基于YOLOv5−SEDC模型的煤矸分割识别方法

杨洋; 李海雄; 胡淼龙; 郭秀才; 张会鹏

doi:10.13272/j.issn.1671-251x.2024010078

基于YOLOv5−SEDC模型的煤矸分割识别方法

doi: 10.13272/j.issn.1671-251x.2024010078

1.
靖远煤业工程勘察设计有限公司，甘肃白银　730913
2.
榆林市能源局，陕西榆林　719000
3.
浙江维思无线网络技术公司，浙江嘉兴　314001
4.
西安科技大学电气与控制工程学院，陕西西安　710054

基金项目: 陕西省秦创原“科学家+工程师”队伍建设项目(2022KXJ-38)；陕西省教育厅服务地方专项计划项目(23JC049)。

详细信息

作者简介:
杨洋（1987—），男，河南周口人，工程师，主要从事煤矿智能化技术研究与应用工作，E-mail：3050522067@qq.com

中图分类号: TD67/94
计量
- 文章访问数: 73
- HTML全文浏览量: 26
- PDF下载量: 7
- 被引次数: 0
出版历程
- 收稿日期: 2024-01-23
- 修回日期: 2024-08-13
- 网络出版日期: 2024-08-12

Coal and gangue segmentation and recognition method based on YOLOv5-SEDC model

1.
Jingyuan Coal Engineering Survey and Design Co., Ltd., Baiyin 730913, China
2.
Yulin Energy Bureau, Yulin 719000, China
3.
Wins Wireless Network Technology Ltd., Jiaxing 314001, China
4.
College of Electrical and Control Engineering, Xi'an University of Science and Technology, Xi'an 710054, China

摘要

摘要: 现有煤矸分割识别技术参数量大、分类速度慢和识别准确度不高；YOLOv5−seg模型在上下采样操作中易造成图像表面的纹理细节和灰度特征信息丢失，降低煤矸识别效率，且在训练过程中过分侧重全局特征，而忽略了对煤矸识别至关重要的局部显著区域和特征。针对上述问题，提出了一种基于YOLOv5−SEDC模型的煤矸分割识别方法。首先接收包含煤矸形状信息的图像，并利用主干网络进行特征提取，生成特征图；其次在YOLOv5−seg模型中集成SENet模块，以保留煤与矸石表面的纹理细节和灰度特征，避免下采样带来的信息丢失；然后采用不同，膨胀率的空洞卷积策略替代传统卷积核，不仅扩大了模型的感受野，还有效减少了模型参数量；最后分割检测头对融合后的特征进行精细处理，实现对煤矸的精确分割和识别。在大柳塔煤矿实际煤矸分选现场搭建煤矸图像采集实验平台，消融实验结果表明，YOLOv5−SEDC模型的煤和矸石识别的精确率较YOLOv5−seg模型平均提高1.3%，参数量减少0.7×10⁶个，检测速度提高了1.4 帧/s。对比实验结果表明：① YOLOv5−SEDC模型的精确率较YOLOv3−tiny，YOLOv5−seg，Mask−RCNN模型分别提高了10.7%，2.7%，1.9%，达到95.8%。② YOLOv5−SEDC模型的召回率较YOLOv3−tiny，YOLOv5−seg，Mask−RCNN模型分别提高了3.0%，2.1%，0.9%，达到89.1%。③ YOLOv5−SEDC模型的平均精度均值较YOLOv3−tiny，YOLOv5−seg，Mask−RCNN模型分别提高了6.4%，6.3%，1.8%，达到95.5%。④ YOLOv5−SEDC模型的F₁较YOLOv3−tiny，YOLOv5−seg，Mask−RCNN模型分别提高了5.2%，4.2%，2.1%，达到92.2%。⑤ YOLOv5−SEDC模型的检测速度较YOLOv3−tiny，YOLOv5−seg，Mask−RCNN模型分别降低了1.9，1.4，2.7 帧/s。可视化结果表明，YOLOv5−SEDC模型对煤和矸石的检测准确度较YOLOv5−seg和Mask−RCNN模型更高，说明了YOLOv5−SEDC模型在煤矸分割识别上具有较好性能。
- 煤矸分割 /
- 煤矸识别 /
- 压缩激励网络 /
- YOLOv5−SEDC /
- YOLOv5−seg /
- 注意力网络 /
- 空洞卷积
Abstract: The existing coal and gangue segmentation and recognition technology has a large number of parameters, slow classification speed, and low recognition accuracy. The YOLOv5-seg model is prone to losing texture details and grayscale feature information on the image surface during up and down sampling operations, which reduces the efficiency of coal and gangue recognition. The YOLOv5-seg model overly focuses on global features during training, while neglecting the locally significant regions and features that are crucial for coal and gangue recognition. In order to solve the above problems, a coal and gangue segmentation and recognition method based on YOLOv5-SEDC model is proposed. Firstly, the method receives an image containing the shape information of coal and gangue, and uses the backbone network for feature extraction to generate a feature map. The method integrates the SENet module into the YOLOv5-seg model to preserve the texture details and grayscale features of coal and gangue surfaces, avoiding information loss caused by down sampling. The method adopts a dilated convolution strategy with different dilation rates instead of traditional convolution kernels. It not only expands the receptive field of the model, but also effectively reduces the number of model parameters. Finally, the segmentation detection head finely processes the fused features to achieve precise segmentation and recognition of coal and gangue. A coal and gangue image acquisition experimental platform is established at the actual coal and gangue sorting site of Daliuta Coal Mine. The ablation experiment results show that the accuracy of coal and gangue recognition of YOLOv5-SEDC model is improved by an average of 1.3% compared to YOLOv5-seg model. The parameter quantity is reduced by 0.7×10⁶, and the detection speed is increased by 1.4 frames/s. The comparative experimental results show the following points. ① The accuracy of the YOLOv5-SEDC model is improved by 10.7%, 2.7%, 1.9% compared to the YOLOv3-tiny, YOLOv5-seg, and Mask-RCNN models, respectively, reaching 95.8%. ② The recall rate of the YOLOv5-SEDC model has increased by 3.0%, 2.1%, and 0.9% compared to the YOLOv3-tiny, YOLOv5-seg, and Mask-RCNN models, respectively, reaching 89.1%. ③ The mAP of the YOLOv5-SEDC model has increased by 6.4%, 6.3%, and 1.8% compared to the YOLOv3-tiny, YOLOv5-seg, and Mask-RCNN models, respectively, reaching 95.5%. ④ The F₁ value of the YOLOv5-SEDC model has increased by 5.2%, 4.2%, 2.1% compared to the YOLOv3-tiny, YOLOv5-seg, and Mask-RCNN models, respectively, reaching 92.2%. ⑤ The detection speed of the YOLOv5-SEDC model is reduced by 1.9, 1.4, and 2.7 frames/s compared to the YOLOv3-tiny, YOLOv5-seg, and Mask-RCNN models, respectively. The visualization results show that the YOLOv5-SEDC model has higher detection accuracy for coal and gangue than the YOLOv5-seg and Mask-RCNN models. It indicates that the YOLOv5-SEDC model has good performance in coal gangue segmentation and recognition.
- coal and gangue segmentation /
- coal and gangue recognition /
- compressed incentive network /
- YOLOv5-SEDC /
- YOLOv5-seg /
- attention network /
- dilated convolution

HTML全文

图 1 SENet网络结构

Figure 1. Structure of squeeze and excitation networks(SENet)

下载: 全尺寸图片幻灯片

图 2 DC效果

Figure 2. Effect of dilated convolutions(DC)

下载: 全尺寸图片幻灯片

图 3 YOLOv5−SEDC网络结构

Figure 3. Structure of YOLOv5-SEDC

下载: 全尺寸图片幻灯片

图 4 相机和光源的安装位置

Figure 4. Installation position of the camera and the light sources

下载: 全尺寸图片幻灯片

图 5 YOLOv5−SEDC模型评估指标曲线

Figure 5. Evaluation index curve of YOLOv5-SEDC model

下载: 全尺寸图片幻灯片

图 6 煤矸分割识别可视化效果

Figure 6. Visual effect of coal and gangue segmentation and recognition

下载: 全尺寸图片幻灯片

表 1 硬件环境配置

Table 1. Hardware environment configuration

硬件	参数
CPU	AMD Ryzen 75800 H
GPU	RTX3070
内存	16 GiB

下载: 导出CSV

表 2 消融实验结果

Table 2. Ablation experiment results

模型	精确率/%		参数量/10⁶个	检测速度/（帧·s⁻¹）
模型	煤	矸石	参数量/10⁶个	检测速度/（帧·s⁻¹）
YOLOv5−seg	92.9	93.5	7.2	3.8
YOLOv5−seg+SENet	94.3	93.8	7.3	3.6
YOLOv5−seg+DC	93.8	93.9	6.4	2.8
YOLOv5−SEDC	95.1	95.8	6.5	2.4

下载: 导出CSV

表 3 模型性能对比结果

Table 3. Models performance comparison results

	精确率/%	召回率/%	$ {\mathrm{mAP}}/\text{%} $	$F_1/ {\text{%}}$	检测速度/（帧·s⁻¹）
YOLOv3−tiny	85.1	86.1	89.1	87.0	4.3
YOLOv5−seg	93.1	87.0	89.2	88.0	3.8
Mask−RCNN	93.9	88.2	93.7	90.1	5.1
YOLOv5−SEDC	95.8	89.1	95.5	92.2	2.4

下载: 导出CSV

参考文献(23)

