Blab AI
Chandigarh University Researchers Unveil Lightweight Transformer Model for Crop Yield Prediction
A new AI model could give Punjab’s farmers a clearer view of what the harvest will bring. On June 27 2026, Chandigarh University presented a transformer‑based system that fuses satellite imagery, climate data and historical production records to forecast crop yields. The unveiling took place at the 2026 International Conference on Signal Processing and Electronics Design (ICSPED) in Chandigarh, a venue that attracted agronomists, data scientists and policy makers.
The research team, based in the University Centre of Research and Development, is led by Assistant Professor Kusum Lata, Professor Navneet Kaur and Professor Simrandeep Singh. Their work centers on refining yield predictions for Punjab’s agricultural heartland, a region where precision farming can translate directly into economic gains.
At the core of the model lies data from the European Space Agency’s Sentinel‑1 synthetic‑aperture radar and Sentinel‑2 optical satellites. Kusum Lata explained, “The transformer model utilizes data from Sentinel‑1 and Sentinel‑2 satellites… The satellite observations are combined with climatic variables… creating a comprehensive picture of crop performance throughout the growing season.” By layering rainfall, temperature, soil moisture and past crop yields, the system builds a holistic view of each field’s health.
In a rigorous evaluation, the model was tested on four major crops—paddy, maize, moong and sugarcane—in Ludhiana district, using data spanning 2019 to 2023. The researchers reported that the transformer outperformed established approaches such as Random Forest and Long Short‑Term Memory (LSTM) models, achieving lower prediction errors and improved computational efficiency. Kusum Lata noted, “The model was evaluated on four major crops… Experimental results demonstrated that the transformer model outperformed… indicating stronger agreement between predicted and actual yields.”
A key advantage is the model’s lightweight architecture. It requires nearly 40 percent fewer parameters than conventional transformer models while delivering faster and accurate predictions. This efficiency makes it suitable for near real‑time forecasting via cloud platforms, the researchers said.
Accurate pre‑harvest yield forecasts can inform agricultural planning, resource allocation, crop insurance and market management. In Punjab, where agriculture drives a large portion of the economy, the technology could enhance resilience and sustainability.
Future work will focus on deploying the model on cloud‑based platforms to broaden adoption and integrate additional data sources. The team aims to support AI‑driven decision‑support systems in agriculture.
The lightweight transformer model represents a step forward in precision farming. While the research shows promising results, further validation across more regions and crop types is needed. Deployment will depend on infrastructure readiness and stakeholder engagement.
The research team, based in the University Centre of Research and Development, is led by Assistant Professor Kusum Lata, Professor Navneet Kaur and Professor Simrandeep Singh. Their work centers on refining yield predictions for Punjab’s agricultural heartland, a region where precision farming can translate directly into economic gains.
At the core of the model lies data from the European Space Agency’s Sentinel‑1 synthetic‑aperture radar and Sentinel‑2 optical satellites. Kusum Lata explained, “The transformer model utilizes data from Sentinel‑1 and Sentinel‑2 satellites… The satellite observations are combined with climatic variables… creating a comprehensive picture of crop performance throughout the growing season.” By layering rainfall, temperature, soil moisture and past crop yields, the system builds a holistic view of each field’s health.
In a rigorous evaluation, the model was tested on four major crops—paddy, maize, moong and sugarcane—in Ludhiana district, using data spanning 2019 to 2023. The researchers reported that the transformer outperformed established approaches such as Random Forest and Long Short‑Term Memory (LSTM) models, achieving lower prediction errors and improved computational efficiency. Kusum Lata noted, “The model was evaluated on four major crops… Experimental results demonstrated that the transformer model outperformed… indicating stronger agreement between predicted and actual yields.”
A key advantage is the model’s lightweight architecture. It requires nearly 40 percent fewer parameters than conventional transformer models while delivering faster and accurate predictions. This efficiency makes it suitable for near real‑time forecasting via cloud platforms, the researchers said.
Accurate pre‑harvest yield forecasts can inform agricultural planning, resource allocation, crop insurance and market management. In Punjab, where agriculture drives a large portion of the economy, the technology could enhance resilience and sustainability.
Future work will focus on deploying the model on cloud‑based platforms to broaden adoption and integrate additional data sources. The team aims to support AI‑driven decision‑support systems in agriculture.
The lightweight transformer model represents a step forward in precision farming. While the research shows promising results, further validation across more regions and crop types is needed. Deployment will depend on infrastructure readiness and stakeholder engagement.
