Fracture identification and 3D reconstruction of coal-rock combinations based on VRA-UNet network
-
摘要:
在煤岩组合体裂隙三维重构中，针对传统阈值分割方法无法准确确定煤岩之间的阈值大小从而导致裂隙分割效果不佳的问题，基于深度学习理论提出了一种新型VRA-UNet煤岩组合体裂隙精确识别模型，为煤岩组合体裂隙精确识别提供了一种优化解决方案。为了提升模型的泛化能力和防止初始化模型参数过于随机，使用VGG16模块作为骨干特征提取网络。针对煤岩组合体裂隙拓扑结构复杂，非均匀性强等问题，在上采样部分引入使用残差连接且具有空间维度和通道维度的注意力模块（ResCBAM）增强模型特征提取能力，缓解模型梯度消失的问题。在下采样的末端加入了利用不同尺度卷积核的非对称空洞金字塔模块（AC-ASPP），通过多尺度的特征提取，提高模型对不同大小裂隙的识别能力。同时，利用煤岩组合体CT扫描图像数据集验证了模型的有效性。研究结果表明：VRA-UNet模型在裂隙提取和识别方面性能良好，平均交并比、像素平均值及识别精度分别为85.22%、90.80%和91.95%；与主流的分割网络UNet、PSPNet、DeeplabV3+、FCN和SegNet相比，VRA-UNet模型的平均交并比分别提高了6.05%、16.7%、10.77%、6.87%和6.4%，像素平均值分别提高了7.13%、13.29%、12.84%、7.4%和7.53%，识别精度分别提高了3.82%、14.45%、7.4%、5.58%和4.31%；VRA-UNet识别出的裂隙结构分形维数与原始CT扫描裂隙结构分形维数保持了良好的一致性，真实还原了煤岩组合体内部裂隙结构的分布特征。
Abstract:In the 3D reconstruction of coal-rock combinations fractures, in response to the problem that traditional threshold segmentation methods cannot accurately determine the threshold size between coal and rock, resulting in poor fracture segmentation performance, a new VRA-UNet coal-rock combinations fracture identification model based on deep learning theory is proposed, providing an optimized solution for accurate identification of coal-rock combinations fractures. Firstly, the VGG16 module is used as the backbone feature extraction network to enhance the model’s generalization ability and prevent the initialization of model parameters from being too random. Secondly, to address the complex fracture topology and strong non-uniformity of coal-rock combinations, an attention module (ResCBAM) with spatial and channel dimensions is introduced into the up-sampling part to enhance the model's feature extraction ability and alleviate the problem of gradient disappearance. Finally, an asymmetric atrous pyramid module (AC-ASPP) utilizing convolution kernels of different scales is added at the end of the downsampling, which reduced the computational complexity and improved the computational efficiency of the model while keeping the receptive field unchanged. The effectiveness of the model is verified using a dataset of CT scan images of coal-rock combinations. The research results indicate that the VRA-UNet model performs well in crack extraction and recognition, with an average intersection to union ratio, pixel average value, and recognition accuracy of 85.22%, 90.80%, and 91.95%, respectively; Compared with mainstream segmentation networks UNet, PSPNet, DeeplabV3+, FCN, and SegNet the average intersection to union ratio of the VRA-UNet model has increased by 6.05%, 16.7%, 10.77%, 6.87%, and 6.4% respectively. The average pixel value has increased by 7.13%, 13.29%, 12.84%, 7.4%, and 7.53% and the recognition accuracy has risen by 3.82%, 14.45%, 7.4%, 5.58%, and 4.31% respectively; The fractal dimension of the fracture structure identified by VRA-UNet maintains good consistency with the fractal dimension of the original CT scan fracture structure, accurately reproducing the distribution characteristics of the internal fracture structure of the coal-rock combinations.
