Automated building and road classifications from hyperspectral imagery through a fully convolutional network and support vector machine

Hyperspectral imagery facilitates the determination of various urban-correlated characteristics, such as features on the Earth's surface, including roads, trees, buildings, and natural and anthropogenic structures. Road network and building extractions are some of the main tasks necessary for emergency management, smart transport, and smart city systems. Currently, most researchers are focusing on deep learning and machine learning methods to classify high-resolution images. Machine learning algorithms have become important tools in modern hyperspectral image analyses. Hyperspectral images cover a variety of spectral bands with rather finite intervals in the electromagnetic spectrum and high spectral data resolutions. It is important to use high-resolution imagery to extract these features. The main objective of this research is to create a current technique for dimensionality reductions, automated building extractions, and road detections obtained from hyperspectral images. Given the above issues, this paper proposes a new dimensionality reduction and classification technique by combining the ICA, PCA, FCN, and SVM classification models. This new model extracts and classifies road and building features from hyperspectral imagery with the highest accuracy. The experimental findings obtained based on the ground truth of a Pavia University dataset and DC Mall dataset show significantly better accuracy compared to the results of existing machine learning approaches.