Integration of super-pixel segmentation and deep-learning methods for evaluating earthquake-damaged buildings using single-phase remote sensing imagery

Rapid identification of post-earthquake collapsed buildings can be used to conduct immediate damage assessments (scope and extent), which could potentially be conducive to the formulation of emergency response strategies. Up to the present, the assessments of earthquake damage are mainly achieved through artificial field investigations, which are time-consuming and cannot meet the urgent requirements of quick-response emergency relief allocation. In this research study, an intelligent assessment method based on deep-learning, super-pixel segmentation, and mathematical morphology was proposed to evaluate the damage degrees of earthquake-damaged buildings. This method firstly utilized the Deeplab v2 neural network to obtain the initial damaged building areas. Then, the simple linear iterative cluster (SLIC) method was employed to segment the test images so as to accurately extract the area boundaries of the earthquake-damaged buildings. Next, the images subdivided by SLIC can be merged according to the initial damaged building areas identified by Deeplab v2 neural network. Finally, a mathematical morphological method was introduced to eliminate the background noise. Experimental results demonstrated that the proposed algorithm was superior to others in both convergent speed and accuracy. Besides, its parameter selection was flexible and easily realized which was of great significance to earthquake damage assessments and provided valuable guidance for the formulation of future emergency response plans after earthquake events.