-
-
![]()
图 9 不同注意力机制网络mIOU指标变化
Figure 9. Changes in mIOU indicator of different attention mechanism networks
![]()
图 11 不同金字塔结构网络mIOU指标变化
Figure 11. Changes in mIOU indicators for networks with different pyramid structures
![]()
图 16 不同分割方法分形维数对比图
Figure 16. Comparison of fractal dimensions using different segmentation methods
表 1 不同主干网络对模型的影响
Table 1 Effect of different backbone on the model
模型
Model平均交并比
mIOU像素平均值
mPA精确率
PrecisionUNet 79.17 83.67 88.13 UNet+Mobilenet 79.46 84.04 88.79 UNet+Resnet50 79.23 83.81 88.67 UNet+VGG16 80.10 84.27 89.07 
下载: 导出CSV
表 2 不同注意力机制对模型的影响
Table 2 Effect of different attention mechanisms on model
模型
Model平均交并比
mIOU像素平均值
mPA精确率
PrecisionUNet 79.17 83.67 88.13 UNet+SE 79.50 83.94 88.73 UNet+ECA 79.56 84.20 89.06 UNet+ResCBAM 81.27 86.22 90.32
下载: 导出CSV
表 3 不同金字塔模块对模型的影响
Table 3 Effect of different pyramid modules on model
模型
Model平均交并比
mIOU像素平均值
mPA精确率
PrecisionUNet 79.17 83.67 88.13 UNet+SPP 81.94 88.60 89.61 UNet+ASPP 82.09 89.13 89.29 UNet+AC-ASPP 83.50 89.95 90.46
下载: 导出CSV
表 4 消融试验结果
Table 4 Results of ablation experiments
主干网络
VGG16注意力机制
ResCBAM非对称空洞
金字塔
AC-ASPP平均交并比
mIOU像素平均值
mPA精确率
Precision× × × 79.17 83.67 88.13 √ × × 80.10 84.27 89.07 √ √ × 83.53 89.54 90.92 √ √ √ 85.22 90.80 91.95 注：“√”代表在基础UNet网络模型中引入这个模块；“×”则反之。
下载: 导出CSV
表 5 不同模型的对比试验
Table 5 Experimental comparison of different models
模型
Model平均交并比
mIOU像素平均值
mPA精确率
PrecisionUNet 79.17 83.67 88.13 PSPNet 68.52 77.51 77.50 DeeplabV3+ 74.45 77.96 84.55 FCN 78.35 83.04 86.37 SegNet 78.82 83.27 87.64 VRA-UNet 85.22 90.80 91.95
下载: 导出CSV
表 6 不同模型单张图像处理时间
Table 6 Single image processing time for different models
模型Model 时间 Time/ms UNet 176 PSPNet 132 DeeplabV3+ 137 FCN 165 SegNet 171 VRA-UNet 202
下载: 导出CSV
表 7 不同分割方法分形维数
Table 7 Fractal dimensions of different segmentation methods
方法 分形维数D 原始CT 1.561 VRA-UNet 1.519 阈值分割 1.441
下载: 导出CSV
-
[1] 程远平,王成浩. 构造煤变形能及在煤与瓦斯突出中的作用[J]. 煤炭学报,2024,49(2):645−663. CHENG Yuanping,WANG Chenghao. Deformation energy of tectonic coal and its role in coal and gas outbursts[J]. Journal of China Coal Society,2024,49(2):645−663.
[2] 陈光波,李谭,杨磊,等. 不同煤岩比例及组合方式的组合体力学特性及破坏机制[J]. 采矿与岩层控制工程学报,2021,3(2):84−94. CHEN Guangbo,LI Tan,YANG Lei,et al. Mechanical properties and failure mechanism of combined bodies with different coal-rock ratios and combinations[J]. Journal of Mining and Strata Control Engineering,2021,3(2):84−94.
[3] 袁亮,王恩元,马衍坤,等. 我国煤岩动力灾害研究进展及面临的科技难题[J]. 煤炭学报,2023,48(5):1825−1845. YUAN Liang,WANG Enyuan,MA Yankun,et al. Research progress of coal and rock dynamic disasters and scientific and technological problems in China[J]. Journal of China Coal Society,2023,48(5):1825−1845.
[4] 赵伟,董虎子,闫志达,等. 深部煤层瓦斯含量分阶赋存规律及其与突出危险的关联[J/OL]. 煤炭学报,1−14[2024−09−25]. https://doi.org/10.13225/j.cnki.jccs.2024.0161. ZHAO Wei,DONG Huzi,YAN Zhida,et al. The hierarchical occurrence law of gas content in deep coal seam and its association with outstanding hazards[J/OL]. Journal of China Coal Society,1−14[2024−09−25]. https://doi.org/10.13225/j.cnki.jccs.2024.0161.
