Classification of RASAT Satellite Images Using Machine Learning Algorithms

The development in the remote sensing and geographic information systems facilitated the monitoring processes of changes in land cover and use. This article aimed to evaluate the classification accuracy of five supervised classification methods: Neural Network, Naive Bayes, K-nearest neighbors, discriminant analysis and Decision Tree using the Turkish RASAT satellite images. The Bursa area in Turkey was taken as a study area to examine the RASAT satellite images. MATLAB and Python programming languages were employed to develop the training dataset and generated the five classifiers. According to the performance analysis using confusion matrix metric, the best overall accuracy was achieved by K-nearest neighbors. the K-nearest neighbors method produced 100% performance accuracy using RASAT satellite image. This comparative analysis showed that the K-nearest neighbors can be used as a trusted method for satellite image classification.