Supervised Machine Learning Based Land Cover Classification Using Sentinel Satellite Imagery

It is crucial to have accurate and updated cover maps for the producing sustainable crops and managing land efficiently. However, developing an accurate Land Cover (LC) mapping for efficient land management is still a major challenge in environment modeling and for agricultural regions with various landscapes that are complicated. To address this, a supervised Machine Learning (ML) based LC classification i.e., Support Vector Machine based LC (SVM-LC) is proposed which used Sentinel Satellite Imagery (Sentinel-2) dataset. The proposed research focuses on using the two parameters of SVM parameters: gamma and penalty, for identifying various crops and to classify LC. Initially, the dataset is preprocessed using Normalized Difference Vegetable Index (NDVI) and Land Surface Water Index (LSWI) followed by extracting features from, Histogram of Oriented Gradient (HOG), Local Gabor Binary Pattern Histogram Sequence (LGBPHS) and haralick texture features. The extracted features are then fed to the optimized SVM for choosing optimal parameters and classifying them for better LC classification. The results obtained from the proposed SVM-LC has shown better LC classification accuracy with 99.43% and precision with 97.46%. The results demonstrated the efficiency and applicability of the proposed approach for efficient land management in agricultural landscapes.