In view of the problems that existing monitoring methods for rock mass joints mainly rely on manual identification, with low detection efficiency and strong subjectivity, a detection and segmentation method for rock mass joints was proposed in this paper. The model improved the YOLOv8 algorithm by introducing the multi-scale feature module (MSBlock) and channel prior convolutional attention (CPCA) mechanism to enhance the image recognition accuracy of the existing YOLOv8 network model. By introducing the MSBlock algorithm, features at different scales and levels were fused, which could both capture the fine local details in the image and recognize the image features macroscopically. By introducing the CPCA mechanism, the attention distribution was adaptively adjusted according to the current image features, enabling the network to more flexibly deal with different targets. A total of 4 057 images collected from the AM Highway Project in Equatorial Guinea were trained and recognized. The results indicate that the performance evaluation indicators of the optimized algorithm, including Box (P), Box (R), Mask (P), Mask (R), rIOU, and CDICE, reach 89.4%, 87.1%, 61.6%, 58.9%, 92.8%, and 86.8%, respectively. Compared with the original model, these indicators increase by 4.6 percentage points, 9.5 percentage points, 8.5 percentage points, 12.4 percentage points, 17.5 percentage points, and 1.1 percentage points, respectively. This multi-scale feature fusion algorithm is more accurate and efficient when handling complex images and has better adaptability and robustness.
CPCA[29](Class-Prototype based Classification Attention)方法是一种新型的轻量级且高效的通道空间注意力机制,它通过深度可分离卷积和注意力机制来捕捉和融合不同尺度的局部特征,以此提高图像分割任务的性能[30]。CPCA在CNN的基础上,结合注意力权重和卷积核,有效地提取和整合图像特征。在处理复杂岩石节理图像时,CPCA能够更好地处理模糊和遮挡的节理特征,通过动态调整注意力权重来增强重要的节理特征的权重,同时忽略不重要的部分,从而提高YOLOv8模型在围岩图像节理提取方面的效率和准确性。CPCA的结构设计,增强了通道和空间信息的融合能力,以及不同层次特征图之间的联系,提升了目标检测的效果:使YOLOv8能够更加精确地识别和定位不同尺度和类别的目标,同时保持计算效率和实时处理的能力。CPCA总体结构包括通道注意力(AC)和空间注意力(AS)两部分,具体流程如下[31]:
由于AdamW(Adam with Weight Decay)在深度学习任务中表现优异,它可以自适应地调整每个参数的学习率,改进了权重衰减的处理方式,可以进行更精准的模型调优,故优化器选择AdamW。训练数据按照8∶2的比例随机划分原始围岩节理数据集。其中,80%数据集用于网络模型的训练,20%数据作为模型的验证集,用于评估模型的检测性能。
MaoJ Y, SongZ P, FanS Y, et al. Exploration and verification of tunnel stability evolution law under jointed rock mass with various attitudes[J]. International Journal of Civil Engineering, 2024, 22(5): 689-703.
[2]
SongZ P, LiX, HuoR K, et al. Intelligent early-warming platform for open-pit mining: Current status and prospects[J].Rock Mechanics Bulletin,2023:100098
SunZe, SongZhanping, YueBo, et al. Inversion model for surrounding rock parameters based on back propagation neural networks improved by stochastic gradient descent algorithm[J]. Tunnel Construction, 2023, 43(12): 2066-2076.
HuangPeng, FanWenzhe, HuLingling, et al. Application of bridge crack recognition algorithm based on image connected domain[J]. Journal of China & Foreign Highway, 2025, 45(2): 126-132.
ChenWei, JiangZhicheng, TianZijian, et al. Unsafe action detection algorithm of underground personnel in coal mine based on YOLOv8[J]. Coal Science and Technology, 2024, 52(S2): 267-283.
[12]
GirshickR, DonahueJ, DarrellT, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus: IEEE, 2014: 580-587.
GaoHongfei, YangRensong, ZhangYunbo, et al. The intelligent identification method and application of tunnel face joints based on the enhanced Mask RCNN algorithm[J]. Industrial Construction, 2023, 53(S2): 583-587.
