Mine hoisting steel wire rope surface damage image recognition based on improved YOLOv8n
-
摘要:
针对矿用提升钢丝绳表面油污覆盖引发背景干扰、绳股间隙较大导致特征混淆及小目标损伤识别难度大等问题，提出了一种基于改进YOLOv8n的矿用提升钢丝绳表面损伤图像识别方法。在YOLOv8n主干网络中引入多尺度注意力模块（MSAM），通过增强损伤特征与油污背景的空间特征区分能力，提升模型抗干扰能力；将YOLOv8n原有的3个检测头替换为4个轻量化小目标检测头，强化对小目标损伤的识别能力；采用深度可分离卷积（DSConv）替代标准卷积，减少了计算量，提高了识别速度。实验结果表明：改进YOLOv8n模型的平均精度均值（mAP）、识别精度和推理速度分别达92.6%，89.7%和43.5帧/s，相比YOLOv8n模型分别提高了3.1%，4.9%，34.7%；与Faster−RCNN，YOLOv5s，YOLOv8n，YOLOv10m，TWRD−Net，YOLOv5−TPH等主流模型相比，改进YOLOv8n模型对小目标损伤识别精度最高，同时保证了较高的实时性；在煤矿现场油污覆盖、绳股间隙较大的复杂场景中，改进YOLOv8n模型未出现漏检情况，且误检情况较少，平均识别准确率达90%。
Abstract:To address issues such as background interference caused by oil stains covering the surface of mine hoisting steel wire ropes, large gaps between rope strands leading to feature confusion, and the difficulty in identifying small target damages, a surface damage image recognition method for mine hoisting steel wire ropes based on an improved YOLOv8n model was proposed. The Multi-Scale Attention Module (MSAM) was introduced into the YOLOv8n backbone network to enhance the model’s ability to distinguish between damage features and oil stain backgrounds, improving its anti-interference capability. The original three detection heads of YOLOv8n were replaced with four lightweight small-target detection heads to strengthen the recognition ability for small target damages. Depthwise Separable Convolutions (DSConv) were used instead of standard convolutions to reduce computational load and improve recognition speed. Experimental results showed that the improved YOLOv8n model achieved an Mean Average Precision (mAP) of 92.6%, a recognition accuracy of 89.7%, and an inference speed of 43.5 frames per second, representing improvements of 3.1%, 4.9%, and 34.7%, respectively, compared to the original YOLOv8n model. Compared with mainstream models such as Faster-RCNN, YOLOv5s, YOLOv8n, YOLOv10m, TWRD-Net, and YOLOv5-TPH, the improved YOLOv8n model exhibited the best accuracy for small target damage recognition while maintaining high real-time performance. In complex field scenarios with oil stain coverage and large gaps between rope strands in coal mines, the improved YOLOv8n model did not miss any damage detections and had fewer false detections, achieving an average recognition accuracy of 90%.
-
-
![]()
图 9 不同模型对实验室钢丝绳识别结果对比
Figure 9. Comparison of recognition results for laboratory steel wire ropes using different models
![]()
图 10 不同模型对矿井实际钢丝绳识别结果对比
Figure 10. Comparison of recognition results for actual mine steel wire ropes using different models
表 1 消融实验结果
Table 1 Results of ablation experiment
DSConv MSAM 检测头 Precision/% Recall/% mAP/% FPS/（帧·s−1） × × × 85.5 87.7 89.8 32.3 √ × × 86.2 88.8 90.7 45.6 × √ × 89.4 88.5 92.0 31.1 × × √ 87.9 89.5 91.2 30.5 √ √ √ 89.7 90.1 92.6 43.5 
下载: 导出CSV
表 2 各模型性能对比结果
Table 2 Performance comparison results of various models
模型 mAP/% FPS/（帧·s−1） Faster−RCNN 85.5 14.2 YOLOv5−TPH 89.9 38.5 TWRD−Net 88.5 45.1 YOLOv10m 92.0 40.0 YOLOv5s 87.9 40.0 YOLOv8n 89.8 45.0 改进YOLOv8n 92.6 43.5
下载: 导出CSV
表 3 不同模型对小目标损伤识别对比结果
Table 3 Comparison results of small target damage recognition of different models
模型 mAP/% FPS/（帧·s−1） Faster−RCNN 82.1 12.8 YOLOv5−TPH 88.4 35.9 TWRD−Net 86.7 40.4 YOLOv10m 87.2 35.2 YOLOv5s 81.2 38.6 YOLOv8n 82.5 42.9 改进 YOLOv8n 92.1 41.5
下载: 导出CSV
表 4 矿井实际钢丝绳识别结果
Table 4 Recognition results of actual mine steel wire ropes
样本 数量/张 识别正确数量/张 识别错误数量/张 准确率/% 钢丝绳1 100 92 8 92 钢丝绳2 100 90 10 90 钢丝绳3 100 88 13 88
下载: 导出CSV
-
[1] 李腾宇,寇子明,吴娟,等. 超千米深井提升机可视化监测系统应用[J]. 煤炭学报,2020,45(增刊2):1069-1078. LI Tengyu,KOU Ziming,WU Juan,et al. Monitoring system of the hoist in the over kilometer deep shaft[J]. Journal of China Coal Society,2020,45(S2):1069-1078.
[2] 路正雄,郭卫,张传伟,等. 平行磁化NdFeB钢丝绳无损检测仪开发[J]. 西安科技大学学报,2021,41(1):139-144. LU Zhengxiong,GUO Wei,ZHANG Chuanwei,et al. Development of a new wire rope non-destructive tester using parallely magnetized NdFeB[J]. Journal of Xi'an University of Science and Technology,2021,41(1):139-144.
[3] 田劼,王洋洋,郭红飞,等. 基于漏磁检测的钢丝绳探伤原理与方法研究[J]. 煤炭工程,2021,53(9):95-100. TIAN Jie,WANG Yangyang,GUO Hongfei,et al. Principle and method of mine steel wire rope flaw detection based on magnetic flux detection[J]. Coal Engineering,2021,53(9):95-100.
[4] 王文庆,刘文辉,李生辉,等. 基于永磁环励磁结构的钢丝绳无损检测设计[J]. 西安邮电大学学报,2023,28(5):92-101. WANG Wenqing,LIU Wenhui,LI Shenghui,et al. A design of non-destructive testing for steel wire ropes based on permanent magnet ring excitation structure[J]. Journal of Xi'an University of Posts and Telecommunications,2023,28(5):92-101.
[5] 王红尧,李小伟,韩亦淼,等. 矿用钢丝绳损伤检测系统设计[J]. 工矿自动化,2020,46(6):92-97. WANG Hongyao,LI Xiaowei,HAN Yimiao,et al. Design of damage detection system for mine-used wire rope[J]. Industry and Mine Automation,2020,46(6):92-97.
[6] MAZUREK P,ROSKOSZ M,KWAŚNIEWSKI J. Novel diagnostic of steel wire rope with passive magnetic methods[J]. IEEE Magnetics Letters,2021,13. DOI: 10.1109/LMAG.2021.3128828.
[7] 张鑫鹏,黄丽霞,沈佳卉,等. 钢丝绳声发射信号传播特性分析[J]. 失效分析与预防,2022,17(1):24-31,36. DOI: 10.3969/j.issn.1673-6214.2022.01.004 ZHANG Xinpeng,HUANG Lixia,SHEN Jiahui,et al. Analysis on transmission characteristics of acoustic emission signal of wire rope[J]. Failure Analysis and Prevention,2022,17(1):24-31,36. DOI: 10.3969/j.issn.1673-6214.2022.01.004
[8] 付红伟,牛彦鹏,李岩森. 电梯钢丝绳检测与维护方法探讨[J]. 中国设备工程,2020(1):120-122. DOI: 10.3969/j.issn.1671-0711.2020.01.063 FU Hongwei,NIU Yanpeng,LI Yansen. Discussion on detection and maintenance method of elevator wire rope[J]. China Plant Engineering,2020(1):120-122. DOI: 10.3969/j.issn.1671-0711.2020.01.063
[9] 王国锋,王守军,陶荣颖,等. 矿井提升机钢丝绳外观缺陷视觉识别技术研究[J]. 工矿自动化,2024,50(5):28-35. WANG Guofeng,WANG Shoujun,TAO Rongying,et al. Research on visual recognition technology for appearance defects of steel wire rope in mine hoist[J]. Journal of Mine Automation,2024,50(5):28-35.
