Integration of Drone-Based Remote Sensing and Machine Learning for Real-Time Agricultural Yield Prediction in Indian Farmlands

Abstract

Drone-based remote sensing now sees regular use in Indian agriculture for field monitoring and crop assessment. Yield prediction at the field level still faces limits due to slow data handling and weak linkage with learning models. This study examines an integrated approach combining drone-based multispectral imagery with machine learning models to support near real-time crop yield prediction under Indian farming conditions.

The study uses high-resolution multispectral imagery collected through multi-rotor drones across selected agricultural plots growing rice, wheat, and cotton. Data collection covered key crop growth stages. Image features included vegetation indices, canopy structure measures, and stage-specific growth indicators. The study linked these features with supervised machine learning models trained using yield values recorded at harvest. Model evaluation relied on standard accuracy measures applied across crop types, field sizes, and operating conditions.

Results show higher yield prediction accuracy when machine learning models use drone-derived features compared with basic statistical approaches. Prediction performance differed across crops and growth stages. Mid-season and pre-harvest stages showed more stable estimates. Field size, crop uniformity, and local environmental conditions influenced model consistency and error levels. Near real-time processing reduced the time gap between data capture and yield estimation.

The findings indicate practical value for integrated drone sensing and machine learning workflows in Indian agriculture. Observed limits relate to data quality variation, model transfer across sites, and sensitivity to environmental factors. The results guide future development of scalable decision-support tools for precision agriculture based on integrated sensing and learning systems