一种煤矿井下多目标检测算法

范守俊; 陈希琳; 魏良跃; 王青玉; 张世源; 董飞; 雷少华

doi:10.13272/j.issn.1671-251x.2024090035

一种煤矿井下多目标检测算法

范守俊^1,,
陈希琳²,
魏良跃¹,
王青玉¹,
张世源²,
董飞^{2, 3, ,},
雷少华⁴

1.
兖矿能源集团股份有限公司，山东济宁　273500
2.
中国矿业大学信息与控制工程学院，江苏徐州　221116
3.
安徽大学互联网学院，安徽合肥　230039
4.
徐州高新区安全应急装备产业技术研究院，江苏徐州　221008

基金项目: 国家自然科学基金项目（52074273）；江苏省自然科学基金项目（BK20231060）；兖矿能源集团科学技术项目（YK2023B07-R47）。

详细信息

作者简介:
范守俊（1977—），男，山东淄博人，高级工程师，主要从事智能检测与信息处理工作，E-mail：shoujun_fan1977@163.com

通讯作者:
董飞（1993—），男，安徽庐江人，讲师，硕士，主要研究方向为信号处理与人工智能，E-mail: feidong@cumt.edu.cn。

中图分类号: TD67
计量
- 文章访问数: 64
- HTML全文浏览量: 18
- PDF下载量: 13
出版历程
- 收稿日期: 2024-09-10
- 修回日期: 2024-12-28
- 网络出版日期: 2024-12-05
- 刊出日期: 2024-12-24

An underground coal mine multi-target detection algorithm

1.
Yankuang Energy Group Company Limited, Jining 273500, China
2.
School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China
3.
School of Internet, Anhui University, Hefei 230039, China
4.
Xuzhou High-tech Zone Safety Emergency Equipment Industry Technology Research Institute, Xuzhou 221008, China

摘要

摘要:
目前基于深度学习的煤矿井下目标检测算法在面对光照强度分布不均、目标环境复杂及多类目标尺度分布不均衡时，对复杂小目标的检测效果不佳，易出现漏检和误检现象。针对上述问题，基于单阶段目标检测算法YOLOv8n，提出了一种基于动态蛇形卷积的特征提取（FEDSC）−双向特征金字塔网络与语义和细节融合的特征融合（FFBD）的煤矿井下多目标检测算法，即采用FEDSC替换YOLOv8n的主干网络，扩大感受野；将FFBD作为颈部网络，减少目标误检和漏检；引入SIoU的解耦检测头作为检测层，提高模型对小目标的适应能力与模型收敛速度。实验结果表明：① FEDSC−FFBD算法的mAP@0.5为97.00%，模型参数量为4.22×10⁶个，每秒浮点运算数为21.7×10⁹。② FEDSC−FFBD算法的mAP@0.5较YOLOv8n算法提升了3.40%，对安全帽小目标的识别准确率为90.90%，较YOLOv8n算法提升了11%。③ 与其他YOLO系列算法相比，FEDSC−FFBD算法的mAP@0.5最高，较YOLOv5s，YOLOv9c，YOLOv10n和YOLOv11n算法分别提升了3.60%，1%，10.50%和6.40%。④ FEDSC−FFBD算法在面对煤矿井下光照强度分布不均、目标环境复杂及尺度分布不均衡的条件下，提高了多类别目标的检测精度，改善了小目标漏检和误检的问题。基于FEDSC−FFBD的煤矿井下多目标检测算法在无图像质量增强算法的前提下，克服了光照强度分布不均对小尺度目标检测带来的挑战。
- 煤矿井下多目标检测 /
- YOLOv8n /
- 动态蛇形卷积 /
- CA注意力机制 /
- 特征提取 /
- 特征融合
Abstract:
Currently, underground coal mine target detection algorithms based on deep learning show poor performance in detecting complex small targets under conditions of uneven light intensity distribution, complex target environments, and imbalanced multi-class target scale distribution, often resulting in missed detection and false detection. To address these issues, based on the single-stage target detection algorithm YOLOv8n, this study proposed an underground coal mine multi-target detection algorithm based on feature extraction by dynamic snake convolution (FEDSC)-feature fusion by bi-directional feature pyramid network and semantic and detail fusion (FFBD). FEDSC replaced the backbone network of YOLOv8n to expand the receptive field, while FFBD acted as the neck network to reduce target false detection and missed detection. Additionally, a decoupling detection head of SIoU was used as the detection layer to improve the model's adaptability to small targets and the convergence speed. The results showed that: ① The mAP@0.5 of the FEDSC-FFBD algorithm was 97.00%, the number of model parameters was 4.22×10⁶, and the number of floating point operations per second was 21.7×10⁹. ② The mAP@0.5 of the FEDSC-FFBD alorithm was 3.40% higher than the YOLOv8n algorithm, and the recognition accuracy of the helmet small target was 90.90%, 11% higher than the YOLOv8n algorithm. ③ Compared with other YOLO series algorithms, the FEDSC-FFBD algorithm achieved the highest mAP@0.5, which was 3.60%, 1%, 10.50%, and 6.40% higher than YOLOv5s, YOLOv9c, YOLOv10n, and YOLOv11n algorithms, respectively. ④ The FEDSC-FFBD algorithm improved the detection accuracy of multi-class targets and reduced missed detection and false detection of small targets under conditions of uneven light intensity distribution, complex target environments, and imbalanced target scale distribution in underground coal mine. The underground coal mine multi-target detection algorithm based on FEDSC-FFBD overcame the challenge of small-scale target detection caused by uneven light intensity distribution without relying on image quality enhancement algorithms.
- multi-target detection in underground coal mine /
- YOLOv8n /
- dynamic serpentine convolution /
- CA attention mechanism /
- feature extraction /
- feature fusion

