基于改进RT−DETR的井下输送带跑偏故障检测算法

安龙辉; 王满利; 张长森

doi:10.13272/j.issn.1671-251x.2024080089

基于改进RT−DETR的井下输送带跑偏故障检测算法

河南理工大学物理与电子信息学院，河南焦作　454000

基金项目: 国家自然科学基金项目(52074305)；河南省科技攻关项目(242102221006)。

详细信息

作者简介:
安龙辉（1999—），男，河南新乡人，硕士研究生，研究方向为目标检测、图像处理，E-mail：alh1117@163.com

通讯作者:
王满利(1981—)，男，河南焦作人，副教授，博士，研究方向为图像处理、人工智能、智慧矿山、工业过程控制，E-mail:wml920@163.com。

中图分类号: TD634
计量
- 文章访问数: 45
- HTML全文浏览量: 7
- PDF下载量: 9
出版历程
- 收稿日期: 2024-08-29
- 修回日期: 2025-03-22
- 网络出版日期: 2025-03-26
- 刊出日期: 2025-03-14

Fault detection algorithm for underground conveyor belt deviation based on improved RT-DETR

School of Physics and Electronic Information, Henan Polytechnic University, Jiaozuo 454000, China

摘要

摘要:
目前输送带跑偏检测研究主要集中于提取输送带边缘的直线特征，该方式需设定特定阈值，易受环境因素的制约，导致检测速度慢、精度不高。针对该问题，提出了一种基于改进RT−DETR的井下输送带跑偏故障检测算法，使用改进RT−DETR直接对一组托辊检测，根据左右托辊的暴露程度识别是否跑偏。针对实时检测转换器（RT−DETR）主干网络进行3个方面的改进：① 为了减少主干网络的参数量和浮点运算数量（FLOPs），使用FasterNet Block替换ResNet34中的BasicBlock；② 为了提升模型的精度和效率，在FasterNet Block结构中，引入结构重参数化的思想；③ 为了提升FasterNet Block在特征提取方面的性能，引入了高效多尺度注意力机制(EMA)，更加有效地捕捉全局和局部特征图。为了拓展感受野并捕获更有效、更广泛的上下文信息，以获得更为丰富的特征表达，采用改进高级筛选特征融合金字塔网络（HS−FPN）来优化多尺度特征融合。实验结果表明，与基准模型相比较，改进RT−DETR模型的参数量和FLOPs分别减少了8.4×10⁶ 个和17.8 G，mAP@0.5达94.5%，严重跑偏检测精度达99.2%，检测速度达41.0 帧/s，优于TOOD，ATSS等目标检测模型，满足煤矿生产对目标检测实时性和准确性的需求。
- 输送带跑偏 /
- 目标检测 /
- 实时检测转换器 /
- 结构重参数化 /
- 高效多尺度注意力机制 /
- 多尺度特征融合
Abstract:
Current research on conveyor belt deviation detection mainly focuses on extracting the straight-line features of belt edges. The method requires setting specific thresholds and is easily affected by environmental factors, resulting in slow detection speed and low accuracy. To address the issue, an underground conveyor belt deviation fault detection algorithm based on an improved real-time detection transformer (RT-DETR) was proposed. The improved RT-DETR was used to directly detect a set of idlers and identify deviation based on the exposure degree of the left and right idlers. Three improvements were made to the RT-DETR backbone network: ① To reduce the number of parameters and floating-point operations (FLOPs), FasterNet Block was used to replace the BasicBlock in ResNet34. ② To enhance model accuracy and efficiency, the concept of structural reparameterization was introduced into the FasterNet Block structure. ③ To improve the feature extraction capability of FasterNet Block, an efficient multi-scale attention (EMA) Module was incorporated to capture both global and local feature maps more effectively. To expand the receptive field and capture more effective and comprehensive contextual information for richer feature representation, an improved high-level screening feature fusion pyramid network (HS-FPN) was adopted to optimize multi-scale feature fusion. Experimental results showed that compared to the baseline model, the improved RT-DETR reduced parameters and FLOPs by 8.4×10⁶ and 17.8 G, respectively. The mAP@0.5 reached 94.5%, with a severe deviation detection accuracy of 99.2% and a detection speed of 41.0 frame per second, outperforming TOOD and ATSS object detection models, meeting the real-time and accuracy requirements of coal mine production.
- conveyor belt deviation /
- target detection /
- real-time detection transformer(RT-DETR) /
- structural reparameterization /
- efficient multi-scale attention mechanism /
- multi-scale feature fusion

