A Comparative Study of Three Supervised Algorithms for Mixed Crop Classification

Abstract

This study focuses on advancing precision agriculture through machine learning algorithms applied to crop classification using PlanetScope multispectral data in Kheda district, Gujarat. Three algorithms—Support Vector Machines (SVM), Spectral Angle Mapper (SAM), and Random Forests (RF)—were tested for their accuracy in classifying crop types. Additionally, the research utilized multi-temporal satellite imagery to monitor crop phenological cycles, enhancing classification reliability. The results highlighted SVM's boundary delineation, SAM's spectral similarity approach, and RF's ensemble learning as effective in distinguishing crops in mixed scenarios. Integrating ground truth data further validated classification accuracy, underscoring the study's contribution to improving agricultural management and planning.