Crop Discrimination and Classification Using Sentinel-2A and Machine Learning Techniques

Abstract

Precision agriculture and agricultural monitoring rely heavily on accurate crop identification.  This study examines the ability of Support Vector Machine and Random Forest classifiers to distinguish cotton, wheat, maize, and sugarcane using Sentinel-2 imagery.  Both classifiers were trained and tested using ground truth data, with accuracy determined by the User's Accuracy, Producer's Accuracy, Overall Accuracy, and Kappa coefficient. The outcome revealed RF as having a better OA (83.33%) and Kappa value (0.77) than SVM (66.66% OA, 0.55 Kappa), reflecting a better classification performance. RF also had a better validation accuracy (97.91%) than SVM (95.83%). Although both models performed well in classifying crops, moderate UA values reveal room for improvement. The research emphasizes the efficacy of RF for crop categorization and recommends future enhancement with the use of sophisticated machine learning methods and multi-temporal data fusion for more efficient agricultural surveillance.