[10] 林嘉森. 基于多路视频采集及图像分析的缆索表面损伤检测系统研究与开发[D]. 南京:东南大学,2018. LIN Jiasen. Research and development of cable surface damage detection system based on multi-channel video acquisition and image analysis[D]. Nanjing:Southeast University,2018.
[11] 施联宾. 基于机器视觉的立井提升系统关键部件运行状态监测研究[D]. 徐州:中国矿业大学,2021. SHI Lianbin. Research on running condition monitoring of key components of hoisting system based on machine vision[D]. Xuzhou:China University of Mining and Technology,2021.
[12] HUANG Xinyuan,LIU Zhiliang,ZHANG Xinyu,et al. Surface damage detection for steel wire ropes using deep learning and computer vision techniques[J]. Measurement,2020,161. DOI: 10.1016/j.measurement.2020.107843.
[13] 曲诚,陈景龙,常元洪,等. 面向钢丝绳微弱损伤智能识别的多尺度注意力网络[J]. 西安交通大学学报,2021,55(7):141-150. DOI: 10.7652/xjtuxb202107016 QU Cheng,CHEN Jinglong,CHANG Yuanhong,et al. Multi-scale attention network for intelligent identification of weak damage on wire ropes[J]. Journal of Xi'an Jiaotong University,2021,55(7):141-150. DOI: 10.7652/xjtuxb202107016
[14] 张玉茜,刘文荣,孙勇,等. 基于Faster−RCNN与自注意力机制的矿山图像异常检测算法[J]. 金属矿山,2024(7):196-201. ZHANG Yuqian,LIU Wenrong,SUN Yong,et al. Mining image anomaly detection algorithm based on Faster-RCNN and self-attention mechanism[J]. Metal Mine,2024(7):196-201.
[15] 王红尧,韩爽,李勤怡. 改进YOLOv5的钢丝绳损伤图像识别实验方法研究[J]. 计算机工程与应用,2023,59(17):99-106. DOI: 10.3778/j.issn.1002-8331.2210-0505 WANG Hongyao,HAN Shuang,LI Qinyi. Experimental research on image recognition of wire rope damage based on improved YOLOv5[J]. Computer Engineering and Applications,2023,59(17):99-106. DOI: 10.3778/j.issn.1002-8331.2210-0505
[16] 高嘉,刘涛,王显峰,等. TWRD−Net:一种用于曳引钢丝绳缺陷的实时检测网络算法[J]. 仪器仪表学报,2023,44(6):223-235. GAO Jia,LIU Tao,WANG Xianfeng,et al. TWRD-Net:a real-time detection network algorithm for traction wire rope defects[J]. Chinese Journal of Scientific Instrument,2023,44(6):223-235.
[17] ZHOU Ping,ZHOU Gongbo,WANG Shihao,et al. Visual sensing inspection for the surface damage of steel wire ropes with object detection method[J]. IEEE Sensors Journal,2022,22(23):22985-22993. DOI: 10.1109/JSEN.2022.3214109
[18] 刘晓磊,吴国群,阚哲. 基于深度学习的煤矿钢丝绳缺损检测方法研究[J]. 煤炭工程,2023,55(11):148-153. LIU Xiaolei,WU Guoqun,KAN Zhe. Research on defect detection method of coal mine wire rope based on deep learning[J]. Coal Engineering,2023,55(11):148-153.
