面向煤矿不安全行为检测的免训练域泛化方法

doi:10.16265/j.cnki.issn1003-3033.2026.S1.0006

中国安全科学学报 ›› 2026, Vol. 36 ›› Issue (S1): 36-41.doi: 10.16265/j.cnki.issn1003-3033.2026.S1.0006

面向煤矿不安全行为检测的免训练域泛化方法

李冉(), 闫力维^**(), 汪昊典

国能数智科技开发(北京)有限公司智能矿山事业部, 北京 100080

收稿日期:2026-02-10 修回日期:2026-05-10 出版日期:2026-06-30
通信作者:
** 闫力维(1981—),男,山西长治人,高级工程师,主要从事智能矿山建设规划和矿山数字化转型研究工作。E-mail:11688301@ceic.com
作者简介:
李冉 (1981—),男,山西大同人,本科,主要从事煤炭行业数智化建设管理工作。E-mail:11688171@ceic.com。
基金资助:
国家能源集团科技创新项目资助(GJNY-25-9)

Training-free domain generalization method for unsafe behavior detection in coal mines

Li Ran(), Yan Liwei^**(), Wang Haodian

Intelligent Mining Division, Guoneng Digital Intelligence Technology Development (Beijing) Co., Ltd., Beijing 100080, China

Received:2026-02-10 Revised:2026-05-10 Published:2026-06-30

摘要/Abstract

摘要：

针对煤矿场景复杂多变环境中人员不安全行为检测的域泛化难题,以及标注数据稀缺、新场景适应成本高的行业痛点,提出一种名为 MineGuard的免训练协同感知框架,首先,通过融合视觉-语言大模型(VLM)通用认知能力与轻量级专家小模型专精检测能力,构建认知-执行分层架构;然后,建立语义驱动的协同推理机制,实现多模型的智能调度与协同验证;最后,构建跨矿区评测基准,验证了框架的泛化性能。结果表明:在煤矿不安全行为数据集上,MineGuard 无需任何微调即可达到 90.22% 的平均精度,较依赖源域训练的传统方法提升 5.65%,展现出优异的跨域泛化性能。

关键词: 煤矿, 不安全行为, 免训练, 域泛化, 大模型

Abstract:

A training-free collaborative perception framework named MineGuard was proposed to address the domain generalization challenge of unsafe personnel behavior detection in complex and dynamic coal mine scenarios, as well as the industrial pain points of scarce annotated data and high adaptation costs for new scenarios. Firstly, the general cognitive capabilities of vision-language large models (VLMs) were integrated with the specialized detection capabilities of lightweight expert models to build a cognition-execution layered architecture. Then, a semantics-driven collaborative reasoning mechanism was established to enable intelligent scheduling and collaborative verification of multiple models. Finally, a cross-mine evaluation benchmark was constructed to validate the framework's generalization performance. The results show that on the coal mine unsafe behavior dataset, MineGuard reaches the mean average precision (mAP) of 90.22% without any fine-tuning, which is 5.65% higher than traditional source-domain-trained methods. It exhibits excellent cross-domain generalization performance.

Key words: coal mine, unsafe behavior, training-free, domain generalization, large model

中图分类号:

X936

李冉, 闫力维, 汪昊典. 面向煤矿不安全行为检测的免训练域泛化方法[J]. 中国安全科学学报, 2026, 36(S1): 36-41.

Li Ran, Yan Liwei, Wang Haodian. Training-free domain generalization method for unsafe behavior detection in coal mines[J]. China Safety Science Journal, 2026, 36(S1): 36-41.

