Archives of Agriculture Research and Technology
[ ISSN : 2832-8639 ]
Automated GUI-Based System for In-Season Crop Classification and Acreage Estimation using Multi Temporal SAR Data and Machine Learning
1Regional Remote Sensing Centre-North, New Delhi, India
2Department of CSE, The NorthCap University, Gurugram, Haryana, India
Corresponding Authors
Keywords
Abstract
Accurate in-season crop acreage statistics are crucial for agricultural policymakers, stakeholders, and the food security community. However, manual work ows for downloading, preprocessing, and classifying Synthetic Aperture Radar (SAR) data - especially over large areas such as states or country - are time-consuming and prone to ine ciencies. is study presents a Graphical User Interface (GUI)-based automated system designed for multi-class crop classi cation and acreage estimation using multi-temporal SAR data and machine learning techniques. e system supports data acquisition from two SAR platforms: Sentinel-1A (VV & VH) and EOS-4 (HH & HV), based on user-de ned areas of interest (district or state level). e automation pipeline includes comprehensive preprocessing steps such as radiometric calibration, speckle ltering, geometric/terrain correction, mosaicing, crop land masking, and layer stacking. Classi cation is performed using six di erent machine learning algorithms, with integrated hyperparameter tuning to enhance model performance. e system outputs class-wise crop area statistics and validates class separability using backscatter temporal pro les. To evaluate its e ectiveness, the system was applied to selected districts and achieved an overall classi cation accuracy of approximately 90% for major crops including paddy, arhar, cotton, and maize. A four-page auto-generated report summarizes the outputs, featuring the classi cation report, confusion matrix, backscatter curves, variable importance plots, and classi ed imagery. e system is scalable, e cient, and user-friendly - requiring minimal technical expertise - making it a valuable tool for stakeholders in the agricultural domain. Its methodology can be readily adapted to other datasets and geographic regions, supporting broader applications in operational crop monitoring and decision-making.