Scientists have developed an AI-based system that uses handheld field sensors and deep learning to predict wheat yield early and with high accuracy. This could strengthen India’s food security plans.
The study, published in Theoretical and Applied Genetics on February 25, 2026, outlines a framework that combines real-time field data with advanced AI to improve the accuracy of crop predictions.
The study was led by Gyanendra Pratap Singh, Director, ICAR National Bureau of Plant Genetic Resources (NBPGR), and ICAR Indian Agricultural Statistics Research Institute (IASRI) Principal Scientists Jyoti Kumari and Girish Kumar Jha. Scientists Sudhir Nabate and Yasavantha Kumar from Agarkar Research Institute (ARI), Pune, were key collaborators along with teams from ICAR and Indian Agricultural Research Institute (IARI).
Accurate yield estimates are essential for managing food supplies, stabilizing markets, and guiding policy. Traditional methods, satellite imagery, and crop statistical models face challenges such as cloud cover, low resolution, and limited ability to capture the complex interactions of weather, soil, and genetics.
To address these gaps, researchers developed a hybrid AI framework called Genetic Algorithm Optimized Deep Neural Network (GA-DNN). The model combines deep neural networks to detect complex patterns in large datasets and genetic algorithms to optimize model parameters through the principles of natural selection.
“The system continuously improves prediction accuracy while learning the complex relationships between plant characteristics and final grain yield,” said Nabate.
Unlike traditional approaches, this study leveraged proximity sensing, handheld or vehicle-mounted devices such as GreenSeeker sensors to collect real-time field data on NDVI (plant greenness and vigor), canopy temperature (plant stress), and plant height (biomass growth).
The AI was trained on data from 3,350 wheat genotypes grown under irrigated and rainfed conditions in New Delhi and Pune. NDVI measurements at three major growth stages: ground cover, flowering, and maturity were the most reliable predictors of yield. GA-DNN outperformed traditional machine learning models and maintained high accuracy even under rainy weather conditions.
“This framework can be integrated into a digital advisory platform for making real-time decisions on irrigation and nutrient management,” Nabate added.
This system has multiple benefits. Breeders can identify superior varieties early, farmers can get reliable pre-harvest yield estimates, and policy makers can develop more accurate production forecasts for sourcing and storage.
According to the researchers, this is India’s first application of a genetic algorithm-optimized deep learning model using hand-held sensor data to predict wheat yield. This research supports precision agriculture efforts and could help ensure stable wheat production even in a changing climate and reduce uncertainty in India’s agricultural system.