[1]	袁亮,张农,阚甲广,等. 我国绿色煤炭资源量概念、模型及预测[J]. 中国矿业大学学报,2018,47(1):1-8. YUAN Liang,ZHANG Nong,KAN Jiaguang,et al. The concept,model and reserve forecast of green coal resources in China[J]. Journal of China University of Mining & Technology,2018,47(1):1-8.
[2]	钱鸣高,许家林,王家臣. 再论煤炭的科学开采[J]. 煤炭学报,2018,43(1):1-13. QIAN Minggao,XU Jialin,WANG Jiachen. Further on the sustainable mining of coal[J]. Journal of China Coal Society,2018,43(1):1-13.
[3]	LI Jianping,DU Changlong,BAO Jianwei,et al. Direct-impact of sieving coal and gangue[J]. Mining Science and Technology,2010,20(4):611-614.
[4]	DUAN Chenlong,ZHOU Chenyang,DONG Liang,et al. A novel dry beneficiation technology for pyrite recovery from high sulfur gangue[J]. Journal of Cleaner Production,2018,172(3):2475-2484.
[5]	MOHANTA S,MEIKAP B C. Influence of mediumparticle size on the separation performance of an air dense medium fluidized bed separator for coal cleaning[J]. Journal of the South African Institute of Mining and Metallurgy,2015,115:761-766.
[6]	李思维,常博,刘昆轮,等. 煤炭干法分选的发展与挑战[J]. 洁净煤技术,2021,27(5):32-37. LI Siwei,CHANG Bo,LIU Kunlun,et al. Development and challenge of dry coal separation[J]. Clean Coal Technology,2021,27(5):32-37.
[7]	曹现刚,李莹,王鹏,等. 煤矸石识别方法研究现状与展望[J]. 工矿自动化,2020,46(1):38-43. CAO Xiangang,LI Ying,WANG Peng,et al. Research status of coal-gangue identification method and its prospect[J]. Industry and Mine Automation,2020,46(1):38-43.
[8]	MCCOY J T,AURET L. Machine learning applications in minerals processing:a review[J]. Minerals Engineering,2019,132:95-109.
[9]	LI Deyong,WANG Guofa ,ZHANG Yong,et al. Coal gangue detection and recognition algorithm based on deformable convolution YOLOv3[J]. IET Image Processing,2022,16(1):134-144.
[10]	SONG Qingjun,LIU Zhijiang,JIANG Haiyan. Coal gangue detection method based on improved YOLOv5[C]. International Conference on Big Data,Artificial Intelligence and Internet of Things Engineering,Xi'an,2022. DOI: 10.1109/ICBAIE56435.2022.9985920.
[11]	GUI Fangjun,YU Shuo,ZHANG Hailan,et al. Coal gangue recognition algorithm based on improved YOLOv5[C]. 2nd International Conference on Information Technology,Big Data and Artificial Intelligence,Chongqing,2021.DOI: 10.1109/ICIBA52610.2021.9687869.
[12]	FU Chengcai,LU Fengli,ZHANG Guoying. Gradient- enhanced waterpixels clustering for coal gangue[J]. International Journal of Coal Preparation and Utilization,2023,43(4):677-690.
[13]	LAI Wenhao,HU Feng,KONG Xixi,et al. The study of coal gangue segmentation for location and shape predicts based on multispectral and improved Mask R-CNN[J]. Powder Technology,2022,407. DOI: 10.1016/J.POWTEC.2022.117655.
[14]	LYU Ziqi,WANG Weidong,ZHANG Kanghui,et al. A synchronous detection-segmentation method for oversized gangue on a coal preparation plant based on multi-task learning[J]. Minerals Engineering,2022,187. DOI: 10.1016/J.MINENG.2022.107806.
[15]	TAGHANAKI S A,ABHISHEK K,COHEN J P,et al. Deep semantic segmentation of natural and medical images:a review[J]. Artificial Intelligence Review,2021,54(1):137-178.
[16]	HAO Shijie,ZHOU Yuan,GUO Yanrong. A brief survey on semantic segmentation with deep learning[J]. Neurocomputing,2020,406:302-321.
[17]	陈彪,卢兆林,代伟,等. 基于轻量化HPG−YOLOX−S模型的煤矸石图像精准识别[J]. 工矿自动化,2022,48(11):33-38. CHEN Biao,LU Zhaolin,DAI Wei,et al. Accurate recognition of coal-gangue image based on lightweight HPG-YOLOX-S model[J]. Journal of Mine Automation,2022,48(11):33-38.
[18]	郝俊峰,李玉涛,来博文. 基于YOLOv5−seg的多模型电石检测分割系统[J]. 现代计算机,2023,29(16):1-7,14. doi: 10.3969/j.issn.1007-1423.2023.16.001 HAO Junfeng,LI Yutao,LAI Bowen. Multi-model calcium carbide detection and segmentationsystem based on YOLOv5−seg[J]. Modern Computer,2023,29(16):1-7,14. doi: 10.3969/j.issn.1007-1423.2023.16.001
[19]	许灿辉,史操,陈以农. 基于膨胀卷积网络的端到端文档语义分割[J]. 中南大学学报(英文版),2021,28(6):1765-1774. XU Canhui,SHI Cao,CHEN Yinong. End-to-end dilated convolution network for document image semantic segmentation[J]. Journal of Central South University,2021,28(6):1765-1774.
[20]	YU Fisher,KOLTUN V. Multi-scale context aggregation by dilated convolutions[C]. International Conference on Learning Representation,Washington,2016. DOI: 10.48550/arXiv.1511.07122.
[21]	饶中钰,吴景涛,李明. 煤矸石图像分类方法[J]. 工矿自动化,2020,46(3):69-73. RAO Zhongyu,WU Jingtao,LI Ming. Coal-gangue image classification method[J]. Industry and Mine Automation,2020,46(3):69-73.
[22]	FERNANDO P G,RACHEL S,SEBASTIEN O. TorchIO:a Python library for efficient loading,preprocessing,augmentation and patch-based sampling of medical images in deep learning[J]. Computer Methods and Programs in Biomedicine,2021,208. DOI: 10.1016/J.CMPB.2021.106236.
[23]	赵杰,孙伟,徐中达,等. 基于形态学预处理的数字图像相关方法研究[J]. 实验力学,2022,37(5):629-637. ZHAO Jie,SUN Wei,XU Zhongda,et al. Study on the method of digital image correlation based morphological pre-processing[J]. Journal of Experimental Mechanics,2022,37(5):629-637.