Research on VOC leakage area identification based on target detection
-
摘要: 为了解决红外气体成像仪在挥发性有机化合物(VOC)泄漏识别中存在的误识别率高、漏检率高、算法执行效率低以及模型泛化能力差等问题，提出了一种基于运动特征增强的VOC泄漏区识别方法。采用视频序列投影变化率统计的方法确定视频稳定性判定阈值，提取稳定状态下运动背景和运动前景；采用优化线性拉伸的方法对运动前景进行特征增强和异常值过滤；将运动前景与原始帧进行图像融合，并利用目标检测算法进行VOC泄漏区域识别；通过模型预训练和迁移学习的方法，以烟雾数据集和少量VOC泄漏数据集进行了识别模型训练，并将模型迁移至RK3588S嵌入式开发板上进行了执行效率测试。结果表明，该算法在交并比为0.5的情况下，平均精度均值为0.88；交并比在0.5~0.95范围内，平均精度均值为0.51，单帧平均识别时间为49 ms，具有较高的识别精度和识别效率，能够满足实时监测需求。本文中的算法能够保持稳定的模型性能且具有一定的抗干扰能力，为VOC泄漏识别提供了一定的参考。
-
关键词:
- 图像处理 /
- 泄漏区域识别 /
- 目标检测 /
- 挥发性有机化合物泄漏 /
- 红外气体成像仪
Abstract: To solve the problem of high misidentification rate, high missed detection rate, low algorithm execution efficiency, and poor model generalization ability in the recognition of volatile organic compound (VOC) leakage area of infrared gas imager, a VOC leakage area recognition method based on motion feature enhancement was proposed. The video stability threshold was determined by using the statistical method of projection change rate of video sequence, and the moving background and moving foreground were extracted under stable state. Optimized linear stretching was used to perform feature enhancement and outlier filtering on the moving foreground. The motion foreground was fused with the original frame, and VOC leakage area identification was performed using the target detection algorithm. Through the method of model pre-training and transfer learning, the smoke dataset and a small amount of VOC leakage dataset were used to train the recognition model, and the model was transferred to the RK3588S embedded development board for execution efficiency test. Experimental results show that the mean average precision of the proposed algorithm is 0.88 when the intersection over union ratio is 0.5, and the mean average precision is 0.51 when the intersection over union ratio ranges from 0.5 to 0.95. The average recognition time of a single frame is 49 ms, which has high recognition accuracy and recognition efficiency, and can meet the requirements of real-time monitoring. The algorithm in this article can maintain stable model performance and has certain anti-interference capabilities providing some reference for VOC leak identification. -
-
![]()
图 10 算法耗时逐帧统计结果
Figure 10. Algorithm execution time frame-by-frame statistical results
表 1 红外气体成像仪核心参数
Table 1 Core parameters of infrared gas imaging device
item technical indicators reolution 640 pixel×512 pixel pixel size 15 μm spectral response range 3.2 μm~3.5 μm noiseequivalent temperature difference 22 mK cooling method stirling refrigeration machine data type 8 bit/14 bit 
下载: 导出CSV
表 2 模型训练关键参数
Table 2 Key parameters of model training
item smoke recognition model VOC recognition model number of categories 1 1 anchor point [10 13 16 30 33 23]
[30 61 62 45 59 119]
[116 90 156 198 373 326][10 13 16 30 33 23]
[30 61 62 45 59 119]
[116 90 156 198 373 326]initial learning rate 0.01 0.01 batch size 20 10 epochs 10 100 number of samples 15000 323
下载: 导出CSV
表 3 算法关键指标统计结果
Table 3 Algorithm key performance indicators statistics results
model accuracy algorithm execution efficiency/(frame·ms-1) algorithm execution efficiency standard deviation/(frame·ms-1) frame difference method — 18 9 contour detection method 0.79 110 126 the algorithm in this article 0.88 (mAP@0.5) 49 22
下载: 导出CSV
-
[1] YAN M Q, ZHU H K, LUO H N, et al. Daily ex1posure to environmental volatile organic compounds triggers oxidative damage: Evidence from a large-scale survey in China[J]. Environmental Science & Technology, 2023, 57(49): 20501-20509.
[2] CHOI Y H, KIM H J, SOHN J R, et al. Occupational exposure to VOC and carbonyl compounds in beauty salons and health risks associated with it in South Korea[J]. Ecotoxicology and Environmental Safety, 2023, 256: 114837. DOI: 10.1016/j.ecoenv.2023.114837
[3] HUANGE Y X, D K, XIONG Y, et al. One-third of global population at cancer risk due to elevated volatile organic compounds levels[J/OL]. Research Square: (2023-09)[2024-01-11]. https://doi.org/10.21203/rs.3.rs-3320416/v1.
[4] NARANJO E, BALIGA S, BERNASCOLLE P F, et al. IR gas imaging in an industrial setting[J]. Proceedings of the SPIE, 2010, 7661: 7661K.
[5] RANGEL J, SCHMOLL R, KROLL A. Catadioptric stereo optical gas imaging system for scene flow computation of gas structures[J]. IEEE Sensors Journal, 2021, 21(5): 6811-6820. DOI: 10.1109/JSEN.2020.3042116
[6] HUSEIN A M, CALVIN, HALIM D, et al. Motion detect application with frame difference method on a surveillance camera[J]. Journal of Physics: Conference Series, 2019, 12309(1): 012017.
