MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202621040489 A) filed by Vivek Kumar; and Dr. Rahul R. Chakre on March 31, 2026, for A System And Method For Explainable Multimodal Learning For Joint Yield-Price Forecasting In Agriculture.
Inventors include Vivek Kumar; and Dr. Rahul R. Chakre.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: A computer-implemented system and method for jointly forecasting agricultural crop yield and commodity wholesale price using explainable multimodal machine learning. The system fuses four publicly available data streams - district-level crop production statistics, daily wholesale market prices, satellite multispectral reflectance (Sentinel-2 and MODIS), and ERA5-Land meteorological reanalysis - into a unified weekly district-level feature panel. Two fusion architectures are provided: Early Fusion concatenates all modalities into a shared CNN-LSTM encoder with separate yield and price prediction heads trained jointly; Late Fusion first trains a yield model on supply-side features and injects derived supply signals into a separate price model. Both produce multi-horizon forecasts with calibrated prediction intervals. An explainability module provides global feature importance, local gradient-based attributions, and interactive scenario analysis. Validated on wheat across twenty-two districts of Haryana, India, the system achieves R-squared 0.877 for yield and R-squared 0.777 for price on the 2025 held-out test set.
Disclaimer: Curated by HT Syndication.