图/表 4

图1

图2

表1

表2

参考文献 18

[1]	林柏泉, 常建华, 翟成. 我国煤矿安全现状及应当采取的对策分析[J]. 中国安全科学学报, 2006, 16(5): 42-46,146.
	Lin Boquan, Chang Jianhua, Zhai Cheng. Analysis of current safety situation in Chinese coal mines and countermeasures to be adopted[J]. China Safety Science Journal, 2006, 16(5): 42-46,146.
[2]	赵亮, 孙魁元, 韩宝虎, 等. 基于人工智能视频分析的选煤厂安全管理研究[J]. 中国安全科学学报, 2021, 31(增刊1): 19-23.
	Zhao Liang, Sun Kuiyuan, Han Baohu, et al. Research on safety management of coal preparation plant based on artificial intelligence video analysis[J]. China Safety Science Journal, 2021, 31(S1): 19-23. doi: 10.16265/j.cnki.issn 1003-3033.2021.S1.004
[3]	Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers[C]. European Conference on Computer Vision, 2020: 213-229.
[4]	Bai Shuai, Chen Keqin, Liu Xuejing, et al. Qwen2.5-VL technical report[J]. arXiv preprint arXiv:2502.13923, 2025.
[5]	张博, 宋元斌, 熊若鑫, 等. 融合人体关节点的安全帽佩戴检测[J]. 中国安全科学学报, 2020, 30(2): 177-182. doi: 10.16265/j.cnki.issn1003-3033.2020.02.028
	Zhang Bo, Song Yuanbin, Xiong Ruoxin, et al. Safety helmet wearing detection fused with human joint points[J]. China Safety Science Journal, 2020, 30(2): 177-182.
[6]	赵冬月, 陈长坤, 易亮. 灾害演化网络时序性风险评估模型[J]. 中国安全科学学报, 2021, 31(3): 171-177. doi: 10.16265/j.cnki.issn1003-3033.2021.03.024
	Zhao Dongyue, Chen Changkun, Yi Liang. A temporal risk assessment model for disaster evolution network[J]. China Safety Science Journal, 2021, 31(3): 171-177. doi: 10.16265/j.cnki.issn1003-3033.2021.03.024
[7]	Wang Zijian, Luo Yadan, Qiu Ruihong, et al. Learning to diversify for single domain generalization[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 834-843.
[8]	Zhao Yinuo, Yuan Jiale, Xu Zhiyuan, et al. Training-free generation of temporally consistent rewards from VLMs[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2025: 8 133-8 143.
[9]	Lee B, Hachiuma R, Wang Yu-Chiang Frank, et al. VLsI: verbalized layers-to-interactions from large to small vision language models[C]. Proceedings of the Computer Vision and Pattern Recognition Conference, 2025: 29 545-29 557.
[10]	Lin Tsung Yi, Maire M, Belongie S, et al. Microsoft COCO: common objects in context[C]. European Conference on Computer Vision, 2014: 740-755.
[11]	Mupparaju S, Thotakura S R, Chintakunta V R R. A review on YOLOv8 and its advancements[C]. International Conference on Data Intelligence and Cognitive Informatics, 2024: 529-545.
[12]	Qiao Sen, Wang Yilin, Li Jian. Real-time human gesture grading based on OpenPose[C]. 2017 10^th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI). IEEE, 2017: 1-6.
[13]	Zhang Yuanhan, Yin Zhenfei, Li Yidong, et al. CelebA-Spoof: large-scale face anti-spoofing dataset with rich annotations[C]. European Conference on Computer Vision, 2020: 70-85.
[14]	Deng Jiankang, Guo Jia, Xue Niannan, et al. ArcFace: additive angular margin loss for deep face recognition[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 4 690-4 699.
[15]	Kirillov A, Mintun E, Ravi N, et al. Segment anything[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 4 015-4 026.
[16]	Liu Wei, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]. European Conference on Computer Vision, 2016: 21-37.
[17]	Girshick R. Fast R-CNN[C]. Proceedings of the IEEE International Conference on Computer Vision, 2015: 1 440-1 448.
[18]	He Kaiming, Gkioxari G, Dollár P, et al. Mask R-CNN[C]. Proceedings of the IEEE International Conference on Computer Vision, 2017: 2 961-2 969.

