改进型金字塔解析网络在矿区浮石识别中的应用研究

郭梨; 吴昊; 顾清华; 黄智奇; 李照康

doi:10.13637/j.issn.1009-6094.2025.047610.13637/j.issn.1009-6094.2025.0476

Chinese

您当前的位置：

首页 >

文章列表页 >

改进型金字塔解析网络在矿区浮石识别中的应用研究

安全评价安全工程 | 更新时间：2026-04-17

- 改进型金字塔解析网络在矿区浮石识别中的应用研究
- 改进型金字塔解析网络在矿区浮石识别中的应用研究
- 安全与环境学报 2026年26卷第4期页码：1307-1315
- 作者机构：
  
  西安建筑科技大学资源工程学院
- 作者简介：
- 基金信息：
  
  国家自然科学基金项目(52474179)
- DOI：10.13637/j.issn.1009-6094.2025.047610.13637/j.issn.1009-6094.2025.0476
  中图分类号： TD804;TP391.41;TP18
- 纸质出版：2026
- 稿件说明：
移动端阅览
[1]郭梨,吴昊,顾清华,等.改进型金字塔解析网络在矿区浮石识别中的应用研究[J].安全与环境学报,2026,26(04):1307-1315.
[1]郭梨,吴昊,顾清华,等.改进型金字塔解析网络在矿区浮石识别中的应用研究[J].安全与环境学报,2026,26(04):1307-1315. DOI： 10.13637/j.issn.1009-6094.2025.047610.13637/j.issn.1009-6094.2025.0476.

DOI：10.13637/j.issn.1009-6094.2025.0476.

摘要

为解决露天矿区复杂环境下浮石识别与特征提取的技术难题，提出了一种基于动态卷积改进型金字塔解析网络模型的浮石检测与三维重建方法。针对浮石区域分布复杂、小目标特征难以捕获的问题，优化金字塔解析网络模型，引入动态卷积和通道注意力机制，增强特征提取能力和分割精度；结合ZED双目相机获取的深度图，与分割结果生成三维点云模型，通过凸包算法实现浮石几何特征和体积的精准测量。在自制浮石数据集上的试验表明，改进模型的mIoU达到81.94%

比原始模型提升5.49百分点；mDice和mPA分别提升至90.90%和89.98%;体积、质量、形状因子和初始高度识别的平均准确率分别达到87.65%、85.55%、82.38%和96.15%。研究表明，改进的金字塔解析网络模型在复杂矿区环境中展现出卓越的浮石检测与几何信息提取能力，为矿区浮石清理与安全评估提供了可靠的技术支持。

Abstract

Loose rocks scattered on open-pit slopes pose significant safety hazards to mining operations by threatening equipment

workers

and production stability. To address the limitations of traditional laser scanning and manual inspection methods

this study develops an enhanced Pyramid Scene Parsing Network(PSPNet) combined with stereo vision-based 3D reconstruction for precise loose rock detection and quantitative feature extraction. The original PSPNet is improved by integrating dynamic convolution

depthwise separable convolution

and Squeeze-and-Excitation(SE) attention mechanisms. These modifications significantly enhance the network's ability to capture multi-scale features and details of small objects while maintaining computational efficiency. A custom dataset comprising 1 205 annotated images of loose rocks was created

encompassing diverse lighting conditions

rock sizes

and complex backgrounds. Extensive data augmentation techniques—such as rotation

scaling

brightness adjustment

Gaussian noise addition

and flipping—were applied to improve the model's robustness. Model training was conducted using PyTorch over 500 iterations on an NVIDIA RTX 4060 GPU

ensuring convergence and stable performance. The segmentation results indicate that the improved PSPNet achieved a mean Intersection over Union(mIoU) of 81.94%

representing a 5.49 percentage-point improvement over the baseline PSPNet. Meanwhile

the mean Dice coefficient and pixel accuracy increased to 90.90% and 89.98%

respectively

confirming its superior capability to delineate small and irregular rock boundaries. Comparative experiments with U-Net

Segformer

DeepLabV3

Mask2 Former

and ConvNeXt further validated the enhanced model's overall advantages across multiple evaluation metrics. For geometric analysis

ZED stereo cameras were employed to acquire depth maps

which were fused with segmentation masks to generate 3D point clouds. A convex hull algorithm

combined with dynamic regional optimization and depth-weight correction

enabled accurate reconstruction of rock morphology. The estimation accuracies for rock volume

mass

shape factor

and initial height reached 87.65%

85.55%

82.38%

and 96.15%

respectively. In conclusion

the proposed framework effectively integrates deep learning-based segmentation with stereo vision-driven 3D reconstruction

providing a reliable tool for loose rock detection

quantitative hazard assessment

and safety management in open-pit mining environments. This study not only enhances the precision of rock feature extraction but also contributes to advancing intelligent monitoring and risk control technologies in complex mining operations.

