作者简介:薛洪文(1985-),男,测绘高级工程师,主要从事测绘工程。E-mail:shui8688@126.com
DOI: 10.13512/j.hndz.2026.01.07
备注
为提升多源数据在地质灾害监测预警中的应用能力,此研究构建了融合无人机摄影、地质传感器、卫星遥感等多源数据的监测预警系统。通过计算数据相似度并设定0.8的融合阈值,对未达标的数据采用5×5卷积核的三层卷积神经网络提取特征并二次融合,解决了多源异构数据的一致性问题。同时,提出深度核支持向量机(DKSVM)模型,利用深度学习自动提取地形坡度、土壤湿度等高阶特征关联,结合支持向量机的结构化风险最小化原理实现分类预测。试验选取滑坡、泥石流、地面塌陷等9类地质灾害与文献[3]~[4]的预警模型进行对比,结果表明:在滑坡、泥石流、地面塌陷等9类地质灾害中本研究系统的灾害预测准确率达98.7%,相较于对比文献效果显著提升,其中滑坡、泥石流等灾害类型的F1值均超过97.6%。数据融合后有效信息占比从65%提升至92%,平均相似度达0.89,验证了模型在复杂地质环境下的鲁棒性与泛化能力,为智能化地质灾害预警提供了技术参考。
To enhance the application capability of multi-source data in geological hazard monitoring and early warning, this study constructed a monitoring and early warning system that integrates multi-source data such as UAV photography, geological sensors, satellite remote sensing and other multi-source data. By calculating data similarity and setting a fusion threshold of 0.8,the system employs a three-layer convolutional neural network with 5×5 convolutional kernel to extract features and performs secondary fusion on data that does not meet the standard, addressing the consistency issue of multi-source heterogeneous data. Additionally,a deep kernel support vector ma‑chine(DKSVM)model is proposed,which utilizes deep learning to automatically extract high-order feature associa‑tions such as terrain slope and soil moisture, and combines the structured risk minimization principle of support vector machine to achieve classification prediction. The study selected nine types of geological hazards,including landslides,debris flows,and ground subsidence,and compared them with the early warning models in referenc‑es [3]to[4]. The results indicate that among the nine types of geological disasters such as landslide,debris flow and ground subsidence, the accuracy of disaster prediction of this system reached 98.7%, which is significantly im‑proved compared with the comparison literature,with F1 values for landslide,debris flow and other disaster types exceeding 97.6%. After data fusion,the proportion of effective information increased from 65 % to 92 %,with an average similarity of 0.89,verifying the robustness and generalization ability of the model in complex geological en‑vironments,and providing a technical reference for intelligent geological disaster early warning.
















一致性。 。其中
为Sigmoid函数,k=2为调节参数。设定融合阈值0.8,高于阈值数据直接融合,低于阈值数据通过卷积神经网络(CNN)提取特征,卷积核大小为 5×5,层数为 3 层,激活函数为ReLU。
