Please wait a minute...
工程设计学报  2020, Vol. 27 Issue (5): 568-576    DOI: 10.3785/j.issn.1006-754X.2020.00.076
设计理论与方法学     
基于语义分割的火车车厢位置检测研究
卢进南, 单德兴
辽宁工程技术大学 机械工程学院, 辽宁 阜新 123000
Research on railway carriage position detection based on semantic segmentation
LU Jin-nan, SHAN De-xing
School of Mechanical Engineering, Liaoning Technical University, Fuxin 123000, China
 全文: PDF(3312 KB)   HTML
摘要: 为了实现在煤炭定量装车站装车过程中实时检测火车车厢位置,为溜槽升降提供触发信号,设计了一种基于语义分割的火车车厢位置检测模型。以FPN (feature pyramid networks,特征金字塔网络)和ResNet101 (residual network,残差网络)为主干网络,提取并融合分辨率、语义强度不同的特征图;结合基于期望最大化(expectation maximization, EM)算法的注意力机制,构建车厢上边框语义分割模型,用于过滤特征图中的噪声,提高图像边界的语义分割精度;设计位置检测模块,计算语义分割后图像中各类别的面积及其比例和车厢上边框外接矩形高度,以获取火车车厢位置信息。结果表明,所构建的车厢上边框语义分割模型在测试集上的mIoU (mean intersection over union,均交并比)为81.21%,mPA (mean pixel accuracy,平均像素精度)为88.64%,相比未引入注意力机制的语义分割模型分别提升了3.91%和7.44%。在煤炭定量装车站现场进行的火车车厢位置检测试验结果表明,基于语义分割的火车车厢位置检测模型的检测精度满足煤炭装车过程中车厢位置检测任务的要求,这为实现煤炭定量装车系统的智能化提供了新思路。
Abstract: In order to realize the real-time detection of railway carriage position during the loading process of the quantitative coal loading station and provide a trigger signal for the chute lifting, a railway carriage position detection model based on the semantic segmentation is proposed. Taking the FPN (feature pyramid networks) and ResNet101 (residual network) as the backbone network, the feature maps with different resolution and semantic strength were extracted and fused.Combined with the attention mechanism based on the expectation maximization (EM) algorithm, a carriage upper frame semantic segmentation model was constructed to filter the noise in the feature map and improve the semantic segmentation accuracy of the image boundary. A position detection module was designed to calculate the area of each category and their proportions in the image after semantic segmentation and the height of circumscribed rectangle of the carriage upper frame, so as to obtain the position information of railway carriage. The results showed that the mIoU (mean intersection over union) and the mPA (mean pixel accuracy) of the constructed carriage upper frame semantic segmentation model on the test set was 81.21% and 88.64%, respectively. Compared with the semantic segmentation model without attention mechanism, it was improved by 3.91% and 7.44%. The railway carriage position detection test was conducted at the coal quantitative loading station site, and the results indicated that the detection accuracy of the railway carriage position detection model based on the semantic segmentation met the requirements of the carriage position detection task in the coal loading process. It provides a new idea for the realization of intelligent quantitative coal loading system.
收稿日期: 2020-04-07 出版日期: 2020-10-28
CLC:  TP 751.1  
作者简介: 卢进南(1979—),男,辽宁丹东人,副教授,博士,从事机电液装备自动化及智能控制基础和技术研究,E-mail:21020331@qq.com,https://orcid.org/0000-0002-8046-2672;
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  

引用本文:

卢进南, 单德兴. 基于语义分割的火车车厢位置检测研究[J]. 工程设计学报, 2020, 27(5): 568-576.

LU Jin-nan, SHAN De-xing. Research on railway carriage position detection based on semantic segmentation. Chinese Journal of Engineering Design, 2020, 27(5): 568-576.

链接本文:

https://www.zjujournals.com/gcsjxb/CN/10.3785/j.issn.1006-754X.2020.00.076        https://www.zjujournals.com/gcsjxb/CN/Y2020/V27/I5/568

No related articles found!