Information Extraction Of Damaged Buildings Caused Byearthquake With The Object-oriented Method From High Spacial Resolution Image

In recent years, remote sensing technology have received a great deal of development, it has become an effective means for disaster monitoring and damage assessment, especially played an indispensable role in the both disaster information acquisition and emergency response after earthquake role. At the damage assessment of earthquake disaster stage, high-resolution remote sensing data can be used to getting the damage information of buildings and road more precisely, but the traditional manual extraction method is inefficient while ensuring the accuracy. The paper presents the object-oriented image classification method to extract the damaged buildings with the information of spectrum, texture, shape which discovered from high-resolution aerial images. The research work and results are as follows:1) Reached the algorithm of remote sensing image segmentation, the method of the image characteristics of the object quantitative analysis and the classification commonly used in classification of object-oriented information extraction techniques. Because of the complex image characteristics after the earthquake, edge pixels of adjacent features is mixed in a high level, therefore established using fuzzy classification method based on membership function to extract the damage buildings.2) This paper proposes a scale selection evaluation method based on texture feature of damaged buildings, the optimal segmentation scale respectively by the standard deviation and Moran index to describe objects within the "homogeneity" and "heterogeneity" between objects, the texture features as evaluation factors to calculate the similarity between samples and objects, through the above index comprehensive evaluation of segmentation scale, the method on the premise of guarantee the global optimal segmentation, also take into consideration the extraction fall damage the structure of the segmentation scale requirements.3) Constructed a random forest classification model which is applicable to damage buildings extraction: on the basis of studying the mathematical definition, classification process and advantage of the random forest classification, selecting the appropriate number of samples in the experimental zone to train a classifier determines the optimal classifier parameters, participate in the characteristic parameters of classification, achieved good results in the damaged buildings extraction process.4) The classification of damaged buildings, using the method based on template matching, according to the expert experience to select different damage degree of damage of building object as a template, take the multi-scale multi-level segmentation strategy on the basis of random forest classification, according to the object within the gray statistical characteristic using hierarchical object by matching, the damaged building object classification. Through the precision analysis, random forest classification method obtains a good result, and compared with the manual extraction of damaged buildings classification result in this method has certain availability.