[5] 赵伟,王凯,周爱桃,等. 扩散主控型煤层定义、特征及瓦斯促抽路径分析[J/OL]. 煤炭学报,1−14[2024−09−25]. https://doi.org/10.13225/j.cnki.jccs.WK24.0320. ZHAO Wei,WANG Kai,ZHOU Aitao,et al. Definition,characteristics and analysis of gas pumping path of diffusion master control briquette[J/OL]. Journal of China Coal Society,1−14[2024−09−25]. https://doi.org/10.13225/j.cnki.jccs.WK24.0320.
[6] 李伟,杨康,邓东,等. 考虑孔弹性效应的煤岩微纳米孔隙瓦斯表观渗透率模型及其在瓦斯抽采中的应用[J]. 岩石力学与工程学报,2024,43(3):587−599. LI Wei,YANG Kang,DENG Dong,et al. A gas apparent permeability model in coal micro/nano-pores considering the poroelastic effect and its application in gas extraction[J]. Chinese Journal of Rock Mechanics and Engineering,2024,43(3):587−599.
[7] 陈光波,李谭,张国华,等. 煤岩组合体破坏前能量积聚规律试验研究[J]. 煤炭学报,2021,46(S1):174−186. CHEN Guangbo,LI Tan,ZHANG Guohua,et al. Experimental study on energy accumulation law before coal rock combination failure[J]. Journal of China Coal Society,2021,46(S1):174−186.
[8] 王登科,张平,魏建平,等. CT可视化的受载煤体三维裂隙结构动态演化试验研究[J]. 煤炭学报,2019,44(S2):574−584. WANG Dengke,ZHANG Ping,WEI Jianping,et al. Experimental study on dynamic evolution of three-dimensional fracture structure of loaded coal based on CT visualization[J]. Journal of China Coal Society,2019,44(S2):574−584.
[9] 王登科,房禹,魏建平,等. 基于深度学习的煤岩Micro-CT裂隙智能提取与应用[J]. 煤炭学报,2024,49(8):3439−3452. WANG Dengke,FANG Yu,WEI Jianping,et al. Intelligent extraction and application of Micro CT fractures in coal and rock based on deep learning[J]. Journal of China Coal Society,2024,49(8):3439−3452.
[10] 王刚,陈雪畅,韩冬阳,等. 基于改进Otsu的煤体CT图像阈值分割算法的研究[J]. 煤炭科学技术,2021,49(1):264−271. WANG Gang,CHEN Xuechang,HAN Dongyang,et al. Research on threshold segmentation algorithm of coal CT images based on improved Otsu[J]. Coal Science and Technology,2021,49(1):264−271.
[11] 郝晨光,郭晓阳,邓存宝,等. 基于Bi-PTI模型的CT数字煤岩孔裂隙精准识别及阈值反演[J]. 煤炭学报,2023,48(4):1516−1526. HAO Chenguang,GUO Xiaoyang,DENG Cunbao,et al. Precise identification and threshold inversion of pores and fissures in CT digital coal rock based on Bi-PTI model[J]. Journal of China Coal Society,2023,48(4):1516−1526.
[12] 李国燕,梁家栋,刘毅,等. MFC-DeepLabV3+:一种多特征级联融合裂缝缺陷检测网络模型[J]. 铁道科学与工程学报,2023,20(4):1370−1381. LI Guoyan,LIANG Jiadong,LIU Yi,et al. MFC-DeepLabV3+:A multi feature cascade fusion crack defect detection network model[J]. Journal of Railway Science and Engineering,2023,20(4):1370−1381.
[13] 张云,童亮,来兴平,等. 基于机器视觉的煤尘环境下掘进空间煤岩界面感知与精准识别[J]. 煤炭学报,2024,49(7):3276−3290. ZHANG Yun,TONG Liang,LAI Xingping,et al. Perception and precise recognition of coal rock interface in excavation space under coal dust environment based on machine vision[J]. Journal of China Coal Society,2024,49(7):3276−3290.
[14] 杜锋,王凯,董香栾,等. 基于CT三维重构的煤岩组合体损伤破坏数值模拟研究[J]. 煤炭学报,2021,46(S1):253−262. DU Feng,WANG Kai,DONG Xiangluan,et al. Numerical simulation study on damage and failure of coal rock composite based on CT three-dimensional reconstruction[J]. Journal of China Coal Society,2021,46(S1):253−262.
