViT) backbone captures long-range spatial representations of leaf-level pathologies, and a parallel Temporal Convolutional Network (TCN) maps multi-spectral temporal sequences to extract field-level crop growth trajectories. A Cross-Task Attention Mechanism (CTAM) connects the dual streams, utilizing a query-key-value attention formulation to mathematically project the physiological impact of identified biological stress directly onto the final crop yield prediction. An Explainable AI (XAI) module based on Integrated Gradients produces high-granularity, pixel-level saliency maps and spectral band attribution maps, providing transparent, visually verifiable evidence of biological markers to enhance stakeholder trust. The model is structurally optimized into a lightweight Edge AI package, achieving 97.4% disease classification accuracy and a 14.2% reduction in yield prediction error with field-ready real-time inference latency.
Disclaimer: Curated by HT Syndication.
