在变电站设备的热故障诊断过程中, 红外热成像检测技术是用于判断电力设备关键部位热状态的有效方法。依靠人工巡检的红外热成像检测技术存在诸多局限已不适用于逐渐信息化、智能化的现代变电站，探索更优性能的检测方法成为研究重点。基于此，本文对基于特征联合与双流互助的变电站设备红外热故障检测方法开展研究。论文主要研究内容如下：
（1） 本文设计“分割+ 检测”的红外热故障检测研究技术路线。在实例分割方面，针对现有的红外图像分割算法较少明确利用目标尺度与检测精度之间的对应关系问题，首先，本文采集西北某变电站红外图像构建数据集，并将其标准化为当前主流数据格式供模型验证；其次，构建基于非对称卷积块的骨干网络、面向精确区域的多尺度候选框提取颈部网络及基于动态mask 的预测头网络，实现了模型从骨干网络高效化、颈部网络分级化，到头部网络动态化的分割技术升级；最后，选取评价指标，对本文各子模块进行消融实验，对改进模型与主流模型进行对比实验。数据表明，在分割的精确率、准确率上本文模型均为最高（分别为84.5% 和83.9%），浮点计算值最低（58.3G）。
（2） 分割模型为热故障检测提供了良好的像素信息，但仍缺少图像语义信息。针对上述问题，本文围绕检测构建面向红外图像热故障检测和实例分割的双流互助模型，首先，目标检测流本文设计基于无卷积步长或池化层的YOLOv5 模型，并通过由空间到深度卷积（space-to-depth convolution, SPD-Conv）模块增加了目标检测精度；其次，实例分割流本文对基于多级特征联合提取的实例分割网络进行优化，采用轻量级实例分割模型作为辅助信模型，对目标检测进行信息增强；最后，设计相关实验进行模型验证，SPD-Conv 消融实验表明引入该模型给4 个版本的YOLOv5 均带来显著的检测精度提升（分别为10.1%，8.6%，4.1%，7.9%），模型对比实验表明本文模型在大中小目标的检测上均有最好表现，分别为53.1%，68.4%，71.5%。
  本文所提基于特征联合与双流互助的变电站设备红外热故障检测方法在自建数据集上取得了良好的检测效果，与同领域其他前沿模型相比具有显著性能优势和较少计算限制。上述模型的设计与验证为变电站电力设备热故障检测方法的工程现场应用提供了理论基础和实验数据。

  In the process of substation equipment thermal fault diagnosis, infrared thermal imagingdetection technology is an effective method to judge the thermal state of key parts of power equipment. The infrared thermal imaging detection technology which relies on manual inspection has many limitations and no longer suitable for the modern substation. The modern substation
is gradually informatization and intelligentize. The exploration of detection methods with better performance has become the research focus. Based on this, the infrared thermal fault detection
method of substation equipment based on feature association and double-current mutual assistance is studied in this paper. The main research contents of this paper are as follows:
  (1) In this paper, the research technical route of infrared thermal fault detection “segmentation+detection” is designed. In terms of segmentation, the existing infrared image segmentation
algorithms rarely make clear use of the corresponding relationship between object scale and detection accuracy. Firstly, this paper collects infrared images from a substation in Northwest of China to build a data set, and standardizes it into the current mainstream data format for model verification. Secondly, the backbone network based on asymmetric convolutional blocks, the multi-scale candidate box extraction neck network oriented to precise region, and the prediction head network based on dynamic mask are constructed. The segmentation technology of the model is upgraded from the high efficiency of the backbone network and the classification of the neck network to the dynamic head network. Finally, the evaluation indexes are selected to evaluate experiments, which include ablation experiments on each submodule and the improved model are compared with the mainstream model. The data show that the precision and accuracy of segmentation are the highest (84.5% and 83.9% respectively), and the floating point calculation value is the lowest (58.3G).
  (2) Segmentation model provides good pixel information for thermal fault detection, but still lacks image semantic information. To solve the above problems, this paper builds a doubleflow
mutual aid model for infrared image thermal fault detection and instance segmentation around detection. First, the YOLOv5 model based on no strided convolutional or pooling layer is designed for object detection flow, and the object detection accuracy is increased by spaceto-depth convolution module. Secondly, this paper optimizes the segmentation network based on multi-stage feature joint extraction, and uses lightweight segmentation model as auxiliary information model to enhance the information of object detection. Finally, relevant experiments are designed to verify the model. SPD-Conv ablation experiments show that the introduction of this model significantly improved the detection average precision of the four versions of YOLOv5 (10.1%, 8.6%, 4.1%, 7.9%, respectively). Model comparison experiments show that the proposed model has the best performance in detecting large, medium and small objects.They are 53.1%, 68.4% and 71.5% respectively.
  The infrared thermal fault detection method of substation equipment based on feature combination and double-stream mutual assistance proposed in this paper has achieved good detection results on the self-built data set. It has significant performance advantages and less computational limitations compared with other mainstream models in same field. The design and verification of the above model provide theoretical basis and experimental data for the engineering field application of thermal fault detection method of power equipment in substation.