[15] 闫志蕊,王宏伟,耿毅德. 基于改进DeeplabV3+和迁移学习的煤岩界面图像识别方法[J]. 煤炭科学技术,2023,51(S1):429−439. YAN Zhirui,WANG Hongwei,GENG Yide. Coal-rock interface image recognition method based on improved DeeplabV3+ and transfer learning[J]. Coal Science and Technology,2023,51(S1):429−439.
[16] JIN H X,CAO L,KAN X,et al. Coal petrography extraction approach based on multiscale mixed-attention-based residual U-net[J]. Measurement Science and Technology,2022,33(7):075402. doi: 10.1088/1361-6501/ac5439
[17] BOUGOURZI F,DISTANTE C,DORNAIKA F,et al. PDAtt-unet:Pyramid dual-decoder attention unet for covid-19 infection segmentation from CT-scans[J]. Medical Image Analysis,2023,86:102797. doi: 10.1016/j.media.2023.102797
[18] LI B,WU F,LIU S K,et al. CA‐Unet++:An improved structure for medical CT scanning based on the Unet++ Architecture[J]. International Journal of Intelligent Systems,2022,37(11):8814−8832. doi: 10.1002/int.22969
[19] 李元海,徐晓华,朱鸿鹄,等. 基于计算机视觉的岩石裂隙识别表征与软件研制[J]. 岩土工程学报,2024,46(3):459−469. doi: 10.11779/CJGE20221239 LI Yuanhai,XU Xiaohua,ZHU Honghu,et al. Identification and characterization of rock fractures based on computer vision and software development[J]. Chinese Journal of Geotechnical Engineering,2024,46(3):459−469. doi: 10.11779/CJGE20221239
[20] ZHAO W,ZHANG H D,YAN Y J,et al. A semantic segmentation algorithm using FCN with combination of BSLIC[J]. Applied Sciences,2018,8(4):500. doi: 10.3390/app8040500
[21] 王安民,高于超,邹俊超,等. 基于深度学习的煤系页岩孔隙结构定量表征[J]. 煤炭科学技术,2023,51(S2):183−190 WANG Anmin,GAO Yuchao,ZOU Junchao,et al. Quantitative characterization of pore structure in coal bearing shale based on deep learning[J]. Coal Science and Technology,2023,51(S2):183−190
[22] 薛东杰,唐麒淳,王傲,等. 煤岩微观相态FCN智能识别与分形重构[J]. 岩石力学与工程学报,2020,39(6):1203−1221. XUE Dongjie,TANG Qichun,WANG Ao,et al. FCN-based intelligent identification and fractal reconstruction of pore-fracture network in coal by micro CT scanning[J]. Chinese Journal of Rock Mechanics and Engineering,2020,39(6):1203−1221.
[23] CHEN J H,LI H,PHILIP CHEN C L. Boosting sharpness-aware training with dynamic neighborhood[J]. Pattern Recognition,2024,153:110496. doi: 10.1016/j.patcog.2024.110496
[24] DENG H Y,ZHU R J,QIU X R,et al. Tensor decomposition based attention module for spiking neural networks[J]. Knowledge-Based Systems,2024,295:111780. doi: 10.1016/j.knosys.2024.111780
[25] 龙丽红,朱宇霆,闫敬文,等. 新型语义分割D-UNet的建筑物提取[J]. 遥感学报,2023,27(11):2593−2602. doi: 10.11834/jrs.20211029 LONG Lihong,ZHU Yuting,YAN Jingwen,et al. New building extraction method based on semantic segmentation[J]. National Remote Sensing Bulletin,2023,27(11):2593−2602. doi: 10.11834/jrs.20211029
[26] 赵志宏,何朋,郝子晔. 一种道路裂缝检测的变尺度VS-UNet模型[J]. 湖南大学学报(自然科学版),2024,51(6):63−72. ZHAO Zhihong,HE Peng,HAO Ziye. A variable-scale VS-UNet model for road crack detection[J]. Journal of Hunan University (Natural Sciences),2024,51(6):63−72.
[27] 张伟光,钟靖涛,呼延菊,等. 基于VGG16-UNet语义分割模型的路面龟裂形态提取与量化[J]. 交通运输工程学报,2023,23(2):166−182. ZHANG Weiguang,ZHONG Jingtao,HU Yanju,et al. Extraction and quantification of pavement alligator crack morphology based on VGG16-UNet semantic segmentation model[J]. Journal of Traffic and Transportation Engineering,2023,23(2):166−182.
