基于改进STDC的井下轨道区域实时分割方法

马天; 李凡卉; 杨嘉怡; 张杰慧; 丁旭涵

doi:10.13272/j.issn.1671-251x.2023080076

基于改进STDC的井下轨道区域实时分割方法

doi: 10.13272/j.issn.1671-251x.2023080076

1.
西安科技大学计算机科学与技术学院，陕西西安　710054
2.
西安科技大学安全科学与工程学院，陕西西安　710054

基金项目: 国家重点研发计划项目（2021YFB4000905）；国家自然科学基金项目（62101432，62102309）；陕西省自然科学基础研究计划项目（2022JM-508）。

详细信息

作者简介:
马天（1982—），男，河南商丘人，副教授，博士，研究方向为图形图像处理、数据可视化，E-mail：matian@xust.edu.cn

通讯作者:
李凡卉（1998—），女，陕西西安人，硕士研究生，研究方向为图像处理，E-mail：julyhuizili@163.com。

中图分类号: TD67
计量
- 文章访问数: 166
- HTML全文浏览量: 68
- PDF下载量: 22
- 被引次数: 0
出版历程
- 收稿日期: 2023-08-22
- 修回日期: 2023-11-15
- 网络出版日期: 2023-11-23

Real time segmentation method for underground track area based on improved STDC

1.
College of Computer Science and Technology, Xi'an University of Science and Technology, Xi'an 710054, China
2.
College of Safety Science and Engineering, Xi'an University of Science and Technology, Xi'an 710054, China

摘要

摘要: 目前中国大部分井下轨道运输场景较为开放，存在作业人员、散落物料或煤渣侵入到轨道上的问题，从而给机车行驶带来威胁。煤矿井下轨道区域多呈线性或弧形不规则区域，且轨道会逐渐收敛，采用目标识别框或检测轨道线的方法划分轨道区域难以精确获得轨道范围，采用轨道区域的分割可实现像素级别的精确轨道区域检测。针对目前井下轨道区域分割方法存在边缘信息分割效果差、实时性低的问题，提出了一种基于改进短期密集连接（STDC）网络的轨道区域实时分割方法。采用STDC作为骨干架构，以降低网络参数量与计算复杂度。设计了基于通道注意机制的特征注意力模块（FAM），用于捕获通道之间的依赖关系，对特征进行有效的细化和组合。使用特征融合模块（FFM）融合高级语义特征与浅层特征，并利用通道和空间注意力丰富融合特征表达，从而有效获取特征并减少特征信息丢失，提升模型性能。采用二值交叉熵损失、骰子损失及图像质量损失来优化详细信息的提取，并通过消除冗余结构来提高分割效率。在自建的数据集上对基于改进STDC的轨道区域实时分割方法进行验证，结果表明：该方法的平均交并比（MIoU）为95.88%，较STDC提高了3%；参数量为6.74 MiB，较STDC降低了18.3%；随着迭代次数增加，优化后的损失函数值持续减小，且较STDC降低更为明显；基于改进STDC的轨道区域实时分割方法的MIoU达95.88%，帧速率为37.8帧/s，参数量为6.74 MiB，准确率为99.46%。该方法可完整识别轨道区域，轨道被准确地分割且边缘轮廓完整准确。
- 井下轨道区域 /
- 语义分割 /
- 短期密集连接网络 /
- 特征注意力 /
- 特征融合 /
- 注意力机制
Abstract: Currently, most underground rail transportation scenarios in China are relatively open. There are problems of operators, scattered materials, or coal slag invading the track. It poses a threat to locomotive operation. The underground track area of coal mines often presents linear or arc-shaped irregular areas, and the track gradually converges. It is difficult to accurately obtain the track range by using object recognition boxes or detecting track lines to divide the track area. Using track area segmentation can achieve pixel level accurate track area detection. Aiming at the problems of poor edge information segmentation and low real-time performance in current underground track area segmentation methods, a real-time track area segmentation method based on improved network short-term dense concatenate (STDC) is proposed. STDC is adopted as the backbone architecture to reduce the amount of network parameters and computational complexity. A feature attention module (FAM) based on channel attention mechanism is designed to capture the dependency relationships between channels and effectively refine and combine features. The feature fusion module (FFM) is used to fuse advanced semantic features with shallow features. The channel and spatial attention are utilized to enrich the fusion feature expression, effectively obtaining features and reducing feature information loss, improving model performance. Binary cross entropy loss, dice loss, and image quality loss are used to optimize the extraction of detailed information, and to improve segmentation efficiency by eliminating redundant structures. By verifying on a self built dataset, the results show the following points. The mean intersection over union (MIoU) of the improved STDC based real-time segmentation method for track area is 95.88, which is 3% higher than STDC. The number of parameters is 6.74 MiB, which is 18.3% lower than STDC. As the number of iterations increases, the optimized loss function value continues to decrease, and the decrease in function value is more significant than that of the original model. The MIoU of the improved STDC based real-time segmentation method for track area reaches 95.88%, frames per second is 37.8 frames/s, the number of parameters is 6.74 MiB, and accuray rate is 99.46%. This method can fully recognize the track area, accurately segment the track, and provide complete and accurate edge contours.
- underground track area /
- semantic segmentation /
- network with short-term dense concatenate /
- feature attention /
- feature fusion /
- feature fusion

