Research on personnel re-identification in complex underground environment
-
摘要: 对煤矿井下视频监控中的人员身份进行智能识别，对提高人员监管效率、减少安全事故发生具有重要意义。受井下环境复杂、视频监控设备性能局限性影响，井下视频监控图像存在分辨率低、遮挡、背景干扰等问题，导致井下人员间差异性较小，人员重识别准确率低。针对上述问题，提出了一种基于通道注意力和距离度量的网络结构，并将其用于井下复杂环境人员重识别。针对监控图像中人员与背景不易区分的问题，在骨干网络中引入通道注意力模块，使其更加关注人员的前景特征而抑制背景信息，并将骨干网络最后一层输出的特征图大小扩大1倍，以获得更多的细粒度特征，丰富人员的特征信息，增强网络对特征的学习能力；在实现不同身份人员分类的基础上，利用人员图像间的绝对距离信息，通过距离度量模块对难以识别的人员图像进行采样和加权处理，增加难样本在反向传播时的权重，使网络更加关注具有判别力的人员特征；联合使用身份损失和距离度量损失优化特征层，使网络提取出更具判别力的人员特征，从而提高重识别准确度。采用Miner-CUMT数据集对提出的井下复杂环境人员重识别方法进行验证，结果表明该方法可充分利用井下不同身份人员的关键信息，使识别网络具有更强的判别能力，提高了井下作业人员身份识别准确度。Abstract: Intelligent identification of personnel in underground video monitoring in coal mines is of great significance for improving the efficiency of personnel supervision and reducing the occurrence of safety accidents. Affected by the complex underground environment and the performance limitations of monitoring video equipment, the underground video monitoring images have problems such as low resolution, occlusion and background interference, resulting in small differences among underground personnel and low accuracy of personnel re-identification. In order to solve the above problems, a network structure based on distance metric and channel attention is proposed. The structure is used for personnel re-identification in complex underground environments. In order to solve the problem that it is not easy to distinguish personnel from background in monitoring images, a channel attention module is introduced into the backbone network to make it pay more attention to the foreground characteristics of personnel and suppress the background information. Moreover, the size of the characteristic map output from the last layer of the backbone network is doubled so as to obtain more fine-grained characteristics, enrich the characteristic information of personnel and enhance the network's ability to learn characteristics. On the basis of realizing the classification of personnel with different identities, using the absolute distance information between the images of personnel, the distance metric module is used to sample and weight the personnel images who are difficult to identify, increase the weight of the difficult samples in the back propagation, and make the network pay more attention to the discriminative personnel characteristics. The identity loss and distance metric loss are jointly used to optimize the characteristic layer, so that the network can extract more discriminative personnel characteristics to improve the re-identification accuracy. The Miner-CUMT data set is used to verify the proposed method for personnel re-identification in complex underground environments. The results show that the method can make full use of the key information of personnel with different identities in the underground, so that the identification network has stronger discrimination ability and improves the accuracy of personnel identification in the underground.
-
-
期刊类型引用(12)
1. 李梅，张鹏鹏，李元琛. AI赋能智能矿山:应用场景及未来展望. 煤炭经济研究. 2025(02): 161-169 . 
百度学术
2. 张美晨，张凯，赵丽娟，史百胜，王宇洋. 煤岩识别及采煤机自适应控制技术研究进展. 山西焦煤科技. 2023(12): 4-10 . 百度学术
3. 张强，张润鑫，刘峻铭，王聪，张赫哲，田莹. 煤矿智能化开采煤岩识别技术综述. 煤炭科学技术. 2022(02): 1-26 . 百度学术
4. 李彦明，孙利海. 基于多源异构信息耦合的煤岩界面识别技术研究. 矿业安全与环保. 2022(05): 6-10 . 百度学术
5. 陈家璘，周正，冯伟东，贺易，李静茹，赵世文. 一种无线传感器网络节点的故障检测算法. 计算技术与自动化. 2021(01): 38-42 . 百度学术
6. 窦国贤，高杨. 一种改进的电网信息系统自动化故障融合监测技术. 计算技术与自动化. 2020(01): 34-38 . 百度学术
7. 赵晓群，曹磊，张伍. 应用模块化+大数据处理技术提高计量新装接线率. 电子测量技术. 2020(05): 150-155 . 百度学术
8. 张林锋，田慕琴，宋建成，贺颖，冯君玲，刘西青. 基于多源数据融合的掘进机截割岩壁硬度识别方法. 振动与冲击. 2020(13): 7-15 . 百度学术
9. 窦国贤，高杨. 基于小波分析方法检测电网信息故障的研究. 计算机测量与控制. 2020(09): 38-41+57 . 百度学术
10. 王思. 基于多传感器的运动员训练信息融合分析系统设计. 计算技术与自动化. 2020(03): 140-146 . 百度学术
11. 王俊成，杜晓坤. 基于小波包分解和模糊神经网络的煤岩界面识别. 科技经济导刊. 2019(04): 48+36 . 百度学术
12. 刘忠超，刘勇军. 煤岩识别现状分析与发展方向. 南阳理工学院学报. 2018(04): 26-30 . 百度学术
其他类型引用(16)
计量
- 文章访问数: 167
- HTML全文浏览量: 13
- PDF下载量: 22
- 被引次数: 28
下载:
