遥感自动提取技术在房屋抗震调查中的应用

Application of the automatic extraction method of remote-sensing images to the seismic capacity investigation of buildings

摘要:

房屋抗震能力调查对于全面摸清地震灾害风险底数,预测地震灾害损失具有十分重要的意义。传统实地调查方法难以大范围开展,而依靠经验估计和人工解译的遥感房屋抗震能力评价的效率仍有待提高。针对此问题,文章以深度学习遥感目标识别算法为基础,提出了多尺度聚合的房屋自动提取方法,并将该方法应用于湖北省房屋抗震能力遥感初判,自动提取房屋建筑共计1 060万余栋,并对其抗震能力进行了分类判别。经与试点县实地调查结果对比,文章方法房屋提取误差总体在10%以内,房屋抗震能力判别准确度在72.3%~90.9%之间。实验结果表明文章方法可为全国自然灾害风险普查工程房屋调查工作提供技术支持。

Abstract:

The investigation of the seismic capacity of buildings is of great significance to comprehensively understand the risk of earthquake disasters and predict the disasters caused by earthquakes. The traditional field investigation method is difficult to apply on a large scale, while the efficiency of the remote-sensing seismic capacity evaluation of buildings, which relies on empirical estimation and manual interpretation, needs improvement. To address the above issues, we propose a multiscale aggregated building extraction method based on a remote-sensing target recognition deep-learning algorithm. The proposed method is applied to the remote-sensing preliminary seismic capacity evaluation of buildings in Hubei Province. Over 10.6 million buildings are automatically extracted, and their seismic capacity is classified and determined. Compared with the field-investigation results in pilot counties, the overall error of building extraction using this method is less than 10%, and the evaluation accuracy of the seismic capacity of buildings is between 72.3% and 90.9%. The experimental results show that the proposed method can support the seismic capacity investigation of buildings in the framework of the national natural disaster risk census project.