A research on feature ontology for hyper-spectral remote sensing images

Abstract

Hyper-spectral remote sensing has a promising application and is a challenging technology about the information extraction from a large amount of spectral data. Previous researches have been concerned on various spectral analysis methods, different approaches of data fusion techniques from both the spectral and spatial data, and even the temporal data, among which feature extraction is the core issue. On the other hand, the technology of ontology has been proved to be of great help to the knowledge representation and reasoning, which is imperative to the information extraction procedure. Our research was based on our previous work on the feature extraction approach on the shape adaptive neighborhood (SAN), devoting to build an efficient ontology towards the features of the spectral, spatial and temporal data of hyper-spectral sensing images. In this paper, the feature ontology of hyper-spectral sensing images was design to contain three main groups of features: 1) the spectral features, including the original spectral curve, the envelope of the curve, a set of mathematical features of the curve such as the derivation and the integration, and the indices of the spectrum absorption; 2) the spatial features, including the location features, the textural features such as the statistical features, the structural features and the random distribution model of the texture, and the shape features on the compactness, the complexity, and the topology of the shapes; and 3) the temporal features, such as the date and the time attributes of the images, and the detected changes from images of different time phases. The ontology was implemented by using the famous ontology editor and knowledge-base framework Protégé, and could be exported into a variety of formats such as RDF(S), OWL, and XML Schema. Copyright © (2009) by the Asian Association on Remote Sensing.