Small to marginal farms with diverse crops and management practices pose a formidable challenge in implementing crop classification algorithms. Coarse-resolution satellite imagery cannot accurately represent fragmented land parcels, for performing crop classification and yield assessment studies. The Indian agro-climatic environment is especially suited to this type of condition. This study aims to classify the crops based on satellite images with very high resolution using machine learning algorithms. The Cartosat-2E satellites has a spatial resolution of 1.6 m, utilized to classify crops in marginal/fragmented land systems. It is widely used to classify the croplands more efficiently and accurately from the data obtained by the field studies in conjunction with data obtained from remote sensing techniques, and using artificial intelligence and machine learning methodologies. This study was carried out in Nandikandi, Sanga Reddy district, Telangana state, India using Cartosat-2E imagery to classify croplands in a fragmented system. The accuracy metrics of both pixel-based and object-based approaches for classifying the fragmented landscapes were compared using Random Forest (RF) and Support Vector Machine (SVM) classification algorithms. It was observed that parcel-based classification using SVM and RF techniques yielded better results with an overall accuracy of 90.74% (F-1 Score of 0.8) and 86.89% (F-1 Score of 0.78), respectively. Our results conclude that SVM outer performed the RF classifier with a parcel-based approach, which can effectively differentiate cropland for crop productivity and damage assessment models.