IMPROVED IN-SEASON CROP CLASSIFICATION PERFORMANCE USING ENSEMBLE LEARNING TECHNIQUE: A CASE STUDY OF LEKODA INSURANCE UNIT, UJJAIN, MADHYA PRADESH

Abstract

The classification of agricultural crop types is an important application of remote sensing. With the improvement in spatial, temporal and spectral resolution of satellite data, a complete seasonal crop growth profile and separability between different crop classes can be studied by using ensemble-learning techniques. This study compares the performance of Random Forest (RF), which is a decision tree based ensemble learning method and Naïve Bayes ( a probabilistic learning technique) for crop classification of Lekoda gram panchayat, Ujjain district, using multi-temporal Sentinel 2 of Rabi 2017–18. The study area contains seven different classes of crop types, and in each class, we have used 65% of the ground data for training and 35% to test the classifier. The performance of RF classifier was found to be better than NB classifier. Kappa coefficient of RF classifier in mid of the crop season (December–January) was found to be 0.93. This result indicates that an accurate in-season crop map of the study area can be generated through integrated use of Sentinel 2 temporal data and RF classifier.