There may be victims in damage buildings after a big earthquake, we must rescue as soon as possible. When the earthquake occurred, we can use the aerial images effectively to grasp the situation of damaged buildings in the wide area. But the classify depends on visual interpretation by the experts and requires long processing time. In this paper, we approached the local complexity vector estimating method that be able to detect the collapsed structures automatically by SVM that used the local complexity vector by the building partition vector in the wide aerial images, and showed the result of the effectivity of the method. We detected the collapsed buildings to use the method from the aerial images immediately after the 2016 Kumamoto earthquake and compared the results to the visual interpretation by the experts. As the result, we could get the about 74.8% accuracy by about 1/40 detection time of the visual interpretation by the experts.