High-Resolution Spectral Reflectance-based Crop Classification and Chlorophyll Content Estimation Using Machine Learning

Precision agriculture progressively relies on remote sensing (RS) technologies to enhance crop classification and monitoring. Among various RS platforms, spectroradiometer offers the highest spectral precision, making them essential for validating the accuracy and performance of other RS methods. Each crop exhibits a unique spectral signature that corresponds to its biophysical characteristics. This spectral information plays a crucial role in accurately classifying crop types and assessing their health status, including water and nutrient availability. Specifically, evaluating crop chlorophyll content enables effective nitrogen management and yield optimization. This study focuses on collecting spectral data using a spectroradiometer (350-1050 nm) at a height of 30 cm above the crop canopy from eight crops, i.e., rice, finger millet, cotton, sunflower, sweet corn, broccoli, cauliflower, and brinjal, classifying the collected data, and measuring chlorophyll content using a Soil Plant Analysis Development (SPAD) meter and predicting the same using key spectral bands and machine learning (ML) techniques. Six supervised ML algorithms, i.e., Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Light Gradient-Boosting Machine (LGBM), Extreme Gradient Boosting (XGBoost), and Multi-Layer Perceptron (MLP) were employed for crop classification. The feature selection process revealed that the spectral range of 710-750 nm is the most significant for crop classification. The MLP model achieved the highest accuracy of 97% during training, 93% in testing, and 85% during validation stage, outperforming other ML classifiers. For chlorophyll content prediction, the RF demonstrated the best performance, with coefficient of determination values of 0.92 for training and 0.72 for testing stage. The ML-based framework, developed in this study, can be applied to various RS platforms, including satellites and unmanned aerial vehicles (UAVs), for crop classification and prediction of chlorophyll content. The developed modelling framework would assist government agencies and policymakers in identifying crop types accurately, enhancing agricultural planning, and optimizing resource allocation to support sustainable on-farm practices.