Smart Harvest Monitoring of paddy Using UAV and ML in Bargarh District, Odisha

Abstract. Accurate crop monitoring during the harvest period is crucial for agricultural planning, procurement, insurance claims, and post- harvest management. This study presents a comprehensive analysis using high-resolution Unmanned Aerial Vehicle (UAV) imagery and machine learning (ML) techniques to classify harvest-stage crops in Rajborsambar Block of Bargarh District, Odisha — a predominantly paddy-growing region in eastern India. UAV surveys were conducted during the harvest phase to capture multispectral and RGB imagery across selected agricultural fields. Based on field observations and ground truth data, crops were categorized into four classes: (1) harvested crop, (2) standing crop, (3) cut and spread crop, and (4) other land use/cover. Notably, traditional satellite imagery such as Sentinel-2 (S2) lacks the spatial detail required to accurately detect cut and spread crops due to coarse resolution and mixed pixel effects. In contrast, UAV imagery, with centimetre-level detail, provides rich surface texture and crop residue patterns that enhance classification accuracy. Key features extracted from the UAV imagery included vegetation indices (Vegetation Atmospherically Resistant Index), canopy cover, and texture metrics. These were used to train and validate multiple machine learning classifiers: Support Vector Machine (SVM), Random Forest (RF). Ground truth data were collected through field surveys and farmer consultations. Among the models, SVM achieved the highest classification accuracy of 82.3%, followed by RF (90.7%). The models showed significant improvement in detecting the cut and spread crop classes, which is typically indistinguishable in medium-resolution satellite imagery.