[1] 肖峻, 张寒, 张宝强, 等. 配电网安全域的维度[J]. 电力系统自动化, 2018, 42(1):16–22.

[2] 单葆国, 刘青, 张莉莉, 等. 新形势下“十四五”后三年中国电力需求形势研判[J].中国电力, 2023, 56(3): 1–11.

[3] 国家能源局. 2021 年全国电力可靠性年度报告[R]. 北京: 国家能源局, 2021.

[4] 陈尧. 红外热成像技术在变电站设备故障诊断中的应用研究[D]. 山东: 山东理工大学, 2021.

[5] 彭明智, 樊培培, 胡永波, 等. 变电站机器人沿轨道自主巡检的方法研究[J]. 中国工程机械学报, 2023, 21(1): 50–55.

[6] 彭向阳, 金亮, 王锐, 等. 变电站机器人智能巡检技术及应用效果[J]. 高压电器, 2019,55(4): 223–232.

[7] Kateb B, Yamamoto V, Yu C, et al. Infrared thermal imaging: a review of the literatureand case report[J]. NeuroImage, 2009, 47: 154–162.

[8] 叶婕. 基于图像分割的变电站电气设备故障自动检测[D]. 西安: 西安科技大学,2012.

[9] 李婷. 变电站高压设备红外图像分割方法[D]. 北京: 华北电力大学, 2015.

[10] 谭靖. 膜计算在变电站电力设备红外图像分割与故障检测中的应用[D]. 成都: 西华大学, 2016.

[11] 张锦文. 变电站电气设备红外图像分割方法研究[D]. 北京: 华北电力大学, 2018.

[12] 田志远. 基于多目标进化聚类红外图像分割的变电站设备故障分析[D]. 武汉: 武汉理工大学, 2019.

[13] Wang F, Guo Y, Li C, et al. Electrical thermal image semantic segmentation: Large-scaledataset and baseline[J]. IEEE Transactions on Instrumentation and Measurement, 2022,71: 1–13.

[14] 刘赫, 赵天成, 刘俊博, 等. 基于深度残差UNet 网络的电气设备红外图像分割方法[J]. Infrared Technology, 2022, 44(12): 1351–1357.

[15] 韦宝泉. 少样本条件下输电线路电气设备图像检测与缺陷识别方法研究[D]. 南昌:华东交通大学, 2022.

[16] 吴迎昌. 基于红外热像技术的电接触故障检测方法研究[D]. 长沙: 湖南大学, 2015.

[17] Xia C, Ren M, Wang B, et al. Infrared thermography-based diagnostics on power equipment:State-of-the-art[J]. High Voltage, 2021, 6(3): 387–407.

[18] Zheng Y, Sun B. Power Cable Terminal Fault Detection Method Based on Infrared Thermal Imaging[J]. Journal of Physics: Conference Series, 2023, 2503(1): 1–6.

[19] Chen S, Luo Y, Yin J, et al. Application of visible light-infrared image fusion technology in power system fault detection[C]//Proceedings of the 2023 Asia Conference on Computer Vision, Image Processing and Pattern Recognition(CVIPPR 2023). Phuket: IEEE, 2023:1–6.

[20] Gao S, Ruan Y, Hong Q, et al. Infrared thermal image fault detection based on YOLOV3-L[C]//2022 IEEE International Conference on Artificial Intelligence and Computer Applications(ICAICA 2022). Dalian: IEEE, 2022: 175–178.

[21] Buffa S, Fouladfar MH, Franchini G, et al. Advanced control and fault detection strategies for district heating and cooling systems—A review[J]. Applied Sciences, 2021, 11(1):455–462.

[22] Zhang Z, Li J, Song Y, et al. A Novel Ultrasound-Vibration Composite Sensor for Defects Detection of Electrical Equipment[J]. IEEE Transactions on Power Delivery, 2022, 37(5):4477–4480.

[23] 赵丽杰, 张俊杰, 张海蛟, 等. 超高压交流变压器高压套管均压球结构关键技术研究[J]. 高压电器, 2016, 52(5): 126–129.

[24] 陈启发, 赵正铨, 王志胜. 无间隙氧化锌避雷器选择手册[M].[S.l.]: 北京: 中国电力出版社, 1999.

[25] 何嘉希, 高文胜, 刘卫东, 等. 六氟化硫断路器灭弧室压力测量导管结构研究[J]. 中国电机工程学报, 2016, 36(1): 310–316.

[26] 李文军, 孙坚, 邹超, 等. 黑体空腔辐射理论和设计[J]. 计量学报, 2011, 32(2):137–140.

[27] 常亮, 邓小明, 周明全, 等. 图像理解中的卷积神经网络[J]. 自动化学报, 2016, 42(9):1300–1312.

