A Futuristic Deep Learning Framework Approach for Land Use-Land Cover Classification Using Remote Sensing Imagery

摘要

Our aim is to propose a new deep learning framework approach which uses an ensemble of convolutional neural network(CNN)for land use-land cover mapping. Every CNN layer was fed with diverse combination of multispectral and geospatial satellite bands provided by Sentinel 2 satellite imagery(spatial resolution of 10 m), topographic and derived texture parameters, of New Delhi(28.6139° N, 77.2090° E)region, India. Several classes were identified like forest, parking, residential areas, slums, wasteland, water bodies. It was observed that our proposed framework outperformed with classification accuracy of 89.43%, compared to the current state-of-the-art algorithms(support vector machine(SVM), K-nearest neighbor(KNN), and random forest(RF)). Accuracy assessment was done by means of following statistic measures(precision, recall, specificity, and area under curve(AUC))and receiver operating characteristic(ROC)curve.