Method of Detecting Substation Equipment in Infrared Images Based on Improved Gaussian Convolution Kernel
-
摘要: 在无锚点算法CenterNet模型的基础上,针对基于红外图像的目标检测算法检测精度低、耗时长的问题,给出了一种基于改进高斯卷积核的变电站设备红外图像检测方法,该目标检测方法模型网络结构精简,模型计算量较小。通过现场变电站巡检机器人设备收集数据样本,进行算法模型的训练及验证,实现红外图像变电站设备精准识别及定位。本文以变电站巡检机器人搭配红外热成像仪采集到的红外图像库为基础,用深度学习方法对数据集进行训练和测试,研究变电站红外图像的目标检测技术。通过深度学习技术判断设备中心点位实现目标分类和回归。实验结果表明,该方法提高了变电站目标检测方法的识别定位精度,为变电站设备红外图像智能检测提供了新的思路。Abstract: Slow and inaccurate target detection algorithms used to analyze infrared images are the focus of this study. An infrared image detection method is proposed for substation equipment using an improved Gaussian convolution kernel, which is based on the CenterNet algorithm without an anchor point. In brief, data samples were first collected using on-site substation inspection robot equipment, the algorithm model was trained and verified, and finally, accurate identification and positioning of infrared image substation equipment was achieved. Specifically, based on the infrared image library collected by the substation inspection robot and the infrared thermal imager, methods of deep learning were applied to train and test a model using the dataset, the target detection technology of substation infrared images was studied, and the equipment center was accurately judged through deep learning technology to achieve target classification and regression. The identification and positioning accuracy of the substation target detection were improved by adopting this proposed method, and it provides new ideas for the intelligent detection of infrared images for substation equipment.
-
表 1 红外图像数据集
Table 1. Infrared image data set
Classname Label name Picture numbers Arrester Arrester 123 Breaker Breaker 166 Current transformer Current transformer 137 Disconnector Disconnector 73 Electricreactor Electricreactor 150 Voltage transformer Voltage transformer 303 Aerial conductor Aerial conductor 86 Condenser Condenser 236 Main transformer Main transformer 224 Tubular busbar Tubular busbar 72 表 2 计算机硬件配置
Table 2. Computer hardware configuration
Name Type CPU Intel Core I7 9700K GPU Nvidia RTX 2080 Ti hard disk 4T SAS 7.2K*1 memory 512 G 表 3 红外图像数据集测试结果
Table 3. Test results of infrared image data set
bn Model mAP Epoch hm_loss wh_loss off_loss Loss DLA-34 before 0.685 200 1.2178887 3.385998 0.229195 1.646682 after 0.705 200 1.1233817 2.775918 0.212195 1.415212 Res-101 before 0.661 200 0.396270 3.179799 0.244151 1.138400 after 0.723 200 0.521400 1.897403 0.232067 1.043208 Res-18 before 0.463 200 0.451421 2.713203 0.262212 0.994953 after 0.582 200 0.813421 2.113203 0.256721 0.87198 表 4 针对变电站真实条件下的性能试验记录表
Table 4. Performance test record for substation
Type Picture numbers Target numbers Correct detection
numberAverage accuracy Miss raio Fallout ratio Total time Aerialconductor 12 16 12 0.750 0.250 0 1.080 Arrester 38 62 54 0.871 0.129 0 3.040 Breaker 45 125 114 0.912 0.024 0.064 4.562 Condenser 83 83 71 0.855 0.133 0.012 8.088 Currenttransformer 33 91 83 0.912 0.030 0.058 2.699 Disconnector 16 16 13 0.813 0.187 0 1.746 Electricreactor 34 65 56 0.862 0.138 0 2.919 Maintransformer 42 42 36 0.857 0.143 0 3.606 Tubular busbar 8 15 12 0.800 0.200 0 0.874 Voltagetransformer 76 146 133 0.911 0.048 0.041 6.519 -
