Classification of Building Structure Types Using UAV Optical Images

Abstract

It is well know that for the same intensity areas, the buildings with different structure types can show different vulnerabilities. Thus, building structure type is one the key parameters for rapid estimation of casualties and injuries after earthquake, which is vital for emergency response and rescue. To estimate building structure types, the buildings are firstly extracted based on the spectrum, texture, and height information of UAV visible images. Then, the structure type of individual extracted buildings is classified using convolution neural network. To evaluate the accuracy of the proposed method, the images of Xuyi county, Huai'an City, Jiangsu Province are acquired using a small rotorcraft UAV. The results show that the user accuracy and cartography accuracy are 80.69% and 78.42%, respectively.