HTML全文

图 1 基于改进STDC的实时分割方法的整体网络结构

Figure 1. Overall network structure of real time segmentation method based on improved short-term dense concatenate (STDC)

下载: 全尺寸图片幻灯片

图 2 STDC模型结构

Figure 2. Module structure of short-term dense concatenate （STDC）

下载: 全尺寸图片幻灯片

图 3 FAM结构

Figure 3. Structure of feature attention module（FAM）

下载: 全尺寸图片幻灯片

图 4 FFM结构

Figure 4. Structure of feature fusion module（FFM）

下载: 全尺寸图片幻灯片

图 5 损失函数优化结果

Figure 5. Loss of the function optimization results

下载: 全尺寸图片幻灯片

图 6 不同方法的轨道分割效果

Figure 6. Track segmentation effect of different methods

下载: 全尺寸图片幻灯片

表 1 消融实验结果

Table 1. Results of ablation experiment

STDC	FAM	FFM	MIoU/%	Params/MiB
√			92.88	8.25
√	√		93.74	8.25
√		√	93.21	6.74
√	√	√	95.88	6.74

下载: 导出CSV

表 2 不同方法的轨道分割性能

Table 2. Track segmentation performance of different methods

方法	主干网络	MIoU/%	FPS/（帧·s⁻¹）	Params/MiB	准确率/%
ENet	−	75.84	23.3	0.35	95.96
SegNet	VGG16	84.31	17.6	29.61	97.53
Deeplab3v+	ResNet18	90.72	9.2	12.38	98.61
BiSeNetV2	ResNet18	82.10	31.3	2.32	97.3
SFNet	ResNet18	94.27	26.6	13.8	99.25
STDCSeg	STDC1	92.88	32.3	8.25	99.12
文献[12]方法	ResNet18	87.19	30.7	4.89	97.89
本文方法	STDC1	95.88	37.8	6.74	99.46

下载: 导出CSV

参考文献(21)

