IIIT-Delhi researchers build AI dataset to monitor crops, predict yields

New Delhi: Researchers from the Indraprastha Institute of Information Technology-Delhi (IIIT-Delhi) and the MS Swaminathan Research Foundation have developed a new multi-satellite dataset that could help artificial intelligence systems monitor crops, identify what is being grown in individual fields and predict agricultural yields.

Called SICKLE (Satellite Imagery for Cropping annotated with Key parameter Labels), the dataset combines satellite imagery with detailed field-level information gathered from agricultural plots. The researchers say such data is important for developing AI tools that can understand crop growth over time rather than simply identify vegetation in satellite images.

SICKLE contains around 2,370 season-wise samples from 388 agricultural plots, combining data from Landsat-8, Sentinel-1 and Sentinel-2. The imagery is available at resolutions of 3, 10 and 30 metres.

The dataset includes information on crop type and variety, growing season, sowing or transplanting, harvesting and productivity. Researchers used it to test AI models for crop identification, crop phenology estimation and yield prediction.

The dataset covers agricultural plots in the Cauvery Delta, spanning four districts and the period from January 2018 to March 2021. It includes 21 crop types, with 351 detailed paddy samples from 145 plots.

The three satellites provide different types of observations, including optical and microwave data. Radar imagery can help when cloud cover limits optical satellite observations, potentially enabling more consistent crop monitoring.

AI systems trained using SICKLE could eventually help researchers and policymakers map

crops and estimate productivity across large agricultural regions, reducing dependence on conventional field surveys.

ABOUT THE AUTHOR

The Hans India