| 28 | 0 | 78 |
| 下载次数 | 被引频次 | 阅读次数 |
落石灾害是一种常见的地质灾害，基于视频图像的落石形态识别是研究落石运动性质及灾害防治的基础。本文建立了一种从落石视频图像直接获取落石形态特征的识别方法。针对落石视频图像，采用工程实测法和隐式标注法建立了落石影像COCO格式数据集，并采用PA-NET模型对落石进行了实例分割，提取了落石掩膜。在此基础上，引入了扁平度、球形度、形状因子等评估落石形态特征的几何参数，并使用实例分割提取的掩膜计算得到了落石形态特征的目标参数。与手工标注的落石几何参数对比分析表明，本文采用PA-NET模型的落石形态特征识别精度满足工程需要，能够为落石灾害的定量评估与防治决策提供技术支持。
Abstract：Rockfall is a common geological hazard, and the recognition of rockfall morphological features based on video imagery serves as a foundation for studying rockfall dynamics and implementing disaster prevention strategies. This study proposes a method for directly extracting rockfall morphological features from video images. A COCO-format dataset of rockfall imagery was constructed using engineering field measurements and implicit annotation methods. The PA-NET model was then applied to perform instance segmentation and extract rockfall masks. Based on the segmentation results, several geometric parameters such as aspectrations sphericity, shape factor were introduced to evaluate the morphological characteristics of the rockfalls. These parameters were calculated from the extracted masks. Comparative analysis with manually annotated geometric parameters demonstrates that the PA-NET-based method achieves sufficient accuracy for engineering applications and offers technical support for quantitative assessment and mitigation of rockfall hazards.
[1]赵屹峰，陈磊磊.泥石流灾害及工程防治对策研究[J].建设科技，2024(2):28-30.ZHAO Yifeng,CHEN Leilei. Research on Debris-flow Hazards and Engineering Prevention and Control Strategies.Construction Science&Technology,2024(2):28-30.
[2]MOOSAVI V,JABOYEDOFF M. Event-triggered Natural Hazard Monitoring with Convolutional Neural Networks on the Edge[C/OL]//IEEE. Proceedings of the IEEE International Smart Cities Conference(ISC2),2018[2024-08-10].http：//DOI:10.48550/arXiv.1810.09409.
[3]伊小娟，罗威，李伟，等.基于Mask R-CNN的地质雷达岩溶预报图像识别研究[J].高速铁路技术，2024(2):78-82.YI Xiaojuan,LUO Wei,LI Wei,et al. Mask R-CNN-Based Image Recognition for Karst Advance Prediction Using Ground-Penetrating Radar(GPR)[J]. High-Speed Railway Technology,2024(2):78-82.
[4]BUSCOMBE D,RITCHIE A. Automated Mapping of Coastal Protection Structures Using Deep Learning with PA-NET and High-resolution Orthoimagery[J].Coastal Engineering,2020,161:103751.
[5]WADELL H.Volume,Shape,and Roundness of Rock Particles[J].Journal of Geology,1932,40(5):443-451.
[6]BARRETT P J. The Shape of Rock Particles—a Critical Review[J].Sedimentology,1980,27(3):291-303.
[7]LIU S,QI L,QIN H,et al. Path Aggregation Network for Instance Segmentation[C/OL]//IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2018:8759-8768[2024-08-16].http//：DOI:10.1109/CVPR.2018.00913.
[8]李俊孟，黄艳利，浦海，等.矸石块体三维形状参数自动获取方法与量化分析[J].中国矿业大学学报，2021,50(3):507-514.LI Junmeng,HUANG Yanli, PU Hai,et al. Automatic Acquisition Method and Quantitative Analysis of ThreeDimensional Shape Parameters of Gangue Blocks[J].Journal of China University of Mining&Technology,2021,50(3):507-514.
[9]WELZL E. Smallest Enclosing Disks(Balls and Ellipsoids)[C/OL]//New Results and New Trends in Computer Science,1991:359-370[2024-07-15].http//：DOI:10.1007/BFb0038202.
[10]GRAHAM R L. An Efficient Algorithm for Determining the Convex Hull of a Finite Planar Set[J].Information Processing Letters,1972,1(4):132-133.
基本信息:
中图分类号:P694;TP391.41
引用信息:
[1]蔡堃,陈建译.基于PA-NET模型的落石形态特征识别方法[J].铁道建筑,2025,65(11):67-72.
基金信息:
中国铁路广州局集团有限公司科研项目(2021K072-z)