欢迎访问《金属矿山》杂志官方网站,今天是 分享到:
×

扫码分享

金属矿山 ›› 2026, Vol. 55 ›› Issue (10): 290-295.

• 地质与测量 • 上一篇    下一篇

融合VGG16与InSAR的矿区开采沉陷边界提取

郝 蓉1 ,牛玉芬2 ,柳 艳1   

  1. 1.山西职业技术学院建筑工程系,山西 太原 030006;2.河北工程大学矿业与测绘工程学院,河北 邯郸 056038
  • 出版日期:2026-11-03 发布日期:2026-11-10
  • 通讯作者: 牛玉芬(1987—),女,讲师,博士。
  • 作者简介:郝 蓉(1990—),女,讲师,硕士。
  • 基金资助:
    国家自然科学基金项目(编号:42307255);河北省自然科学基金项目(编号:D2023402033);河北省教育厅科学研究项目(编号: QN2024231)。

Mining Subsidence Boundary Extraction in Mining Areas by Fusing VGG16 and InSAR

HAO Rong1 ,NIU Yufen2 ,LIU Yan1   

  1. 1.Department of Architectural Engineering,Shanxi Polytechnic College,Taiyuan 030006,China; 2.School of Mining and Geomatics Engineering,Hebei University of Engineering,Handan 056038,China
  • Online:2026-11-03 Published:2026-11-10

摘要: 矿区开采引发的地表沉陷严重威胁了矿区安全与生态环境,传统监测方法存在效率低、成本高等局限, 为提升沉陷边界提取的自动化与精度,提出了一种融合视觉几何组16层卷积神经网络(Visual Geometry Group 16-lay er Convolutional Neural Network,VGG16)网络与合成孔径雷达干涉测量(Synthetic Aperture Radar,InSAR)技术的新方 法。该方法通过InSAR获取矿区时序形变数据,结合VGG16的多尺度特征提取能力实现边界智能分割。我国西北某 典型煤矿区试验表明:在地表覆盖类型复杂区域,VGG16边界提取精度更高,最高可达97.6%,平均精度约95%;RF 算法提取精度次之,最高96.4%,平均精度约92%;VGG16网络的分类准确率稳定性更好,最高可达95.4%;RF算法 分类效果次之,最高94.6%。对于对于沉陷程度差异明显区域,VGG16最高边界提取精度保持在98.2%,平均精度约 96%;RF最高97.8%,平均精度约91%;VGG16最高分类准确率达到91.4%,RF次之,最高为87.2%。研究反映出, 该方法有助于提升矿区开采沉陷边界的提取精度,可为矿区沉陷监测与生态修复提供技术参考。

关键词: 开采沉陷 , VGG16 , InSAR , 边界提取

Abstract: Surface subsidence caused by mining severely threatens the safety and ecological environment of mining areas. Conventional monitoring methods suffer from limitations such as low efficiency and high cost.To enhance the automation and accuracy of subsidence boundary extraction,this study proposes a novel method that integrates a Visual Geometry Group 16-lay er convolutional neural network (VGG16) with interferometric synthetic aperture radar (InSAR).In this method,time-series deformation data of the mining area are acquired using InSAR,and the multi-scale feature extraction capability of VGG16 is employed to achieve intelligent boundary segmentation.Tests in a typical coal mining area in northwestern China showed that, In areas with complex land cover types,VGG16 achieved higher boundary extraction accuracy,reaching up to 97.6%,with an average accuracy of about 95%;the RF algorithm ranked second in extraction accuracy,with a maximum of 96.4% and an av erage accuracy of about 92%;the VGG16 network showed better stability in classification accuracy,reaching up to 95.4%;the RF algorithm ranked second in classification performance,reaching up to 94.6%.For areas with significant differences in sub sidence degree,VGG16 maintained the highest boundary extraction accuracy at 98.2%,with an average accuracy of about 96%;RF reached a maximum of 97.8%,with an average accuracy of about 91%;VGG16 achieved a maximum classification accuracy of 91.4%,followed by RF,with a maximum of 87.2%.These findings indicate that the proposed method can improve the accuracy of mining-induced subsidence boundary extraction and provide a technical reference for subsidence monitoring and ecological restoration in mining areas.

中图分类号: