基于BP神经网络的GB-SAR大气改正模型

A GB-SAR Atmospheric Correction Model Based on BP Neural Network

摘要: 基于微波传感器成像的新型地基合成孔径雷达(GB-SAR)系统，对于地表形变的干涉测量精度达到亚毫米级，兼具遥感和地面测量途径的技术优势，在地质灾害调查及应急监测中的应用潜力巨大。因地基SAR系统的视场范围较大，监测区域内的水汽扰动对精度的影响较为显著，如何保证连续成像监测过程的精度，一直是该应用研究领域中的核心问题。针对现有地基SAR大气改正方法的局限与不足，本文提出基于BP神经网络大气改正模型，并以宝兴县硗碛水库边坡为典型研究区域，论证该模型的可靠性和可行性。实验结果表明，BP神经网络大气改正模型的使用，在温度和水汽变化较快的日间连续监测过程中取得了较好的应用效果，总体精度控制在±0.7 mm以内。同期对滑坡的监测分析亦揭示了滑坡点具体位置及发展态势，获得蠕变滑移速率接近8 mm/天，表明该滑坡对周边道路、建筑的安全性与稳定性已构成威胁。

Abstract: As a newly arising microwave sensor imaging system, ground-based synthetic aperture radar (GB-SAR) integrated the advantages of both remote sensing and ground measurement, and has great potential in geological disaster investigation and emergency monitoring. Because the GB-SAR system has a wide field of view, the water vapor disturbance in the local monitoring area may induce significant influence to the monitoring accuracy. How to ensure the precision of the continuous imaging monitoring process has always been the core issue in the application research field. To overcome the limitations and shortcomings of existing GB-SAR atmospheric correction methods, this paper proposed an atmospheric correction model based on BP neural network. For validation purpose, the landslide around Qiaoqi water reservoir in Baoxing County, Ya’an City, was selected as a typical study area to implement the experiment. The statistical analysis showed that the BP neural network atmospheric correction model has achieved good application results in continuous monitoring process by the situation of the rapid temperature and water vapor changes during the daytime. The overall precision is controlled within ±0.7 mm. In addition, the monitoring and analysis results of the landslide also revealed the specific location and development trend of the landslide point. It is worth noting that the creep slip rate is close to 8 mm/day, which indicated that the landslide poses a threat to the safety and stability of surrounding roads and buildings.