Accurate recognition of coal-gangue image based on lightweight HPG-YOLOX-S model
-
摘要: 针对现有基于视觉技术的煤矸石分选方法存在模型参数量大、特征提取能力差、识别精度低等问题,提出了一种基于轻量化Ghost−S网络与混合并联注意力模块(HPAM)YOLOX−S模型(HPG−YOLOX−S模型)的煤矸石识别方法。首先,在YOLOX−S模型主干网络中加入HPAM,以增强图像中重要信息,抑制次要信息,加强主干网络的特征提取能力。其次,将YOLOX−S模型主干网络替换为参数量更小的Ghost−S网络,提高利用率与特征融合能力。最后,在预测层中采用SIOU损失函数来替换YOLOX−S模型的损失函数,提升检测与定位精度,加强对目标的提取能力。为验证所提方法对大块煤矸石的检测效果,将HPG−YOLOX−S模型与YOLOX−S模型进行对比,结果表明,HPG−YOLOX−S模型对煤与矸石的识别准确率分别为99.53%和99.60%,较YOLOX−S模型识别准确率分别提高了2.51%,1.27%。有效性验证结果表明, HPG−YOLOX−S模型的精确率、召回率和F1值均在94%以上,较YOLOX−S模型分别提高了5.68%,3.51%,2.91%; HPG−YOLOX−S模型的参数为7.8 MB,较YOLOX−S模型降低了1.2 MB。消融试验结果表明,HPG−YOLOX−S模型的平均精度均值较YOLOX−S模型提高了9.17%。热力图可视化试验结果表明,HPG−YOLOX−S模型关注煤与矸石的纹理和轮廓等表面差异,对煤矸石目标的全局关注度更加显著。
-
关键词:
- 煤矸石检测 /
- 图像识别 /
- 轻量化网络 /
- HPG−YOLOX−S /
- 混合并联注意力模块
Abstract: The existing coal-gangue separation methods based on vision technology have problems of large model parameter amount, poor feature extraction capability and low recognition precision. In order to solve the above problems, a coal-gangue recognition method based on YOLOX-S model combined lightweight Ghost-S network and hybrid parallel attention module (HPAM) named HPG-YOLOS-S model is proposed. Firstly, HPAM is added to the backbone network of YOLOX-S model. Thus the important information in an image is enhanced, and the secondary information is inhibited. The feature extraction capability of the backbone network is enhanced. Secondly, the backbone network of YOLOX-S model is replaced by Ghost-S network with smaller parameter quantity. The utilization rate and feature fusion capability are improved. Finally, in the predection layer, the SIOU loss function is used to replace the loss function of YOLOX-S model to impsrove the detection and positioning precision and enhance the extraction capability of the target. In order to verify the detection effect of the proposed method on large coal-gangue, the HPG-YOLOX-S model is compared with YOLOX-S model. The results show that the identification accuracy of the HPG-YOLOX-S model for coal and gangue is 99.53% and 99.60% respectively, which is 2.51% and 1.27% higher than those of YOLOX-S model. The results of validation show that the precision rate, recall rate and F1 value of the HPG-YOLOX-S model are all above 94%, which are 5.68%, 3.51% and 2.91% higher than those of YOLOX-S model respectively. The parameters amount of the HPG-YOLOX-S model is 7.8 MB, which is 1.2 MB lower than that of YOLOX-S model. The ablation experiment results show that the mean average precision of the HPG-YOLOX-S model is 9.17% higher than that of YOLOX-S model. The experiment result of visualization of the thermodynamic diagram shows that the HPG-YOLOX-S model focuses on the surface differences between coal and gangue, such as texture and contour. The model pays more attention to the overall target of coal-gangue. -
表 1 表1 对大块煤矸石的检测效果
Table 1. Detection effect of bulk coal-gangue
模型 类别 准确率/% YOLOX−S 煤 97.09 矸石 98.35 HPG−YOLOX−S 煤 99.53 矸石 99.60 表 2 模型性能参数对比结果
Table 2. Comparison results of the model performance parameters
检测模型 精确率/% 召回率/% F1值/% 参数量/MB YOLOX−S 91.6 91.3 93.1 9.0 YOLOX−M 88.2 82.5 94.9 12.3 YOLOX−L 89.5 85.1 86.5 14.5 HPG−YOLOX−S 96.8 94.5 95.8 7.8 表 3 消融试验结果
Table 3. Ablation test results
模型 模块3 mAP/% HPAM Ghost−S SIOU YOLOX−S × × × 90.5 √ × × 93.6 √ √ × 94.2 √ √ √ 98.8 -
