Study on the intelligent extraction of seismic damage based on the Mean-Shift segmentation

The limitations of parameter determination manually exist in the current object based image analysis (OBIA) of the commercial software, such as Definiens eCognition or ENVI FX, In this paper, an intelligent method based on Mean-Shift and support vector machines (SVM) algorithm of OBIA is proposed to extract the building damage caused by catastrophic earthquakes, which aims to improve the accuracy of classification and get the best bandwidth parameter of the Mean-Shift segmentation automatically by computing the Kappa index that has been used in a feedback loop. The improved method is applied to extract seismic damage information in a test area of the city of Dujiangyan after the 2008 Wenchuan earthquake by using post-earthquake aerial images acquired on May 18, 2008. The results of the experiments indicate that the improved OBIA is more effective and robust than traditional method of OBIA.