[28] 司念文, 张文林, 屈丹, 等. 卷积神经网络表征可视化研究综述[J]. 自动化学报, 2022,48(8): 1890–1920.

[29] 谢婷婷. 基于卷积神经网络的两阶段图像分类研究[D]. 成都: 电子科技大学, 2022.

[30] 胡斌, 陈淼, 姚志军. 信息解码视角下的脑科学研究现状与未来展望[J]. 兰州大学学报：医学版, 2023, 49(4): 1–5.

[31] 乔玲玲. 图像分割算法研究及实现[D]. 武汉: 武汉理工大学, 2009.

[32] 赵泉华, 高郡, 李玉. 基于区域划分的多特征纹理图像分割[J]. 仪器仪表学报, 2015,36(11): 2519–2530.

[33] Achanta R, Shaji A, Smith K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE transactions on pattern analysis and machine intelligence, 2012,34(11): 2274–2282.

[34] Chen J, Li Z, Huang B. Linear spectral clustering superpixel[J]. IEEE Transactions on image processing, 2017, 26(7): 3317–3330.

[35] Achanta R, Susstrunk S. Superpixels and polygons using simple non-iterative clustering[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR 2017). Honolulu: IEEE, 2017: 4651–4660.

[36] 陈如俊, 普运伟, 吴锋振, 等. 基于超像素分割与卷积神经网络的高光谱图像分类[J].激光与光电子学进展, 2023, 60(1610010): 1–9.

[37] 冯磊. 联合多层次特征的显著性检测算法研究及在图像分割中的应用[D]. 秦皇岛:燕山大学, 2022.

[38] 赵蓉. 基于深度学习的图像分割算法应用研究[D]. 南京: 南京邮电大学, 2022.

[39] Viola P, Jones M. Rapid object detection using a boosted cascade of simple features[C]//Proceedings of the 2001 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2001). Kauai: IEEE, 2001: 8–14.

[40] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//Proceedings of the 2005 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2005).San Diego: IEEE, 2005: 886–893.

[41] Felzenszwalb P, McAllester D, Ramanan D. A discriminatively trained, multiscale, deformable part model[C]//Proceedings of the 2008 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2008). Anchorage: IEEE, 2008: 1–8.

[42] Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2016). Las Vegas: IEEE, 2016: 779–788.

[43] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017). Honolulu:IEEE, 2017: 7263–7271.

[44] Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.

[45] Bochkovskiy A, Wang CY, Liao HYM. Yolov4: Optimal speed and accuracy of object detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[46] Zhu X, Lyu S, Wang X, et al. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proceedings of the 2021 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2021). Virtual:IEEE, 2021: 2778–2788.

[47] Li C, Li L, Jiang H, et al. YOLOv6: A single-stage object detection framework for industrial applications[J]. arXiv preprint arXiv:2209.02976, 2022.

[48] Wang CY, Bochkovskiy A, Liao HYM. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the 2023 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2023). Vancouver: IEEE, 2023:

7464–7475.

[49] Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision (ECCV 2016). Amsterdam: Springer,2016: 21–37.

[50] Fu CY, Liu W, Ranga A, et al. Dssd: Deconvolutional single shot detector[J]. arXiv preprint arXiv:1701.06659, 2017.

[51] Li Z, Zhou F. FSSD: feature fusion single shot multibox detector[J]. arXiv preprint arXiv:1712.00960, 2017.

[52] Jaeger PF, Kohl SA, Bickelhaupt S, et al. Retina U-Net: Embarrassingly simple exploitation of segmentation supervision for medical object detection[C]//Machine Learning for Health Workshop. .[S.l.]: [s.n.] , 2020: 171–183.

[53] Law H, Deng J. Cornernet: Detecting objects as paired keypoints[C]//Proceedings of the 15th European Conference on Computer Vision (ECCV 2018). Munich: Springer, 2018:734–750.

[54] Zhou X, Wang D, Krähenbühl P. Objects as points[J]. arXiv preprint arXiv:1904.07850,2019.

[55] Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers[C]//Proceedings of the 16th European Conference on Computer Vision (ECCV 2020).Edinburgh: Springer, 2020: 213–229.

[56] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2014). Columbus: IEEE, 2014: 580–587.

[57] He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence,2015, 37(9): 1904–1916.

[58] Girshick R. Fast r-cnn[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015). Boston: IEEE, 2015: 1440–1448.

[59] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137–1149.

[60] Lin TY, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2017). Honolulu: IEEE, 2017: 2117–2125.

[61] Cai Z, Fan Q, Feris RS, et al. A unified multi-scale deep convolutional neural network for fast object detection[C]//Proceedings of the 14th European Conference on Computer Vision (ECCV 2016). Amsterdam: Springer, 2016: 354–370.

[62] Abadi M, Agarwal A, Barham P, et al. Tensorflow: Large-scale machine learning on heterogeneous distributed systems[J]. arXiv preprint arXiv:1603.04467, 2016.