PengLei, ZhouChun, HuFeng, et al. Intelligent detection and segmentation of tunnel face joints based on deep learning[J]. Yangtze River, 2023, 54(10): 243-250.
ZhangNong, YuanYuxin, HanChangliang, et al. Research on crack detection and localization algorithm for advancing face in coalmine roadways based on Mask R-CNN[J]. Journal of Mining & Safety Engineering, 2023, 40(5): 925-932.
JingYing, RuanShunling, LuCaiwu, et al. An intelligent detection method for open-pit slope fracture based on the improved Mask R-CNN[J]. Journal of Chongqing University, 2023, 46(2): 67-80.
[21]
LiuW, AnguelovD, ErhanD, et al.SSD:Single shot multibox detector[C]//Proceedings of the European Conference on Computer Vision.Chan: Springer,2016:21-37.
ShaoYanhua, ZhangDuo, ChuHongyu, et al. A review of YOLO object detection based on deep learning[J]. Journal of Electronics & Information Technology, 2022, 44(10): 3697-3708.
ZengZhichao, XuYue, WangJingyu, et al. A water surface target detection algorithm based on SOE-YOLO lightweight network[J]. Journal of Graphics, 2024, 45(4): 736-744.
MaChaowei, ZhangHao, MaXinming, et al. Method for the lightweight detection of wheat disease using improved YOLOv8[J]. Transactions of the Chinese Society of Agricultural Engineering, 2024, 40(5): 187-195.
GaoMingxing, GuanXuefeng, FanJingli, et al. Intelligent recognition of road surface disease based on improved YOLOv5-DeepSORT algorithm[J]. Forest Engineering, 2023, 39(5): 161-174.
[34]
WangG, ChenY F, AnP, et al. UAV-YOLOv8: A small-object-detection model based on improved YOLOv8 for UAV aerial photography scenarios[J]. Sensors, 2023, 23(16): 7190.
ZhaoXiaohu, XieLixun, MuDengcong, et al. Metal surface defect detection method based on TCM-YOLO network[J]. Computer Engineering, 2025, 51(6): 338-348.
ZhouFengjun, KangHuaiqiang, GaoShen, et al. Detection and recognition of concrete cracks in underground engineering based on improved YOLOv8 model[J]. Computer Engineering and Science, 2025, 47(6): 1079-1089.
[42]
BodlaN, SinghB, ChellappaR, et al. Soft-NMS: improving object detection with one line of code[C]//Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017:5562-5570.
[43]
LouH, DuanX, GuoJ, et al.DC-YOLOv8: Small-size object detection algorithm based on camera sensor[J].Electronics,2023,12(10):2323.
[44]
TangY Z, QianY. High-speed railway track components inspection framework based on YOLOv8 with high-performance model deployment[J]. High-Speed Railway, 2024, 2(1): 42-50.
LuHongtao, ZhangQinchuan. Applications of deep convolutional neural network in computer vision[J]. Journal of Data Acquisition and Processing, 2016, 31(1): 1-17.
[47]
HuangH J, ChenZ G, ZouY, et al. Channel prior convolutional attention for medical image segmentation[J]. Computers in Biology and Medicine, 2024, 178: 108784.
ZhuChengjie, CaiZizheng, ZhuHongbo. Road surface pothole detection based on improved YOLOv8[J]. Journal of Chongqing Technology and Business University (Natural Science Edition), 2026, 43(2): 93-100.
[52]
ChenY, YuanX, WangJ, et al. YOLO-MS: Rethinking multi-scale representation learning for real-time object detection[J]. IEEE Trans Pattern Anal Mach Intell, 2025, 47(6): 4240-4252.
WenYan, GuoXiaoqian, TangYouzhi, et al. Research on crack identification technology based on Sobel+Matched filter joint algorithm[J]. Journal of China & Foreign Highway, 2025, 45(4): 33-41.
GaoLiang, RaoFaqiang, YangZhongmin, et al. Intelligent recognition technology of highway cracks based on semantic segmentation[J]. Journal of China & Foreign Highway, 2024, 44(5): 241-247.
[57]
RussellB C, TorralbaA, MurphyK P, et al. LabelMe: A database and web-based tool for image annotation[J]. International Journal of Computer Vision, 2008, 77(1): 157-173.