HTML全文

图 1 FEDSC−FFBD算法结构

Figure 1. FEDSC-FFBD algorithm structure

下载: 全尺寸图片幻灯片

图 2 动态蛇形卷积感受野扩大

Figure 2. Dynamic snake convolution receptive field expansion

下载: 全尺寸图片幻灯片

图 3 C2f_DSC结构

Figure 3. C2f _ DSC structure

下载: 全尺寸图片幻灯片

图 4 CA注意力机制结构

Figure 4. CA attention mechanism structure

下载: 全尺寸图片幻灯片

图 5 SDI结构

Figure 5. SDI structure

下载: 全尺寸图片幻灯片

图 6 数据集中部分图像样本

Figure 6. Image samples in the dataset

下载: 全尺寸图片幻灯片

图 7 煤矿井下多目标检测结果

Figure 7. Multi-target detection results in underground coal mine

下载: 全尺寸图片幻灯片

图 8 加入DSConv+CA注意力机制前后检测模型的特征可视化热力图对比

Figure 8. Comparison of feature visualization heat maps of the detection model before and after the addition of DSConv + CA attention mechanism

下载: 全尺寸图片幻灯片

表 1 实验平台环境配置

Table 1 Environment configuration of the experimental platform

名称	版本信息
CPU	Intel（R） i7−13700KF 3.40 GHz
操作系统	Windows 10
内存/GiB	32
GPU	NVIDIA GeForce RTX 4080
CUDA	12.1
Python	3.11
PyTorch	2.1.1

下载: 导出CSV

表 2 目标检测结果对比

Table 2 Comparison of target detectiont results

算法	AP/%							mAP@0.5/%	P/%	FLOPs/10⁹	PM/10⁶个
算法	管道	轨道	人	安全帽	有衣物	胶带	无衣物	mAP@0.5/%	P/%	FLOPs/10⁹	PM/10⁶个
YOLOv8n	99.50	98.80	97.10	79.90	93.40	99.50	86.80	93.60	93.80	8.1	3.00
YOLOv8m	99.50	99.20	96.50	77.90	93.90	99.50	93.20	94.20	94.30	79.1	25.80
YOLOv8l	99.50	99.40	97.50	82.40	95.50	99.50	87.20	94.40	94.90	165.4	43.60
YOLOv8x	99.50	99.30	96.70	85.10	93.80	99.50	91.20	95.00	93.80	257.4	68.10
YOLOv8s	99.50	99.10	96.80	81.90	93.30	99.50	93.50	94.80	92.70	28.5	11.10
YOLOv5n	99.50	99.40	97.20	74.50	92.20	99.50	77.10	91.30	91.50	7.2	2.50
YOLOv5m	99.50	99.30	95.90	76.70	93.70	99.50	81.60	92.30	93.40	64.4	25.10
YOLOv5l	99.50	99.20	95.50	74.80	94.10	99.50	85.90	92.70	94.70	135.3	53.10
YOLOv5x	99.50	98.70	97.40	77.90	92.10	99.50	91.70	93.80	91.50	246.9	97.20
YOLOv5s	99.50	99.00	96.00	81.30	95.20	99.50	83.50	93.40	95.50	23.8	9.10
YOLOv6n	99.50	98.10	95.50	72.20	95.10	99.50	85.80	92.30	94.90	11.6	4.16
YOLOv9c	99.50	99.10	97.10	88.10	96.20	99.50	92.00	96.00	93.10	84.1	21.30
YOLOv10n	99.50	97.10	86.50	65.90	88.90	99.50	68.40	86.50	87.90	8.4	2.70
YOLOv11n	99.50	98.80	96.20	67.90	90.10	99.50	81.90	90.60	89.33	6.3	2.58
FEDSC−FFBD	99.50	98.40	96.70	90.90	96.10	99.50	97.90	97.00	95.90	21.70	4.22