[1] Junwei Hsieh, Yungtai Hsu, Hongyuan Mark Liao, et al. Video-based human movement analysis and its application to surveillance systems[J]. IEEE Transactions on Multimedia, 2008, 10(3): 372-384. doi: 10.1109/TMM.2008.917403 [2] 黄文清, 汪亚明, 周志宇. 计算机视觉技术在工业领域中的应用[J]. 浙江理工大学学报: 自然科学版, 2002, 19(2): 28-32. https://www.cnki.com.cn/Article/CJFDTOTAL-ZJSG200202005.htmHUANG Wenqing, WANG Yaming, ZHOU Zhiyu. Application of computer vision technology in the field of industry[J]. Journal of Zhejiang Institute of Science and Technology, 2002, 19(2): 28-32. https://www.cnki.com.cn/Article/CJFDTOTAL-ZJSG200202005.htm [3] Leonid Sigal, Michael J Black. Guest editorial: state of the art in image- and video-based human pose and motion estimation[J]. International Journal of Computer Vision, 2010, 87(1): 1-3. http://dl.acm.org/citation.cfm?id=1713483 [4] Yuki Kitahara, Seiji Takahashi, Noriyuki Kuramoto. Ion attachment mass spectrometry combined with infrared image furnace for thermal analysis: evolved gas analysis studies[J]. Analytical Chemistry, 2009, 81(8): 3155-3158. doi: 10.1021/ac802746d [5] 王勇, 梅生伟, 何光宇. 变电站一次设备数字化特征和实现[J]. 电力系统自动化, 2010, 34(13): 94-99. https://www.cnki.com.cn/Article/CJFDTOTAL-DLXT201013019.htmWANG Yong, MEI Shengwei, HE Guangyu. The characteristics and realization of digitization of primary equipment in substations[J]. Automation of Electric Power Systems, 2010, 34(13): 94-99. https://www.cnki.com.cn/Article/CJFDTOTAL-DLXT201013019.htm [6] Felzenszwalb P F, Girshick R B, Mcallester D, et al. Object detection with discriminatively trained part-based models[J]. IEEE Transactions on Software Engineering, 2010, 32(9): 1627-1645. http://ieeexplore.ieee.org/document/5255236 [7] 方路平, 何杭江, 周国民. 目标检测算法研究综述[J]. 计算机工程与应用, 2018, 54(13): 11-18. doi: 10.3778/j.issn.1002-8331.1804-0167FANG Luping, HE Hangjiang, ZHOU Guomin. Research overview of object detection methods[J]. Computer Engineering and Applications, 2018, 54(13): 11-18. doi: 10.3778/j.issn.1002-8331.1804-0167 [8] 阮激扬. 基于YOLO的目标检测算法设计与实现[D]. 北京: 北京邮电大学, 2019.RUAN Jiyang. Design and Implementation of Target Detection Algorithm Based on YOLO[D]. Beijing: Beijing University of Posts and Telecommunications, 2019. [9] 李鹏飞, 刘瑶, 李珣, 等. YOLO9000模型的车辆多目标视频检测系统研究[J]. 计算机测量与控制, 2019, 27(8): 21-24. https://www.cnki.com.cn/Article/CJFDTOTAL-JZCK201908006.htmLI Pengfei, LIU Yao, LI Xun, et al. A detection method of multi-target for vehicles based on YOLO9000 model[J]. Computer Measurement & Control, 2019, 27(8): 21-24. https://www.cnki.com.cn/Article/CJFDTOTAL-JZCK201908006.htm [10] 季航, 贾镕, 刘晓, 等. 一种基于YOLOv3的红外目标检测系统[J]. 