关键词

Keywords

references

叶剑，盛建龙.地下矿山溜井浮石的探测及处理方法研究[J].化工矿物与加工，2014,43(10):27-30.YE J,CHENG J L.Detection and processing method for pumice in chute of underground mines[J].Industrial Minerals & Processing,2014,43(10):27-30.

YANG Z,WU H,DING H J,et al.Enhanced U-Net model for rock pile segmentation and particle size analysis[J].Minerals Engineering,2023,203:108352.

李刚，杜亚波，杨庆贺，等.基于多尺度特征融合和注意力机制的矿区道路障碍检测[J].中国安全科学学报，2024,34(9):87-98.LI G,DU Y B,YANG Q H,et al.Obstacle detection on mining roads based on multi-scale feature fusion and attention mechanism[J].China Safety Science Journal,2024,34(9):87-98.

刘辉，李国强，朱晓峻，等.基于深度学习的井筒变形预测模型与应用[J].煤炭学报，2025,50(2):732-747.LIU H,LI G Q,ZHU X J,et al.Exploration and application of deep learning based wellbore deformation forecasting model[J].Journal of China Coal Society,2025,50(2):732-747.

陈根深，刘刚，陈麒玉，等.基于多源数据集成与Web-GIS的三维矿山空间数据管理系统关键技术及应用[J].软件导刊，2023,22(11):18-28.CHEN G S,LIU G,CHEN Q Y,et al.Key technologies and applications of 3D spatial data management system for mining based on multi-source data integration and Web-GIS[J].Software Guide,2023,22(11):18-28.

董宏翔，安毅，谢丽蓉，等.基于局部特征增强的矿山大规模激光点云语义分割[J].中国激光，2024,51(17):286-297.DONG H X,AN Y,XIE L R,et al.Semantic segmentation of large-scale laser point cloud in mines based on local feature enhancement[J].Chinese Journal of Lasers,2024,51(17):286-297.

李荟，韩晓飞，朱万成，等.基于多源信息融合的矿山边坡滑坡灾害研究现状与展望[J].工矿自动化，2024,50(6):6-15.LI H,HAN X F,ZHU W C,et al.Current status and prospects of research on landslide disasters in mine slopes based on multi-source information fusion[J].Industry and Mine Automation,2024,50(6):6-15.

ZHAO H S,SHI J P,QI X J,et al.Pyramid scene parsing network[C]//Proceedings of the Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA.Piscataway,NJ,USA:Institute of Electrical and Electronics Engineers,2017:2881-2890.

HE K M,ZHANG X Y,REN S Q,et al.Deep residual learning for image recognition[C]//Proceedings of the Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,June 27-30,2016,Las Vegas,NV,USA.Piscataway,NJ,USA:Institute of Electrical and Electronics Engineers,2016:770-778.

HOWARD A G,ZHU M L,CHEN B,et al.MobileNets:efficient convolutional neural networks for mobile vision applications[C]//Proceedings of the Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA.Piscataway,NJ,USA:Institute of Electrical and Electronics Engineers,2017:21-26.

LEI C,BU S Q,ZHONG J,et al.Distribution network reconfiguration:a disjunctive convex hull approach[J].IEEE Transactions on Power Systems,2023,38(6):5926-5929.

RONNEBERGER O,FISCHER P,BROX T.U-Net:convolutional networks for biomedical image segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI).Berlin:Springer,2015:234-241.

XIE E Z,WANG W H,YU Z D,et al.SegFormer:simple and efficient design for semantic segmentation with transformers[J].Advances in Neural Information Processing Systems,2021,34:12077-12090.

CHEN L C,ZHU Y,PAPANDREOU G,et al.Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Proceedings of the European Conference On Computer Vision (ECCV),September 8-14,Munich,Germany.Berlin:Springer,2018:801-818.

CHENG B,MISRA I,SCHWING A G,et al.Masked-attention mask transformer for universal image segmentation[C]//Proceedings of the Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,June 18-24,2022,New Orleans,LA,USA.Piscataway,NJ,USA:Institute of Electrical and Electronics Engineers,2022:1290-1299.

LIU Z,MAO H,WU C Y,et al.A convnet for the 2020s[C]//Proceedings of the Institute of Electrical and Electronics Engineers Conference on Computer Vision and Pattern Recognition,June 18-24,2022,New Orleans,LA,USA.Piscataway,NJ,USA:Institute of Electrical and Electronics Engineers,2022:11976-11986.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于LSTM的堆积铝粉自燃早期痕迹气体预测模型

改进YOLOv8算法在长输管道漏磁缺陷检测中的应用研究

时空融合深度学习模型的有毒重气扩散预测

轻量化古建筑青砖损伤检测方法研究

机器视觉下吊装作业吊物与吊钩实时监测方法