[1] 谢和平,王金华,王国法,等. 煤炭革命新理念与煤炭科技发展构想[J]. 煤炭学报,2018,43(5):1187-1197. doi: 10.13225/j.cnki.jccs.2018.0517XIE Heping,WANG Jinhua,WANG Guofa,et al. New ideas of coal revolution and layout of coal science and technology development[J]. Journal of China Coal Society,2018,43(5):1187-1197. doi: 10.13225/j.cnki.jccs.2018.0517 [2] 王中举,毛馨凯,孙江. 基于视频解析的智能煤矸分选技术研究[J]. 工矿自动化,2021,47(增刊1):122-125.WANG Zhongju,MAO Xinkai,SUN Jiang. Research on intelligent coal-gangue separation technology based on video analysis[J]. Industry and Mine Automation,2021,47(S1):122-125. [3] 康利,黄金辉,刘畅. GDRT型 $ \text{γ} $ 射线智能干法分选系统在六家煤矿的应用[J]. 煤炭加工与综合利用,2017(3):22-34. doi: 10.16200/j.cnki.11-2627/td.2017.03.007KANG Li,HUANG Jinhui,LIU Chang. Application of GDRT$ \text{γ} $ -ray intelligent dry separation system in six coal mines[J]. Coal Processing & Comprehensive Utilization,2017(3):22-34. doi: 10.16200/j.cnki.11-2627/td.2017.03.007[4] 梁兴国. TDS智能干选机在井下排矸充填技术的应用[J]. 选煤技术,2020(2):30-34. doi: 10.16447/j.cnki.cpt.2020.02.008LIANG Xingguo. Application of the technique for underground destoning and stowing using TDS intelligent dry cleaning machine[J]. Coal Preparation Technology,2020(2):30-34. doi: 10.16447/j.cnki.cpt.2020.02.008 [5] 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. [6] MOHANTA S,MEIKAP B C. Influence of medium particle 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. [7] 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. doi: 10.13226/j.issn.1006-6772.21042555LI Siwei,CHANG Bo,LIU Kunlun,et al. Development and challenges of dry coal separation[J]. Clean Coal Technology,2021,27(5):32-37. doi: 10.13226/j.issn.1006-6772.21042555 [8] MCCOY J T,AURET L. Machine learning applications in minerals processing:A review[J]. Minerals Engineering,2019,132:95-109. doi: 10.1016/j.mineng.2018.12.004 [9] 何敏,王培培,蒋慧慧. 基于SVM和纹理的煤和煤矸石自动识别[J]. 计算机工程与设计,2012,33(3):1117-1121. doi: 10.3969/j.issn.1000-7024.2012.03.055HE Min,WANG Peipei,JIANG Huihui. Recognition of coal and stone based on SVM and texture[J]. Computer Engineering and Design,2012,33(3):1117-1121. doi: 10.3969/j.issn.1000-7024.2012.03.055 [10] 陈立,杜文华,曾志强,等. 基于小波变换的煤矸石自动分选方法[J]. 工矿自动化,2018,44(12):60-64. doi: 10.13272/j.issn.1671-251x.2018020047CHEN Li,DU Wenhua ZENG Zhiqiang,et al. Automatic separation method of coal and gangue based on wavelet transform[J]. Industry and Mine Automation,2018,44(12):60-64. doi: 10.13272/j.issn.1671-251x.2018020047 [11] 王家臣,李良晖,杨胜利. 不同照度下煤矸图像灰度及纹理特征提取的实验研究[J]. 煤炭学报,2018,43(11):3501-3061. doi: 10.13225/j.cnki.jccs.2018.0866WANG Jiachen,LI Lianghui,YANG Shengli. Experimental study on gray and texture features extraction of coal and gangue image under different illuminance[J]. Journal of China Coal Society,2018,43(11):3501-3061. doi: 10.13225/j.cnki.jccs.2018.0866 [12] 何江,张科星. 基于机器视觉的AlexNet网络煤矸石检测系统[J]. 煤炭技术,2022,41(3):205-208.HE Jiang,ZHANG Kexing. Coal gangue detection system based on machine vision of AlexNet net work[J]. Coal Technology,2022,41(3):205-208. [13] 徐志强,吕子奇,王卫东,等. 煤矸智能分选的机器视觉识别方法与优化[J]. 煤炭学报,2020,45(6):2207-2216. doi: 10.13225/j.cnki.jccs.zn20.0307XU Zhiqiang,LYU Ziqi,WANG Weidong,et al. Machine vision recognition method and optimization for intelligent separation of coal and gangue[J]. Journal of China Coal Society,2020,45(6):2207-2216. doi: 10.13225/j.cnki.jccs.zn20.0307 [14] 来文豪,周孟然,胡锋,等. 基于多光谱成像和改进YOLOv4的煤矸石检测[J]. 光学学报,2020,40(24):72-80.LAI Wenhao,ZHOU Mengran,HU Feng,et al. Coal gangue detection based on multi-spectral imaging and improved YOLOv4[J]. Acta Optica Sinica,2020,40(24):72-80. [15] LIN T Y,GOYAL P,GIRSHICK R,et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,42(2):318-327.