Developing a Spectral-Based Strategy for Urban Object Detection From Airborne Hyperspectral TIR and Visible Data

Abstract

Classification and detection of urban objects have been big challenges for years. High spatial resolution hyperspectral thermal infrared (HSR-HTIR) is a novel source of data that became available in recent years for urban object detection. In this research, a novel method is proposed for integration of HTIR and very high spatial resolution (VHSR) visible image to classify urban objects. First, atmospheric corrections were enforced to the HSR-HTIR. Second, for the first time, projection pursuit (PP) band reduction method was applied to a novel source of data, and the results achieved are better than those obtained by applying principal component analysis (PCA) as a well-known band reduction approach. Then, various features derived from HSR-HTIR and VHSR images were fed to a pixel-based support vector machine (SVM) classification algorithm, and seven urban classes detected. Afterward, an innovative strategy, using object-rule-based postprocessing approach, was introduced for postclassification of the raw classification results. Finally, a decision-based overlaying process was carried out to produce the final map. The classification results obtained indicate the high potential of using only spectral features. Consequently, its implementation becomes more feasible and the accuracies obtained are competitive in comparison to the results announced previously by the IEEE Geoscience and Remote Sensing Society (GRSS) Data Fusion contest 2014.