Research on detection method of Ringelmann emittance of exhaust based on improved YOLOv5s algorithm

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Oct.2024, Vol.47 No.10: 1341-1347, 1361

DOI: 10.3969/j.issn.1003-5060.2024.10.007

Authors

CHENG Shuo $ ^{1,2} $, WANG Huanqin $ ^{2} $, HU Juntao $ ^{1,3} $, XIA Wangjin $ ^{2} $, YU Fajun $ ^{2} $, FANG Yong $ ^{1,3} $

(1. Academy of Opto-electric Technology, Hefei University of Technology, Hefei 230009, China; 2. Institute of Intelligent Machines, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China; 3. Intelligent Manufacturing Institute, Hefei University of Technology, Hefei 230051, China)

Abstract and Keywords

Abstract: For the problems of low accuracy and poor environmental adaptability of traditional detection methods of Ringelmann emittance of exhaust, this paper proposes a detection method of Ringelmann emittance of exhaust based on improved YOLOv5s algorithm. Considering the variable shape and complex background of exhaust, adaptively spatial feature fusion (ASFF) and global attention mechanism (GAM) are added to the existing YOLOv5s network to improve the detection accuracy of exhaust targets. At the same time, in order to reduce the impact of environmental factors such as illumination on exhaust target detection, based on the high temperature characteristics of exhaust, infrared images are used to improve the accuracy of exhaust region detection. Based on the standard Ringelmann emittance, the Ringelmann Emittance level of exhaust in the detected area is determined. The experimental results show that the detection accuracy of the improved YOLOv5s is as high as 95.3%, which is 3.4% higher than that of the existing YOLOv5s; the influence of illumination and other environmental factors on the detection results of exhaust targets is reduced, and the robustness of the algorithm is improved; the final determination accuracy of Ringelmann emittance of exhaust can reach level 0.5,

Keywords: Ringelmann emittance; vehicle exhaust target detection; Ringelmann emittance level determination; infrared image; YOLOv5s algorithm

作者信息

程硕 $ ^{1,2} $，王焕钦 $ ^{2} $，胡俊涛 $ ^{1,3} $，夏王进 $ ^{2} $，虞发军 $ ^{2} $，方勇 $ ^{1,3} $

(1. 合肥工业大学光电技术研究院, 安徽合肥 230009; 2. 中国科学院合肥物质科学研究院智能机械研究所, 安徽合肥 230031; 3. 合肥工业大学智能制造技术研究院, 安徽合肥 230051)

摘要和关键词

摘要: 针对传统尾气黑度测量方法精度低、环境适应性差等问题, 文章提出一种基于改进 YOLOv5s 算法的尾气黑度测量方法。考虑到尾气形状多变、背景复杂, 在现有 YOLOv5s 网络中添加自适应特征融合 (adaptively spatial feature fusion, ASFF) 和全局注意力机制 (global attention mechanism, GAM), 提高尾气目标的检测准确度; 同时, 为减少光照等环境因素对尾气目标检测的影响, 基于尾气的高温特性, 利用红外图像提高尾气区域检测准确度; 并基于标准的林格曼黑度对被检测区域内的尾气黑度进行等级判定。实验结果表明: 改进后的 YOLOv5s 对红外尾气目标的检测准确率高达 95.3%, 比现有 YOLOv5s 检测准确度提高了 3.4%; 同时还降低了光照等环境因素对尾气目标检测结果的影响, 改善了算法的鲁棒性; 最终尾气黑度判定精度达到 0.5 级, 可有效满足现有移动源尾气黑度高精度检测需求。

合肥工业大学学报(自然科学版)

Navigation