[19] 王发明,倪昕东,张旗,等. 基于MobileNetV2−CBAM的机收场景下冬小麦成熟期在线分类识别方法[J]. 农业机械学报,2024,55(增刊1):71-80,100. WANG Faming,NI Xindong,ZHANG Qi,et al. Online classification and identification method of winter wheat maturity under mechanical harvesting scenario based on MobileNetV2-CBAM[J]. Transactions of the Chinese Society for Agricultural Machinery,2024,55(S1):71-80,100.
[20] 黄昆,齐肇建,王娟敏,等. 基于改进YOLOv8的密集行人检测模型[J/OL]. 计算机工程:1-11[2024-11-28]. https://doi.org/10.19678/j.issn.1000-3428.0069026. HUANG Kun,QI Zhaojian,WANG Juanmin,et al. Dense pedestrian detection model based on improved YOLOv8[J/OL]. Computer Engineering:1-11[2024-11-28]. https://doi.org/10.19678/j.issn.1000-3428.0069026.
[21] 刘艳丽,王浩,张帆. 基于轻量卷积和模型优化的电弧故障检测方法[J]. 仪器仪表学报,2024,45(10):38-49. LIU Yanli,WANG Hao,ZHANG Fan. Arc fault detection method based on lightweight convolution and model optimization[J]. Chinese Journal of Scientific Instrument,2024,45(10):38-49.
-
期刊类型引用(12)
1. 任助理，朱红洋，袁瑞甫，王伸. 三维激光扫描技术在地下矿山中的应用综述与展望. 河南理工大学学报(自然科学版). 2025(03): 89-100 . 
百度学术
2. 马庆禄，汪军豪，张杰，邹政. 激光雷达与惯性测量单元同步融合下的园区三维建图. 光学精密工程. 2024(03): 422-434 . 百度学术
3. 周波，唐桂彬. 基于遥感GIS信息融合的测绘机器人滑模跟踪控制系统. 计算机测量与控制. 2024(06): 125-132 . 百度学术
4. 高毅楠，姚顽强，蔺小虎，郑俊良，马柏林，冯玮，高康洲. 煤矿井下多重约束的视觉SLAM关键帧选取方法. 煤炭学报. 2024(S1): 472-482 . 百度学术
5. 倪志平，卢光云. 激光雷达实时采集数据的车辆智能定位方法研究. 激光杂志. 2024(09): 218-222 . 百度学术
6. 覃学宁，钟辉，岳志伟，何瑞冠，冷泉，李飞. 基于轨迹重构与贝叶斯推理的空中机器人灯塔距离测绘技术. 计算机测量与控制. 2024(10): 291-297 . 百度学术
7. 胡青松，李敬雯，张元生，李世银，孙彦景. 面向矿井无人驾驶的IMU与激光雷达融合SLAM技术. 工矿自动化. 2024(10): 21-28 . 本站查看
8. 马亮，高亮，廉博翔，张琦，蔺小虎，姜之跃. 基于已知点约束的高精度煤矿巷道三维点云建模方法. 工矿自动化. 2024(11): 78-83+151 . 本站查看
9. 马宏伟，苏浩，薛旭升，李超，郭逸风，王星，周文剑，崔闻达，喻祖坤，成佳帅. 煤矿井下移动机器人激光标靶定位方法研究. 煤炭科学技术. 2024(11): 60-73 . 百度学术
10. 景宁波，马宪民，郭卫，秦学斌. 改进动态半径的矿井激光雷达点云滤波算法. 西安科技大学学报. 2023(02): 406-413 . 百度学术
11. 高海跃，王凯，王保兵，王丹丹. 基于全局点云地图的煤矿井下无人机定位方法. 工矿自动化. 2023(08): 81-87+133 . 本站查看
12. 王媛媛. 基于多传感信息融合的学前教育机器人导航定位研究. 自动化与仪器仪表. 2023(09): 233-236 . 百度学术
其他类型引用(2)
计量
- 文章访问数: 109
- HTML全文浏览量: 27
- PDF下载量: 36
- 被引次数: 14