电子设计工程, 2019, 27(22): 61-64. https://www.cnki.com.cn/Article/CJFDTOTAL-GWDZ201922013.htmJI Hang, JIA Rong, LIU Xiao, et al. An infrared target detection system based on YOLOv3[J]. Electronic Design Engineering, 2019, 27(22): 61-64. https://www.cnki.com.cn/Article/CJFDTOTAL-GWDZ201922013.htm [11] 栾浩, 王力, 姜敏, 等. 基于改进SSD的目标检测方法[J]. 软件, 2020, 41(1): 29-35. https://www.cnki.com.cn/Article/CJFDTOTAL-RJZZ202001008.htmLUAN Hao, WANG Li, JIANG Min, et al. Object detection method based on improved SSD[J]. Computer Engineering & Software, 2020, 41(1): 29-35. https://www.cnki.com.cn/Article/CJFDTOTAL-RJZZ202001008.htm [12] 谢奇芳, 姚国清, 张猛. 基于Faster R-CNN的高分辨率图像目标检测技术[J]. 国土资源遥感, 2019, 31(2): 38-43. https://www.cnki.com.cn/Article/CJFDTOTAL-GTYG201902006.htmXIE Qifang, YAO Guoqing, ZHANG Meng. Research on high resolution image object detection technology based on Faster R-CNN[J]. Remote Sensing for Land & Resources, 2019, 31(2): 38-43. https://www.cnki.com.cn/Article/CJFDTOTAL-GTYG201902006.htm [13] DUAN K, BAI S, XIE L, et al. CenterNet: keypoint triplets for object detection[C]//IEEE International Conference on Computer Vision, 2019: 6569-6578. [14] 杨海燕, 蒋新华, 聂作先. 基于并行卷积神经网络的人脸关键点定位方法研究[J]. 计算机应用研究, 2015, 32(8): 283-285. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ201508068.htmYANG Haiyan, JIANG Xinhua, NIE Zuoxian. Facial key points location based on parallel convolutional neural network[J]. Application Research of Computers, 2015, 32(8): 283-285. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ201508068.htm [15] 刘云鹏, 裴少通, 武建华, 等. 基于深度学习的输变电设备异常发热点红外图片目标检测方法[J]. 南方电网技术, 2019, 13(2): 27-33. https://www.cnki.com.cn/Article/CJFDTOTAL-NFDW201902006.htmLIU Yunpeng, PEI Shaotong, WU Jianhua, et al. Deep learning based target detection method for abnormal hot spots infrared images of transmission and transformation equipment[J]. Southern Power System Technology, 2019, 13(2): 27-33. https://www.cnki.com.cn/Article/CJFDTOTAL-NFDW201902006.htm [16] 林海波, 王浩, 张毅. 改进高斯核函数的人体姿态分析与识别[J]. 智能系统学报, 2015, 10(3): 436-441. https://www.cnki.com.cn/Article/CJFDTOTAL-ZNXT201503021.htmLIN Haibo, WANG Hao, ZHANG Yi. Human postures recognition based on the improved Gauss kernel function[J]. CAAL Transactions on Intelligent Systems, 2015, 10(3): 436-441. https://www.cnki.com.cn/Article/CJFDTOTAL-ZNXT201503021.htm [17] WU Jiajun, XUE Tianfan, Joseph J. Lim. Single image 3D interpreter network[C]// European Conference on Computer Vision, 2016: 1-18. [18] 莫邵文, 邓新蒲, 王帅, 等. 基于改进视觉背景提取的运动目标检测算法[J]. 光学学报, 2016(6): 196-205. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB201606025.htmMO Shaowen, DENG Xinpu, WANG Shuai, et al. Moving object detection algorithm based on improved visual background extractor[J]. Acta Optica Sinica, 2016(6): 196-205. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB201606025.htm [19] 朱伯伟, 庾农, 肖松. 红外极小目标检测算法研究[J]. 航空电子技术, 2011, 42(3): 5-11. https://www.cnki.com.cn/Article/CJFDTOTAL-HKDZ201103001.htmZHU Bowei, YU Nong, XIAO Song. Approaches on infrared point targets detection algorithm[J]. Avionics Technology, 2011, 42(3): 5-11. https://www.cnki.com.cn/Article/CJFDTOTAL-HKDZ201103001.htm