Objects recognition in urban environment using multiband imagery, is a difficult process, implying the use of elaborated and complex image processing methods, which are used to enhance the detection efficiency. The urban mosaics are characterized by multiple materials (e.g. manmade, urban vegetation, bare soil, transport infrastructure, etc.), which are combined together to form a complex patchwork. This study aim to take advantage of the multiband imagery, to assess the feasibility degree of the urban objects detection, and to explore some of the applications related to the multiband hyperspectral imagery classification.
AbstractUrban objects classification by spectral library: feasibility and applications JURSE 2017, DUBAI, 06-08 Mars 2017 1. Hyperspectral data and spectral library acquisition 2. Method
Spectral variability
Urban objects recognitionAn original technique based on the urban objects identification by spectral library, using airborne hyperspectral imagery, is presented in this study. The potential to distinguish some of the common roofing's materials in the city of Kaunas was demonstrated. Nevertheless, some issues need to be addressed in the future, as the high interclass and intraclass correlations regarding some urban materials, leading to misclassifications. Asbestos type roofing's and bitumen roofing's were difficult to identify and presents high correlation with the other materials like asphalt roads and pavements. To improve the spectral classification results, we proposed a morphological rectification based on attributes filtering, and rules construction, the rectification seemed suitable, and permits to eliminate the misclassified pixels (i.e. artefacts, roads, shadows, etc.). For the future, we are planning to 1) enrich the spectral library with more materials, 2) test other datasets including more bands, and 3) use other classifiers (e.g. SVM, neural networks).