Earthquake-Induced Building Damage Assessment on SAR Multi-Texture Feature Fusion

Abstract

Multi-temporal RS data is not often available in time after an earthquake, so it is useful to assess the building situation with a single post-event SAR. Aiming at the problem that the single texture feature extraction method has inadequate information in the collapsed buildings classification, this paper proposes a SAR texture feature classification method that combines multiple features. Taking the 2016 Kumamoto earthquake as an example, the four methods based on gray histogram, GLCM, LBP, and Gabor filter are used to extract texture features and fused, then random forest classification is applied to obtain the collapse information of earthquake-damaged buildings. In addition, it is compared with the classification results of 26 texture features after principal component analysis. The results of two sets of experiments show that the extraction accuracy based on multi-feature fusion is higher than that of a single texture feature extraction method, and the multi-feature fusion classification result after principal component analysis improves the accuracy while improving the recognition efficiency.