方法	不安全行为场景检测准确率对比
方法	人员未佩戴安全帽	人员越界	未穿工装	人员跌倒	人员睡岗	平均准确率
SSD	64.17	75.41	64.33	69.54	68.18	68.33
Faster R-CNN	80.52	86.43	75.18	76.49	75.14	78.75
Mask R-CNN	83.65	89.97	78.65	79.53	78.31	82.02
DETR	85.37	90.62	88.47	80.97	77.42	84.57
MineGuard	90.72	93.51	91.69	88.61	86.57	90.22

模型类型	方法	图片定位能力	文本描述能力	是否支持联合输出	是否支持分割
小模型	SSD	支持	不支持	否	否
	Faster R-CNN	支持	不支持	否	否
	Mask R-CNN	支持	不支持	否	否
	DETR	支持	不支持	否	否
大模型	SAM	支持	不支持	否	是
大模型	Qwen2.5VL	支持	支持	是	否
混合模型	MineGuard	支持	支持	是	是

面向煤矿不安全行为检测的免训练域泛化方法

Training-free domain generalization method for unsafe behavior detection in coal mines

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 4

参考文献 18

相关文章 15

编辑推荐

Metrics

本文评价

[1]	贺忠磊, 王海山, 刘立仁. 井工煤矿灾害综合管控平台架构设计[J]. 中国安全科学学报, 2026, 36(S1): 101-107.
[2]	赵艺, 张耀耀, 田素川, 赖万安. 老石旦煤矿软弱特厚煤层高效采充关键技术研究[J]. 中国安全科学学报, 2026, 36(S1): 108-114.
[3]	刘雁宾, 王雪松, 方毅, 孙海南, 李伟, 张燕青. 基于大模型与边缘计算的电厂违章识别管控系统[J]. 中国安全科学学报, 2026, 36(S1): 190-194.
[4]	任意, 赵开功, 李长明. 安全视角下煤矿作业成本法研究[J]. 中国安全科学学报, 2026, 36(S1): 22-28.
[5]	高文才, 孙德宇, 刘建宇, 张全柱. 煤矿电动车辆防爆型智能充电器系统设计[J]. 中国安全科学学报, 2026, 36(S1): 55-60.
[6]	刘娇, 孟煜晖, 郝立敏, 王露露, 佟瑞鹏. 矿工不安全行为影响因素概念模型构建及验证[J]. 中国安全科学学报, 2026, 36(8): 37-45.
[7]	成连华, 刘如萍, 朱文宇, 郭慧敏, 任慧娜. 基于模糊DEMATEL-FCM新生代建筑工人不安全行为研究[J]. 中国安全科学学报, 2026, 36(7): 180-189.
[8]	秦凯, 邓志刚, 舒龙勇, 魏帅豪. 煤矿灾害监测系统数据可靠性评价[J]. 中国安全科学学报, 2026, 36(5): 190-198.
[9]	周宇, 吴鑫, 陈洁. 远距监控下重工业车间人员不安全行为检测算法[J]. 中国安全科学学报, 2026, 36(5): 243-250.
[10]	余洋, 蒋琳, 胡启军, 何乐平, 蔡其杰, 白羽. 基于多元拓扑图卷积网络的建筑工人不安全行为识别[J]. 中国安全科学学报, 2026, 36(4): 19-27.
[11]	张洪硕, 高科, 于健康, 李启文. 煤矿综放采空区岩块自由堆积PNM重构分析[J]. 中国安全科学学报, 2026, 36(2): 153-162.
[12]	谭波, 隋龙琨, 柯巍, 刘衍, 朱权洁, 何宁. 双模型算法耦合识别井下人员不安全行为的方法[J]. 中国安全科学学报, 2026, 36(2): 18-26.
[13]	任利鹏, 佟瑞鹏. 智能化煤矿安全风险多主体动态治理演化博弈研究[J]. 中国安全科学学报, 2026, 36(2): 253-261.
[14]	王乐瑶, 李鑫, 武琪, 佟瑞鹏. 行为风险画像的理论基础、构成要素与构建方法[J]. 中国安全科学学报, 2026, 36(2): 37-43.
[15]	田方圆, 李雯琪, 解学才, 邱伟帅, 李红霞. 基于IMB模型的年长矿工不安全行为致因机制[J]. 中国安全科学学报, 2026, 36(1): 42-49.