HTML全文

图 1 RT−DETR模型结构

Figure 1. RT-DETR model structure

下载: 全尺寸图片幻灯片

图 2 FRE Block结构

Figure 2. FRE Block structure

下载: 全尺寸图片幻灯片

图 3 PConv的工作原理

Figure 3. Working principle of PConv

下载: 全尺寸图片幻灯片

图 4 EMA整体结构

Figure 4. Overall structure of EMA

下载: 全尺寸图片幻灯片

图 5 CAA−HSFPN结构

Figure 5. CAA-HSFPN structure

下载: 全尺寸图片幻灯片

图 6 测量方法

Figure 6. Measurement method

下载: 全尺寸图片幻灯片

图 7 数据集中各类别图像数量

Figure 7. Number of images in each category in dataset

下载: 全尺寸图片幻灯片

图 8 输送带正常运行时的检测结果对比

Figure 8. Comparison of detection results during normal operation of conveyor belt

下载: 全尺寸图片幻灯片

图 9 输送带向右轻微跑偏时的检测结果对比

Figure 9. Comparison of detection results when the conveyor belt deviates slightly to the right

下载: 全尺寸图片幻灯片

图 10 输送带向右严重跑偏时的检测结果对比

Figure 10. Comparison of detection results when the conveyor belt deviates significantly to the right

下载: 全尺寸图片幻灯片

图 11 输送带向左轻微跑偏时的检测结果对比

Figure 11. Comparison of detection results when the conveyor belt deviates slightly to the left

下载: 全尺寸图片幻灯片

图 12 输送带向左严重跑偏时的检测结果对比

Figure 12. Comparison of detection results when the conveyor belt deviates significantly to the left

下载: 全尺寸图片幻灯片

表 1 网络训练超参数

Table 1 Network training hyperparameters

参数名称	参数设置	参数名称	参数设置
训练次数	300	学习率动量	0.937
批次大小	4	对象查询	300
初始学习率	0.01	解码器	4

下载: 导出CSV

表 2 FRE Block消融实验结果

Table 2 Results of FRE Block ablation experiments

FasterNet Block	EMA	重参数化	参数量/10⁶ 个	FLOPs/G	mAP@0.5/%				模型体积/MiB	帧率/ （帧·s⁻¹）
FasterNet Block	EMA	重参数化	参数量/10⁶ 个	FLOPs/G	正常	轻微跑偏	严重跑偏	均值	模型体积/MiB	帧率/ （帧·s⁻¹）
×	×	×	31.1	88.8	97.4	80.5	98.0	92.0	61.5	52.2
√	×	×	24.4	72.3	98.5	81.4	98.6	92.8	48.5	52.1
√	×	√	24.4	72.3	98.2	83.5	98.8	93.5	48.7	51.3
√	√	×	24.6	76.1	97.7	80.5	99.0	92.4	48.9	43.3
√	√	√	24.6	76.1	98.8	82.9	99.1	93.6	49.1	41.9

下载: 导出CSV

表 3 CAA−HSFPN对比实验结果

Table 3 Comparative experimental results of CAA-HSFPN

特征融合机制	参数量/10⁶ 个	FLOPs/G	mAP@0.5/%				模型体积/MiB	帧率/ （帧·s⁻¹）
特征融合机制	参数量/10⁶ 个	FLOPs/G	正常	轻微跑偏	严重跑偏	均值	模型体积/MiB	帧率/ （帧·s⁻¹）
CCFM	31.1	88.8	97.4	80.5	98.0	92.0	61.5	52.2
HSFPN	28.2	83.5	97.6	81.8	98.9	92.8	55.6	53.2
BiFPN	30.4	94.5	97.1	80.4	98.7	92.1	60.3	50.3
CAA-HSFPN	28.7	86.7	98.4	83.3	98.9	93.5	56.5	48.9

下载: 导出CSV

表 4 改进RT−DETR消融实验结果

Table 4 Results of improved RT-DETR ablation experiments

FRE− Block	CAA− HSFPN	参数量/10⁶ 个	FLOPs/G	mAP@0.5/%				模型体积/MiB	帧率/ （帧·s⁻¹）
FRE− Block	CAA− HSFPN	参数量/10⁶ 个	FLOPs/G	正常	轻微跑偏	严重跑偏	均值	模型体积/MiB	帧率/ （帧·s⁻¹）
×	×	31.1	88.8	97.4	80.5	98.0	92.0	61.5	52.2
√	×	24.6	76.1	98.8	82.9	99.1	93.6	49.1	41.9
×	√	28.7	86.7	98.4	83.3	98.9	93.5	56.5	48.9
√	√	22.7	71.0	98.3	86.0	99.2	94.5	44.0	41.0

