Extraction of earthquake-collapsed buildings based on correlation change detection of multi-texture features in SAR images

An earthquake is a sudden natural disaster that results in the great loss of human lives and the compromised safety of properties. Rapid assessment of building damage after an earthquake is crucial for earthquake emergency rescue and damage assessment. Synthetic Aperture Radar (SAR) is an effective earthquake disaster analysis and evaluation method with a unique ability to overcome the impact of bad weather after an earthquake. Change detection based on SAR images is one of the important methods for remote sensing seismic information recognition. The conventional SAR image change detection method is mainly based on intensity images. Damaged buildings after an earthquake have complex and diverse forms, and no regularity exists in the intensity image. Therefore, capturing all the change information in the image under the set standard is difficult. However, the change in texture features is stable and cannot be affected easily by a change in ground features. Therefore, the inclusion of texture features in the calculation can completely obtain the change information of the image. Many parameters can describe texture features. If all the features are used, then the complexity of the algorithm is increased, and the feature information becomes redundant, thus reducing the accuracy of information recognition. To address irregular changes in intensity images after an earthquake and the numerous and difficult-to-optimize texture feature parameters, we propose a correlation change detection method based on principal component analysis of texture features. Principal component analysis is used to fuse multi-texture information based on the analysis of image texture features to avoid redundancy of features. Then, the window size is set to calculate the correlation between the extracted principal component components, and the correlation classification threshold is set for the detection of seismic building information. The process is mainly divided into four steps: (1) texture feature analysis; (2) principal component analysis; (3) correlation analysis; and (4) threshold setting and classification. The study considers Mashiki, which is the area that was most seriously damaged by the Kumamoto earthquake, as the study area. ALOS-2 SAR image data are used to verify the effectiveness of the proposed method, and the results are compared with those of the correlation change detection method on the basis of intensity image and with those of the difference change detection method on the basis of intensity image. Results show that the proposed method can efficiently extract buildings with different damage levels with an overall extraction accuracy that reaches 87.2%. The overall extraction accuracy is higher than that of the two change detection methods based on intensity images. The method not only obtains high extraction precision but also reduces misclassification probability. Principal component analysis can cover the useful information in the features and avoid the redundancy of features effectively. The change detection method based on SAR image texture feature can distinguish the intact and the destroyed buildings effectively. The proposed method can be used for allocating earthquake emergency rescue forces, damage assessment, and reconstruction. © 2018, Science Press. All right reserved.