Researchers at Chandigarh University say they have built an AI system that estimates how much a crop will yield before anyone brings in the harvest. It reads satellite images, weather records and past production figures, and it was tested on farmland in Punjab.
The team is led by Kusum Lata, an assistant professor in computer science engineering. Dr Navneet Kaur, a professor in the same department, and Dr Simrandeep Singh of the university’s Centre of Research and Development worked on it with her. They presented the study at the 2026 International Conference on Signal Processing and Electronics Design (ICSPED), held at Chandigarh College of Engineering and Technology.
What goes into the model
The system is a lightweight transformer. It pulls from Sentinel-1 and Sentinel-2, two Earth observation satellites run by the European Space Agency. Their imagery tracks crop growth, vegetation health, soil moisture and field conditions through the season.
Lata said that data is merged with rainfall, temperature and soil moisture readings, plus historical production records, “creating a comprehensive picture of crop performance throughout the growing season.”
She also set it apart from older machine learning tools. “Unlike conventional machine learning models, the newly developed lightweight transformer can identify critical crop growth stages and learn complex temporal patterns that influence final yields,” Lata said. The design, she added, needs fewer computing resources, so it could run in large-scale monitoring systems.
Test results
The team checked the model against four crops grown in Ludhiana district: paddy, maize, moong and sugarcane. Data covered 2019 to 2023.
It beat Random Forest and Long Short-Term Memory (LSTM) models, two methods in wide use. Predicted yields lined up more closely with actual ones, errors were lower, and the computing was more efficient. The university did not publish error figures.
The one number it did give concerns size. The architecture needs nearly 40 percent fewer parameters than conventional transformer models, and it returns predictions faster. That makes it a fit for near real-time uses such as regional crop monitoring, production forecasting and early warning systems.
Why Punjab, and what comes next
Field surveys, the usual way to gauge a harvest, take time, need a lot of labour and cover only so much ground. Farmers are also dealing with changing weather and rising demand for food, the researchers noted.
They argue that an accurate forecast before harvest would help farmers and governments alike. Planning, resource allocation, crop insurance and market management could all draw on it. Agriculture plays a central role in the economy of Punjab, they said.
The next step is cloud-based forecasting in near real time. The researchers see that as a path to wider use of AI-driven decision support in farming.
The university issued the announcement on June 27, 2026.