下载: 导出CSV

表 5 输送带跑偏数据集上各模型实验结果

Table 5 Experimental results of each model on conveyor belt deviation data set

模型	参数量/10⁶ 个	FLOPs/G	mAP@0.5/%				模型体积/MiB	帧率/ （帧·s⁻¹）
模型	参数量/10⁶ 个	FLOPs/G	正常	轻微跑偏	严重跑偏	均值	模型体积/MiB	帧率/ （帧·s⁻¹）
TOOD	32.0	199	97.2	77.4	88.0	87.5	247.2	22.2
ATSS	38.9	110	97.1	78.8	84.9	86.9	298.3	11.4
Deformable DETR	40.1	193	92.6	59.0	87.5	79.9	486.0	16.2
Conditional DETR	43.4	101	90.6	60.6	86.8	79.3	508.9	23.5
YOLOv7	36.5	103.2	95.7	86.8	93.8	92.1	73.0	55.5
YOlOv8m	25.8	78.7	95.9	84.2	92.7	91.0	50.8	60.8
YOlOv9c	50.7	236.6	96.3	85.5	95.7	92.5	100.3	29.1
Faster−YOLOv7	22.7	35.6	95.1	86.5	94.9	92.2	45.0	93.9
SlimNeck−YOLOv7	31.4	90.4	94.5	85.6	93.3	91.1	61.6	56.4
GAM−YOLOv8	34.0	85.2	96.4	84.2	93.7	91.4	68.4	61.7
本文方法	22.7	71.0	98.3	86.0	99.2	94.5	44.0	41.0

下载: 导出CSV

参考文献(24)

