Texture-wavelet analysis for automating wind damage detection from aerial imageries

In this era of climatic changes, heavy wind damages to buildings and structures have become a major issue. Impacts of such disasters can be reduced by a rapid but accurate identification of damaged location and faster reconstruction. In this paper, automatic roof-damaged building identification as well as the damaged area pattern recognition is carried out from aerial images using a novel technique, texture-wavelet analysis. Initially, wavelet features extraction, followed by feature selection using a decision tree algorithm and Support Vector Machine (SVM) classification is performed for roof-damaged building identification. After separating the damaged buildings from the non-damaged ones, the pattern of the roof-damaged area is attained using texture-wavelet analysis and finally percentage area of damaged roof is measured. Comparison is done with the conventional feature extraction methods. The validation is performed with the manually obtained data as well as the field survey information.