Abstract. Remote sensing satellite images plays a significant role in mapping land use/land cover LULC. Machine learning ML provide robust functions for satellite image classification. The objective of this paper is to extend the capability of GIS specialists in geospatial area with minimum knowledge in computer science to easily perform ML satellite image classification. A framework consisting 7 stages established. Tools of steps developed in two programing environments, which are ArcGIS for geospatial datasets structuring and Anaconda for ML training and classification. During the development, authors constrained to reduce the complexity of big data of satellite images and limited memory of computers to make tools available for implementation in PC. In addition, automation and improving the performance accuracy. TensorFlow-Keras library employed to perform the classification using neural networks. A case study using RASAT satellite image in Ankara-Turkey utilized to perform the analysis. The developed classifier gained 80% performance accuracy. The complete RASAT satellite image processed and smoothly classified based on blocks methods. The developed tools successfully tested and applied in geospatial area and can be effectively execute in PC by GIS specialist.
REVIEW ON SATELLITE IMAGE CLASSIFICATION BY MACHINE LEARNING AND OPTIMIZATION APPROACHES
