基于无人机多模态数据的滑坡“形变–温度”协同演化模型构建

Construction of “Deformation and Temperature” Co-Evolution Model for Landslides Based on UAV Multi-Modal Data

摘要: 滇东北小江断裂带复杂地质条件下的滑坡监测长期受制于传统手段空间分辨率有限及多参数协同分析不足等问题。以昭通小龙洞灰岩–粘土复合地层滑坡为研究对象，本文集成大疆M3T无人机多模态传感系统，开展高分辨率4K可见光影像(分辨率0.7 cm)与热红外影像(640 × 512，灵敏度50 mK)的同步获取，构建滑坡“形变–温度”协同演化模型。研究结果表明，基于改进的SfM算法实现的形变监测精度达到厘米级(相对精度1/1000)，空间分辨率较GNSS技术提升约两个数量级；坡肩区2.1℃ ± 0.4℃的热异常与滑带渗流表现出显著相关性R² = 0.86，在引入热弹性系数校正β = 0.08 MPa/K后，模型拟合精度提高23%。基于该模型构建的动态预警系统，将突发型滑坡响应时间由9小时缩短至2.5小时。研究成果不仅为复杂地形区滑坡的多参量耦合监测提供了新思路，也为边远地区地质灾害防治与应急管理提供方法技术和基础数据。

Abstract: Landslide monitoring in the Xiaojiang Fault Zone of northeastern Yunnan has long been constrained by the limited spatial resolution of conventional techniques and the lack of multi-parameter synergistic analysis. Taking the Xiaolongdong landslide, developed in a limestone-clay composite stratum in Zhaotong, as a case study, this paper integrates the DJI M3T UAV multi-modal sensing system to simultaneously acquire high-resolution 4K visible images (0.7 cm spatial resolution) and thermal infrared images (640 × 512, sensitivity 50 mK), thereby constructing a “deformation-temperature” co-evolution model of landslides. The results demonstrate that the improved Structure-from-Motion (SfM) algorithm achieves centimeter-level deformation monitoring accuracy (relative accuracy 1/1000), with a spatial resolution nearly two orders of magnitude higher than that of GNSS. A thermal anomaly of 2.1˚C ± 0.4˚C in the slope crest zone exhibits a strong correlation with slip zone seepage (R² = 0.86). After introducing a thermoelastic correction coefficient (β = 0.08 MPa/K), the model fitting accuracy improved by 23%. Furthermore, the dynamic early-warning system based on this model reduced the response time for sudden landslides from 9 hours to 2.5 hours. These findings provide a novel approach for multi-parameter coupled monitoring of landslides in complex terrains and offer methodological and data support for geological hazard prevention and emergency management in remote regions.