Abstract: Every minute of post-earthquake rescue is precious. Effective use of post-disaster data and accurate assessment of rescue routes can effectively avoid further damage caused by the earthquake. The earthquake did not stop at once, and the aftershocks after the disaster also threaten the lives and property safety of the people in the disaster area. Therefore, formulating a follow-up sustainable rescue plan based on the geology of the disaster area and the distribution of residents is the top priority of post-disaster rescue, which not only enables the rescue to be carried out steadily, but also allows for further planning of residential resettlement. This paper proposes an adaptive method of ground object segmentation algorithm by comparing neural network and traditional machine learning methods. According to the complexity of aerial imagery, neural network and traditional algorithm are combined to compare the identified ground objects with the original measurement data. Differential changes can be used to judge road distortions and geological changes after the disaster and provide an accurate scientific basis for rescue work.

Key words: earthquake, rescue, after disaster, Qinghai, neural networks, change detection