下载: 导出CSV

表 3 消融实验结果

Table 3 Ablation experiment results

模型	AP/%							mAP0.5/%	P/%	FLOPs/10⁹	PM/10⁶个
模型	管道	轨道	人	安全帽	有衣物	胶带	无衣物	mAP0.5/%	P/%	FLOPs/10⁹	PM/10⁶个
YOLOv8n	99.50	98.80	97.10	79.90	93.40	99.50	86.80	93.60	93.80	8.10	3.00
M1	99.50	98.40	97.50	88.90	96.50	99.50	87.20	95.36	95.50	12.70	2.96
M2	99.50	98.60	96.00	89.00	96.50	99.50	88.70	95.40	95.60	12.90	2.97
M3	99.50	99.50	97.50	83.60	96.50	99.50	89.50	95.10	94.30	13.00	2.99
M4	99.50	99.30	96.10	88.90	96.50	99.50	90.50	95.76	93.40	13.50	3.09
M5	99.50	99.40	96.80	93.40	97.00	99.50	92.30	96.84	93.30	21.70	4.22
M6	99.50	98.40	96.50	78.90	93.50	99.50	80.90	92.50	95.30	8.10	3.0
M7	99.50	98.60	96.60	88.60	96.60	99.50	91.60	95.90	93.80	12.70	2.96
M8	99.50	99.30	94.20	86.40	95.90	99.50	98.20	96.10	93.10	12.90	2.97
M9	99.50	98.70	97.30	83.60	97.00	99.50	99.50	96.40	95.40	13.00	2.99
M10	99.50	99.10	97.50	90.50	96.50	99.50	94.50	96.70	95.30	13.50	3.09
M11	99.50	98.40	96.70	90.90	96.10	99.50	97.90	97.00	95.90	21.70	4.22

下载: 导出CSV

表 4 不同注意力机制下的消融实验结果对比

Table 4 Comparison of ablation experiment results under different attention mechanisms

注意力机制	算法	AP/%							mAP@0.5/%	P/%	FLOPs/ 10⁹	PM/ 10⁶个
注意力机制	算法	管道	轨道	人	安全帽	有衣物	胶带	无衣物	mAP@0.5/%	P/%	FLOPs/ 10⁹	PM/ 10⁶个
DAM	FEDSC	99.50	99.20	96.50	87.80	97.00	99.50	83.10	94.70	92.30	13.1	3.23
	FEDSC−BiFPN	99.50	98.40	96.80	90.20	96.10	99.50	96.70	96.70	93.30	13.2	3.25
	FEDSC−BiFPN−DSC	99.50	98.80	96.00	87.70	95.40	99.50	90.70	95.40	90.60	13.7	3.35
	FEDSC−FFBD	99.50	99.40	97.60	85.50	95.30	99.50	82.50	94.20	92.30	21.6	4.39
SEM	FEDSC	99.50	98.70	94.90	86.20	93.70	99.50	83.50	93.70	92.60	12.9	2.97
	FEDSC−BiFPN	99.50	98.70	97.10	87.20	95.70	99.50	99.50	82.10	94.30	13.0	2.99
	FEDSC−BiFPN−DSC	99.50	99.30	96.10	85.70	96.60	99.50	90.30	95.30	90.60	13.5	3.08
	FEDSC−FFBD	99.50	99.40	96.30	87.40	94.70	99.50	80.60	93.90	96.30	21.4	4.13
CBAM	FEDSC	99.50	99.10	96.80	89.10	96.60	99.50	85.00	95.10	94.10	12.9	2.98
	FEDSC−BiFPN	99.50	98.90	94.90	87.10	96.60	99.50	84.30	94.40	92.80	13.0	3.0
	FEDSC−BiFPN−DSC	99.50	98.70	96.70	86.40	94.90	99.50	84.80	94.40	94.10	13.5	3.09
	FEDSC−FFBD	99.50	99.30	96.50	88.20	97.40	99.50	90.40	95.80	95.10	21.4	4.14
EMA	FEDSC	99.50	99.50	96.70	89.00	94.20	99.50	95.40	96.20	92.30	12.9	2.97
	FEDSC−BiFPN	99.50	98.80	97.10	90.00	93.10	99.50	94.70	96.10	94.00	12.9	2.98
	FEDSC−BiFPN−DSC	99.50	99.40	97.30	88.20	97.70	99.50	90.80	96.10	94.20	13.4	3.08
	FEDSC−FFBD	99.50	99.40	97.00	91.50	96.40	99.50	79.20	94.60	94.50	21.4	4.13
CA	FEDSC	99.50	99.30	94.20	86.40	95.90	99.50	98.20	96.10	93.10	12.9	2.97
	FEDSC−BiFPN	99.50	98.70	97.30	83.60	97.00	99.50	99.50	96.40	95.40	13.0	2.99
	FEDSC−BiFPN−DSC	99.50	99.10	97.50	90.50	96.50	99.50	94.50	96.70	95.30	13.5	3.09
	FEDSC−FFBD	99.50	98.40	96.70	90.90	96.10	99.50	97.90	97.00	95.90	21.7	4.22