[7] WANG L M, TONG Zh, JI B, et al. Temporal difference networks for efficient action recognition[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE Press, 2021: 1895-1904. .
[8] XIN Y H, HOU J, DONG L M, et al. A self-adaptive optical flow method for the moving object detection in the video sequences[J]. Optik—International Journal for Light and Electron Optics, 2014, 125(19): 5690-5694. DOI: 10.1016/j.ijleo.2014.06.092
[9] WANG X, FENG L I, XIN L, et al. Moving targets detection for sa-tellite-based surveillance video[C]//IGARSS 2019-2019 IEEE International Geoscience and Remote Sensing Symposium. Yokohama, Japan: IEEE Press, 2019: 5492-5495.
[10] HE W, LI J X, QI Q, et al. SIM-MFR: Spatial interactions mechanisms based multi-feature representation for background modeling[J]. Journal of Visual Communication and Image Representation, 2022, 88: 103622. DOI: 10.1016/j.jvcir.2022.103622
[11] 王建平, 李俊山, 杨亚威, 等. 基于红外成像的乙烯气体泄漏检测[J]. 液晶与显示, 2014, 29(4): 623-628. WANG J P, LI J Sh, YANG Y W, et al. Ethylene gas leakage detection based on infrared imaging[J]. Journal of Liquid Crystals and Displays, 2014, 29(4): 623-628(in Chinese).
[12] 刘路民根, 张耀宗, 栾琳, et al. 一种基于形状的红外图像泄漏气体检测方法[J]. 应用光学, 2019, 40(3): 468-472. LIU L M G, ZHANG Y Z, LUAN L, et al. A shape-based infrared image gas leakage detection method[J]. Journal of Applied Optics, 2019, 40(3): 468-472(in Chinese).
[13] HONG Sh Zh, YING H, YU H W, et al. A VOC gas detection algorithm based on infrared thermal imaging[C]//2019 Chinese Control and Decision Conference (CCDC). Nanchang, China: IEEE Press, 2019: 329-334.
[14] BADAWI D, PAN H Y, CETIN S C C, et al. Computationally efficient spatio-temporal dynamic texture recognition for volatile organic compound (VOC) leakage detection in industrial plants[J]. IEEE Journal of Selected Topics in Signal Processing, 2020, 14(4): 676-687. DOI: 10.1109/JSTSP.2020.2976555
[15] XU Y, DONG J X, ZHANG B, et al. Background modeling methods in video analysis: A review and comparative evaluation[J]. CAAI Transactions on Intelligence Technology, 2016, 1(1): 43-60.
[16] ZHENG Y, FAN L Zh. Moving object detection based on running average background and temporal difference[C]//2010 IEEE International Conference on Intelligent Systems and Knowledge Engineering. Hangzhou, China: IEEE Press, 2010: 270-272.
[17] MEGHANA R K, CHITKARA Y, MOHANA A. Background-mo-delling techniques for foreground detection and tracking using Gaussian mixture model[C]//2019 3rd International Conference on Computing Methodologies and Communication (ICCMC). Erode, India: IEEE Press, 2019: 1129-1134.
[18] ZHU M Zh, WANGE H B. Fast detection of moving object based on improved frame-difference method[C]//2017 6th International Conference on Computer Science and Network Technology (ICCSNT). Dalian, China: IEEE Press, 2017: 299-303.
[19] 费宬, 康佳龙, 刘俊良, 等. 基于FPGA的短波红外图像灰度级拉伸算法实现[J]. 太赫兹科学与电子信息学报, 2022, 20(7): 713-717. FEI F, KANG J L, LIU J L, et al. Implementation of short-wave infrared image gray-level stretching algorithm based on FPGA[J]. Journal of Terahertz Science and Electronic Information, 2022, 20(7): 713-717(in Chinese).
[20] 周永康, 朱尤攀, 曾邦泽, 等. 宽动态红外图像增强算法综述[J]. 激光技术, 2018, 42(5): 718-726. DOI: 10.7510/jgjs.issn.1001-3806.2018.05.025 ZHOU Y K, ZHU Y P, ZENG B Z, et al. A review of wide dynamic range infrared image enhancement algorithms[J]. Laser Technology, 2018, 42(5): 718-726(in Chinese). DOI: 10.7510/jgjs.issn.1001-3806.2018.05.025
[21] 魏艳平. 线性变换与局部均衡融合的红外图像增强[J]. 激光技术, 2024, 48(5): 705-710. DOI: 10.7510/jgjs.issn.1001-3806.2024.05.014 WEI Y P. Infrared image enhancement using linear transformation and local equalization fusion[J]. Laser Technology, 2024, 48(5): 705-710(in Chinese). DOI: 10.7510/jgjs.issn.1001-3806.2024.05.014
[22] WANG Y C, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: Trainable bag-of-free bies sets new state-of-the-art for real-time object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, Canada: IEEE Press, 2023: 7464-7475.
[23] CHENG Y H, YIN J L, CHEN B H, et al. Smoke 100k: A database for smoke detection[C]//2019 IEEE 8th Global Conference on Consumer Electronics (GCCE). Osaka, Japan: IEEE Press, 2019: 596-597.
计量
- 文章访问数: 37
- HTML全文浏览量: 3
- PDF下载量: 4