Land Use/Cover Novel Dataset Based on Deep Learning: Case Study of Fayoum, Egypt

Abstract

AbstractLand use and land cover classification is essential for monitoring studies, resource management, and planning activities. The accuracy of such classification depends on set of factors such as the resolution of input data/images and the size of dataset. The resolution and size of input dataset/images determine the definitions of classes; So, the low resolution images define low number of classes. We select a case study of Fayoum governorate, Egypt for this research; the previous land use/cover classification researches for that studying area are applied on low resolution satellite images with 30m spatial resolution and small size of dataset. These researches were conducted using supervised classifiers; the objectives of this context are building a novel land use/cover dataset for Fayoum governorate using deep learning model. In the proposed study, we used images with high resolution of 3 m, the proposed dataset is called FayPDT. The proposed classification model is trained and conducted on two types of data; raster and vector data; using deep and machine leaning algorithms such as Artificial Neural Network (ANN), Random Forests (RF), and SVM. The proposed classification model is called Deep learning classification model. The Deep classification model is tested with the precision, recall, f-score and kappa index, and the test result for the proposed model is 97.1% overall accuracy for Artificial neural network.KeywordsDeep learningMap processingClassifier models