下载: 导出CSV

参考文献(20)

[1]	程德强,钱建生,郭星歌,等. 煤矿安全生产视频AI识别关键技术研究综述[J]. 煤炭科学技术,2023,51(2):349-365. CHENG Deqiang,QIAN Jiansheng,GUO Xingge,et al. Review on key technologies of AI recognition for videos in coal mine[J]. Coal Science and Technology,2023,51(2):349-365.
[2]	王树奇,刘贝,邹斐. 一种新的矿井监控视频增强目标检测算法[J]. 西安科技大学学报,2019,39(2):347-353. WANG Shuqi,LIU Bei,ZOU Fei. A new image enhancement target detection algorithm based on monitoring video in coal mine tunnel[J]. Journal of Xi'an University of Science and Technology,2019,39(2):347-353.
[3]	寇发荣,肖伟,何海洋,等. 基于改进YOLOv5的煤矿井下目标检测研究[J]. 电子与信息学报,2023,45(7):2642-2649. DOI: 10.11999/JEIT220725 KOU Farong,XIAO Wei,HE Haiyang,et al. Research on target detection in underground coal mines based on improved YOLOv5[J]. Journal of Electronics & Information Technology,2023,45(7):2642-2649. DOI: 10.11999/JEIT220725
[4]	崔铁军,王凌霄. YOLOv4目标检测算法在煤矿工人口罩佩戴监测工作中的应用研究[J]. 中国安全生产科学技术,2021,17(10):66-71. CUI Tiejun,WANG Lingxiao. Research on application of YOLOv4 object detection algorithm in monitoring on masks wearing of coal miners[J]. Journal of Safety Science and Technology,2021,17(10):66-71.
[5]	董彦强,程德强,张云鹤,等. 基于注意力和重构特征融合的轻量级煤矿安全帽检测方法[J]. 计算机工程与应用,2024,60(15):297-306. DOI: 10.3778/j.issn.1002-8331.2304-0421 DONG Yanqiang,CHENG Deqiang,ZHANG Yunhe,et al. Lightweight coal mine safety helmet detection method based on attention and reconfiguration feature fusion[J]. Computer Engineering and Applications,2024,60(15):297-306. DOI: 10.3778/j.issn.1002-8331.2304-0421
[6]	李伟山,卫晨,王琳. 改进的Faster RCNN煤矿井下行人检测算法[J]. 计算机工程与应用,2019,55(4):200-207. DOI: 10.3778/j.issn.1002-8331.1711-0282 LI Weishan,WEI Chen,WANG Lin. Improved faster RCNN approach for pedestrian detection in underground coal mine[J]. Computer Engineering and Applications,2019,55(4):200-207. DOI: 10.3778/j.issn.1002-8331.1711-0282
[7]	LI Dongjun,ZHANG Zhenxin,XU Zhihua,et al. An image-based hierarchical deep learning framework for coal and gangue detection[J]. IEEE Access,2019,7. DOI: 10.1109/ACCESS.2019.2961075.
[8]	LIANG Bin,WANG Zhongbin,SI Lei,et al. A novel pressure relief hole recognition method of drilling robot based on SinGAN and improved faster R-CNN[J]. Applied Sciences,2022,13(1). DOI: 10.3390/APP13010513.
[9]	章赛,纪凡,卢才武,等. 低照度下改进YOLOX的煤矿无人电机车轨道障碍物检测方法[J]. 安全与环境学报,2024,24(3):952-961. ZHANG Sai,JI Fan,LU Caiwu,et al. An improved YOLOX detection method for tracking obstacles of unmanned electric locomotives in coal mines under low lighting[J]. Journal of Safety and Environment,2024,24(3):952-961.
[10]	唐俊,李敬兆,石晴,等. 基于Faster−YOLOv7的带式输送机异物实时检测[J]. 工矿自动化,2023,49(11):46-52,66. TANG Jun,LI Jingzhao,SHI Qing,et al. Real time detection of foreign objects in belt conveyors based on Faster-YOLOv7[J]. Journal of Mine Automation,2023,49(11):46-52,66.
[11]	HUANG Kaifeng,LI Shiyan,CAI Feng,et al. Detection of large foreign objects on coal mine belt conveyor based on improved[J]. Processes,2023,11(8). DOI: 10.3390/PR11082469.
[12]	LUO Bingxin,KOU Ziming,HAN Cong,et al. A faster and lighter detection method for foreign objects in coal mine belt conveyors[J]. Sensors,2023,23(14). DOI: 10.3390/S23146276.
[13]	杨文轲,孟祥瑞,王向前. 基于改进YOLOv7的煤矿工人不安全行为识别方法[J]. 哈尔滨商业大学学报(自然科学版),2024,40(6):664-670. YANG Wenke,MENG Xiangrui,WANG Xiangqian. Method for identifying unsafe behaviors of coal mine workers based on improved YOLOv7[J]. Journal of Harbin University of Commerce(Natural Sciences Edition),2024,40(6):664-670.
[14]	田佳伟, 唐子山. 基于边缘计算和ST−YOLO的矿井智能监控技术研究[J]. 煤炭工程, 2024, 56（7）: 165-173. TIAN Jiawei, TANG Zishan. Mine intelligent monitoring technology based on edge computing and ST-YOLO.[J]. Coal Engineering, 2024, 56（7）: 165-173.
[15]	QI Yaolei,HE Yuting,QI Xiaoming,et al. Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation[C]. IEEE/CVF International Conference on Computer Vision ,Paris,2023:6070-6079.
[16]	HOU Qibin,ZHOU Daquan,FENG Jiashi. Coordinate attention for efficient mobile network design[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Nashville,2021:13713-13722.
[17]	TAN Mingxing,PANG Ruoming,LE Q V. EfficientDet:scalable and efficient object detection[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Seattle,2020:10778-10787.
[18]	ZHOU Shilong,ZHOU Haijin. Detection based on semantics and a detail infusion feature pyramid network and a coordinate adaptive spatial feature fusion mechanism remote sensing small object detector[J]. Remote Sensing,2024,16(13). DOI: 10.3390/RS16132416.
[19]	CHEN Fuxun,ZHANG Lanxin,KANG Siyu,et al. Soft-NMS-enabled YOLOv5 with SIOU for small water surface floater detection in UAV-captured images[J]. Sustainability,2023,15(14). DOI: 10.3390/SU151410751.
[20]	XIE Weining,SUN Xiaoyong,MA Weifeng. A light weight multi-scale feature fusion steel surface defect detection model based on YOLOv8[J]. Measurement Science and Technology,2024,35(5). DOI: 10.1088/1361-6501/AD296D.