[28] CHILLAKURU P,ANANTHAJOTHI K,DIVYA D. Three stage classification framework with ranking scheme for distracted driver detection using heuristic-assisted strategy[J]. Knowledge-Based Systems,2024,293:111589. doi: 10.1016/j.knosys.2024.111589
[29] YANG J,CHEN Y,YU J L. Convolutional neural network based on the fusion of image classification and segmentation module for weed detection in alfalfa[J]. Pest Management Science,2024,80(6):2751−2760. doi: 10.1002/ps.7979
[30] BAHDANAU D,CHO K,BENGIO Y. Neural machine translation by jointly learning to align and translate[J]. arxiv preprint arxiv, 2014: 1409.0473[2024−09−25]. https://doi.org/10.48550/arXiv.1409.0473
[31] SHI Z Q,JIN N,CHEN D B,et al. A comparison study of semantic segmentation networks for crack detection in construction materials[J]. Construction and Building Materials,2024,414:134950. doi: 10.1016/j.conbuildmat.2024.134950
[32] ZHANG G,YAN H F,ZHANG D Y,et al. Enhancing model performance in detecting lodging areas in wheat fields using UAV RGB Imagery:Considering spatial and temporal variations[J]. Computers and Electronics in Agriculture,2023,214:108297. doi: 10.1016/j.compag.2023.108297
[33] HU J,SHEN L,ALBANIE S,et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(8):2011−2023. doi: 10.1109/TPAMI.2019.2913372
[34] YANG Y K,ZHOU W,JISKANI I M,et al. Extracting unstructured roads for smart Open-Pit mines based on computer vision:Implications for intelligent mining[J]. Expert Systems with Applications,2024,249:123628. doi: 10.1016/j.eswa.2024.123628
[35] SHU H,WANG K W,GUO L F,et al. Automated collection of facial temperatures in dairy cows via improved UNet[J]. Computers and Electronics in Agriculture,2024,220:108614. doi: 10.1016/j.compag.2024.108614
[36] ALI J,ZHENG C H,LYU T,et al. Enhanced bioelectroremediation of heavy metal contaminated groundwater through advancing a self-standing cathode[J]. Water Research,2024,256:121625. doi: 10.1016/j.watres.2024.121625
[37] MALEKMOHAMMADI A,SORYANI M,KOZEGAR E. Mass segmentation in automated breast ultrasound using an enhanced attentive UNet[J]. Expert Systems with Applications,2024,245:123095. doi: 10.1016/j.eswa.2023.123095
[38] KANG S,LI D F,LI B L,et al. Maturity identification and category determination method of broccoli based on semantic segmentation models[J]. Computers and Electronics in Agriculture,2024,217:108633. doi: 10.1016/j.compag.2024.108633
[39] FATMA Ş F,AYDOĞAN S,AKAY D. Investigating the carbon border adjustment mechanism transition process with linguistic summarization method:A situational analysis of exporting countries[J]. Advanced Engineering Informatics,2024,61:102528. doi: 10.1016/j.aei.2024.102528
[40] FERNÁNDEZ J G,MEHRKANOON S. Broad-UNet:Multi-scale feature learning for nowcasting tasks[J]. Neural Networks,2021,144:419−427. doi: 10.1016/j.neunet.2021.08.036
[41] FANG C Y,CHEN H B,LI L,et al. A novel Adaptive Zone-fusion network for precise waxberry semantic segmentation to improve automated-harvesting in a complex orchard environment[J]. Computers and Electronics in Agriculture,2024,221:108937. doi: 10.1016/j.compag.2024.108937
[42] BUI N A,OH Y,LEE I. Oil spill detection and classification through deep learning and tailored data augmentation[J]. International Journal of Applied Earth Observation and Geoinformation,2024,129:103845. doi: 10.1016/j.jag.2024.103845
[43] WANG D K,LI L,ZHANG H T,et al. Intelligent identification of coal fractures using an improvedU-shaped networ[J]. Advances in Geo-Energy Research,2025,15(2):129−142. doi: 10.46690/ager.2025.02.05
计量
- 文章访问数: 133
- HTML全文浏览量: 23
- PDF下载量: 46