[1]	CHU Qi,MENG Guoying,FAN Xun. Analysis of speed and belt deviation of the conveyor belt[J]. Advanced Materials Research,2011,339:444-447.
[2]	徐世昌,程刚,袁敦鹏,等. 基于三维点云的带式输送机跑偏及堆煤监测方法[J]. 工矿自动化,2022,48(9):8-15,24. XU Shichang,CHENG Gang,YUAN Dunpeng,et al. Belt conveyor deviation and coal stacking monitoring method based on three-dimensional point cloud[J]. Journal of Mine Automation,2022,48(9):8-15,24.
[3]	ZHANG Mengchao,SHI Hao,YU Yan,et al. A computer vision based conveyor deviation detection system [J]. 2020,10(7). DOI:10.3390/ app10072402.
[4]	ZHANG Mengchao,JIANG Kai,CAO Yueshuai,et al. A deep learning-based method for deviation status detection in intelligent conveyor belt system[J]. Journal of Cleaner Production,2022,363. DOI:10.1016/j. jclepro.2022.132575.
[5]	XU Xinchao,ZHAO Hanguang,FU Xiaotian,et al. Real-time belt deviation detection method based on depth edge feature and gradient constraint[J]. Sensors,2023,23(19). DOI: 10.3390/s23198208.
[6]	WU Xiangfan,WANG Chusen,TIAN Zuzhi,et al. Research on belt deviation fault detection technology of belt conveyors based on machine vision[J]. Machines,2023,11(12). DOI: 10.3390/machines11121039.
[7]	LIU Yi,MIAO Changyun,LI Xianguo,et al. Research on deviation detection of belt conveyor based on inspection robot and deep learning[EB/OL]. [2024-06-10]. https://onlinelibrary.wiley.com/doi/10.1155/2021/3734560.
[8]	ZHAO Yian,LYU Wenyu,XU Shangliang,et al. DETRs beat YOLOs on real-time object detection[EB/OL]. [2024-06-10]. https://arxiv.org/html/2304.08069v3.
[9]	CHEN Jierun,KAO S H,HE Hao,et al. Run,don't walk:chasing higher FLOPS for faster neural networks[EB/OL]. [2024-06-10]. https://arxiv.org/abs/2303.03667v3.
[10]	DING Xiaohan,ZHANG Xiangyu,MA Ningning,et al. RepVGG:making VGG-style ConvNets great again[EB/OL]. [2024-06-10]. https://arxiv.org/abs/2101.03697.
[11]	OUYANG Daliang,HE Su,ZHAN Guozhong,et al. Efficient multi-scale attention module with cross-spatial learning[C]. IEEE International Conference on Acoustics,Speech and Signal Processing,Rhodes Island,2023:776-780.
[12]	CHEN Yifei,ZHANG Chenyan,CHEN Ben,et al. Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J]. Computers in Biology and Medicine,2024,170. DOI: 10.1016/j.compbiomed.2024.107917.
[13]	CAI Xinhao,LAI Qiuxia,WANG Yuwei,et al. Poly kernel inception network for remote sensing detection[EB/OL]. [2024-06-10]. https://arxiv.org/abs/2403.06258v2.
[14]	TAN Mingxing,PANG Ruoming,LE Q V,et al. EfficientDet:scalable and efficient object detection[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Seattle,2020:10778-10787.
[15]	FENG Chengjian,ZHONG Yujie,GAO Yu,et al. TOOD:task-aligned one-stage object detection[C]. IEEE/CVF International Conference on Computer Vision,Montreal,2021:3490-3499.
[16]	ZHANG Shifeng,CHI Cheng,YAO Yongqiang,et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[EB/OL]. [2024-06-10]. https://arxiv.org/abs/1912.02424v4.
[17]	ZHU Xizhou,SU Weijie,LU Lewei,et al. Deformable DETR:deformable transformers for end-to-end object detection[EB/OL]. [2024-06-10]. https://arxiv.org/pdf/2010.04159v1.
[18]	MENG Depu,CHEN Xiaokang,FAN Zejia,et al. Conditional DETR for fast training convergence[C]. IEEE/CVF International Conference on Computer Vision,Montreal,2021:3651-3660.
[19]	WANG C Y,BOCHKOVSKIY A,LIAO H Y M. YOLOv7:trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Vancouver,2023:7464-7475.
[20]	REIS D,KUPEC J,HONG J,et al. Real-time flying object detection with YOLOv8[EB/OL]. [2024-06-10]. https://arxiv.org/abs/2305.09972.
[21]	WANG C Y,YEH I H,MARK LIAO H Y M. YOLOv9:learning what you want toLearn using programmable gradient information[C]. European Conference on Computer Vision,Milan,2024:1-21.
[22]	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.
[23]	冯恒健,韩李涛,张鹏飞,等. 基于改进YOLOv7的高效行人检测方法[J]. 计算机应用,2024,44(增刊1):290-296. FENG Hengjian,HAN Litao,ZHANG Pengfei,et al. Efficient Pedestrian Detection Method based on improved YOLOv7[J]. Journal of Computer Applications,2019,44(S1):290-296.
[24]	WANG Zhenyue,YUAN Guowu,ZHOU Hao,et al. Foreign-object detection in high-voltage transmission line based on improved YOLOv8m[J]. Applied Sciences,2023,13(23). DOI: 10.3390/app132312775.

施引文献(8)

期刊类型引用(7)

1.	林海飞，季鹏飞，孔祥国，李树刚，白杨，龙航，李柏，和递. 我国低渗煤层井下注气驱替增流抽采瓦斯技术进展及前景展望. 煤炭学报. 2023(02): 730-749 . 百度学术
2.	马熠坤，杨艳国，谯永刚，赵勇，邓存宝，赵金典. 近距离特厚夹矸煤层多向消突技术研究. 矿业研究与开发. 2023(07): 131-137 . 百度学术
3.	文建东. 特厚煤层综采面瓦斯涌出规律及优化治理技术研究. 中国煤炭. 2023(11): 49-55 . 百度学术
4.	薛伟超. 顶板定向瓦斯抽采钻孔合理层位确定及其抽采效果分析. 中国煤炭. 2022(04): 41-47 . 百度学术
5.	张剑钊，陶云奇. 煤矿井下定向钻孔防治瓦斯技术研究进展. 能源与环保. 2022(04): 276-280+298 . 百度学术
6.	闫保永，孟祥辉. ?200 mm定向钻孔一次成孔技术与装备的研究. 矿业安全与环保. 2022(05): 49-52+58 . 百度学术
7.	谢明慧. 利用低位抽采巷穿层治理煤巷条带瓦斯技术应用. 煤炭与化工. 2021(08): 99-100 . 百度学术