A Deep Learning Framework Approach for Urban Area Classification Using Remote Sensing Data

Abstract

The main aim of this study is to propose a Deep Learning framework approach for Urban area classification. The research proposes a multilevel Deep Learning architecture to detect the Urban/Non-Urban Area. The support models/parameters of the structure are Support Vector Machine (SVM), convolution of (Neural Networks) NN, high resolution sentinel 2 data, and several texture parameters. The experiments were conducted for the study region Lucknow which is a fast-growing metropolis of India, using Sentinel 2 satellite data of spatial resolution 10-m. The performance observed by the proposed ensembles of CNNs outperformed those of current state of art machine algorithms viz; SVM, Random Forest (RF) and Artificial Neural Network (ANN). It was observed that our Proposed Approach (PA) furnished the maximum classification accuracy of 96.24%, contrasted to SVM (65%), ANN (84%) and RF (88%). Several statistical parameters namely accuracy, specificity, sensitivity, precision and AUC, have been evaluated for examining performance during training and validation phase of the models. © 2020, Springer Nature Singapore Pte Ltd.