[1]	王国法,王虹,任怀伟,等. 智慧煤矿2025情景目标和发展路径[J]. 煤炭学报,2018,43(2):295-305. doi: 10.13225/j.cnki.jccs.2018.0152 WANG Guofa,WANG Hong,REN Huaiwei,et al. 2025 scenarios and development path of intelligent coal mine[J]. Journal of China Coal Society,2018,43(2):295-305. doi: 10.13225/j.cnki.jccs.2018.0152
[2]	胡青松,孟春蕾,李世银,等. 矿井无人驾驶环境感知技术研究现状及展望[J]. 工矿自动化,2023,49(6):128-140. doi: 10.13272/j.issn.1671-251x.18115 HU Qingsong,MENG Chunlei,LI Shiyin,et al. Research status and prospects of perception technology for unmanned mining vehicle driving environment[J]. Journal of Mine Automation,2023,49(6):128-140. doi: 10.13272/j.issn.1671-251x.18115
[3]	ASSIDIQ A A,KHALIFA O O,ISLAM M R,et al. Real time lane detection for autonomous vehicles[C]. International Conference on Computer and Communication Engineering,Kuala Lumpur,2008:82-88.
[4]	FATIH K,YUSUF S A. Vision-based railroad track extraction using dynamic programming [C]. International IEEE Conference on Intelligent Transportation Systems,Saint Louis,2009:42-47.
[5]	谢昭莉,王壬,张德全. 基于图像识别的井下机车轨道检测方法[J]. 计算机工程,2012,38(14):147-149. XIE Zhaoli,WANG Ren,ZHANG Dequan. Track detection method of underground locomotive based on image recognition[J]. Computer Engineering,2012,38(14):147-149.
[6]	李晓明,郎文辉,马忠磊,等. 基于图像处理的井下机车行人检测技术[J]. 煤矿机械,2017,38(4):167-170. doi: 10.13436/j.mkjx.201704059 LI Xiaoming,LANG Wenhui,MA Zhonglei,et al. Pedestrian detection technology for mine locomotive based on image processing[J]. Coal Mine Machinery,2017,38(4):167-170. doi: 10.13436/j.mkjx.201704059
[7]	王国法,刘峰,庞义辉,等. 煤矿智能化——煤炭工业高质量发展的核心技术支撑[J]. 煤炭学报,2019,44(2):349-357. doi: 10.13225/j.cnki.jccs.2018.2041 WANG Guofa,LIU Feng,PANG Yihui,et al. Coal mine intellectualization:the core technology of high quality development[J]. Journal of China Coal Society,2019,44(2):349-357. doi: 10.13225/j.cnki.jccs.2018.2041
[8]	韩江洪,乔晓敏,卫星,等. 基于空间卷积神经网络的井下轨道检测方法[J]. 电子测量与仪器学报,2018,32(12):34-43. doi: 10.13382/j.jemi.2018.12.005 HAN Jianghong,QIAO Xiaomin,WEI Xing,et al. Downhole track detection method based on spatial convolutional neural network[J]. Journal of Electronic Measurement and Instrumentation,2018,32(12):34-43. doi: 10.13382/j.jemi.2018.12.005
[9]	卫星,刘邵凡,杨国强,等. 基于改进双边分割网络的井下轨道检测算法[J]. 计算机应用研究,2020,37(增刊1):348-350. WEI Xing,LIU Shaofan,YANG Guoqiang,et al. Track detection algorithm via modified bilateral segmentation network[J]. Application Research of Computers,2020,37(S1):348-350.
[10]	鲍新平,汪涛. 基于AI视觉智能识别的煤矿斜井轨道运输安全管理系统[J]. 工矿自动化,2023,49(增刊1):72-75. BAO Xinping,WANG Tao. A safety management system for coal mine inclined shaft rail transportation based on intelligent AI visual recognition[J]. Journal of Mine Automation,2023,49(S1):72-75.
[11]	杨荣锦,张秀峰,龚莉娜,等. 基于深度学习的车道线检测方法综述[J]. 大连民族大学学报,2021,23(1):40-44. doi: 10.3969/j.issn.1009-315X.2021.01.009 YANG Rongjin,ZHANG Xiufeng,GONG Li'na,et al. Survey of lane detection methods based on deep learning[J]. Journal of Dalian Minzu University,2021,23(1):40-44. doi: 10.3969/j.issn.1009-315X.2021.01.009
[12]	周华平,郑锐. 基于改进BiSeNet的煤矿井下轨道检测算法[J]. 湖北民族大学学报(自然科学版),2021,39(4):398-403. ZHOU Huaping,ZHENG Rui. Underground rail detection algorithm based on improved BiSeNet[J]. Journal of Hubei Minzu University(Natural Science Edition),2021,39(4):398-403.
[13]	YU Changqian,WANG Jingbo,PENG Chao,et al. BiSeNet:bilateral segmentation network for real-time semantic segmentation[C]. European Conference on Computer Vision,Berlin,2018:334-349.
[14]	YU Changqian,GAO Changxin,WANG Jingbo,et al. BiSeNet V2:bilateral network with guided aggregation for real-time semantic segmentation[J]. International Journal of Computer Vision,2021,129(11). DOI: 10.1007/S11263-021-01515-2.
[15]	FAN Mingyuan,LAI Shenqi,HUANG Junshi,et al. Rethinking bisenet for real-time semantic segmentation[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Nashville,2021:9716-9725.
[16]	ZHU Xizhou,CHENG Dazhi,ZHANG Zheng,et al. An empirical study of spatial attention mechanisms in deep networks[C]. IEEE/CVF International Conference on Computer Vision,Piscataway,2019:6688-6697.
[17]	HAO Yuying,LIU Yi,CHEN Yizhou,et al. EISeg:an efficient interactive segmentation annotation tool based on paddlepaddle[C]. Computer Vision and Pattern Recognition,New Orleans,2022. DOI: 10.48550/arXiv.2210.08788.
[18]	PASZKE A,CHAURASIA A,KIM S,et al. Enet:a deep neural network architecture for real-time semantic segmentation[C]. IEEE Conference on Computer Vision and Pattern Recognition,Las Vegas,2016. DOI: 10.48550/arXiv.1606.02147.
[19]	BADRINARAYANAN V,KENDALL A,CIPOLLA R. SegNet:a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(12):2481-2495.
[20]	CHEN L C,ZHU Yukun,PAPANDREOU G,et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]. European Conference on Computer Vision,Munich,2018:801-818.
[21]	LI Xiangtai,YOU Ansheng,ZHU Zhen,et al. Semantic flow for fast and accurate scene parsing[C]. European Conference on Computer Vision,Berlin,2020:775-793.