Deep Learning Implementation on SAR Data for Built-Up Extraction: Case Study Hyderabad City

India is growing at a speedier pace, and with the largest population in the world as country and of the population is in urban areas, is still growing. As per the 2011 census, 30% of the population resides in urban India, with 8000 cities. To streamline the growth and harness resources, the Indian Government of India has various projects such as AMRUT, Jal Jeevan Mission, Jal Shakti Abhiyan, etc. Monitoring built-up growth through satellite data is vital for regional planning and development. Microwave data has the advantage of day-night acquisition and cloud penetration, which can be exploited for many applications in areas usually covered by clouds. The urban areas, which are growing exponentially, need to be monitored, for which satellite data provides an excellent source because of its temporal coverage. The increasing availability of SAR satellite data using various bands such as C and S from Indian and other satellite missions enables the exploration of the potential of information extraction in different applications such as agriculture, Urban & Infrastructure, etc., While Advancement of data processing techniques, and computational infrastructure enable conceptualize and develop the application based on Deep Learning (DL) techniques in built-up extraction. The current study is conducted to process SAR data from NOVASAR, Sentinnel-1, and EOS 4 satellites using deep learning (DL) techniques that have been used for built-up area extraction. A Dense Neural Network (DNN) based model using Python, Tensor Flow, Keras and GDAL have been used. Hence, the current study aims to develop a Dense Neural Network (DNN) model to extract the built-up areas over Hyderabad, India, from multiple SAR satellite data, i.e., Sentinel-1, NOVASAR and EOS-4, using various polarizations datasets. It is found that EOS-4 datasets gave the best results among the three.