施引文献(13)

期刊类型引用(8)

1.	张玉涛，路旭，李亚清，张园勃，车博. 低甲烷气氛下褐煤的低温氧化特性研究. 安全与环境学报. 2024(02): 517-524 . 百度学术
2.	王斌，贾澎涛，郭风景，孙刘咏，林开义. 基于多特征融合的煤自燃温度深度预测模型. 中国矿业. 2024(02): 84-90 . 百度学术
3.	田富超，贾东旭，陈明义，梁运涛，朱红青，张同浩. 采空区复合灾害环境下含瓦斯煤自燃特征研究进展. 煤炭学报. 2024(06): 2711-2727 . 百度学术
4.	叶正亮，龚选平，尚博，胡冕. 煤自燃全过程精准预测预报指标体系研究与应用. 矿业研究与开发. 2023(09): 141-145 . 百度学术
5.	贾澎涛，林开义，郭风景. 基于PSO-SRU深度神经网络的煤自燃温度预测模型. 工矿自动化. 2022(04): 105-113 . 本站查看
6.	董轩萌，郭立稳，张少华，董宪伟，王福生. 氧化煤阻化性能的FTIR研究. 工业安全与环保. 2021(07): 34-39 . 百度学术
7.	骆大勇，刘振. 许疃煤矿煤炭自燃指标气体优选与应用. 矿业安全与环保. 2021(05): 69-74 . 百度学术
8.	刘宝，穆坤，叶飞，汪帆，王静婷. 基于相关向量机的煤自燃预测方法. 工矿自动化. 2020(09): 104-108 . 本站查看