随着太阳能光伏发电技术的不断推广和应用，光伏发电系统的安全运行问题也越来越受到重视。热斑故障不但降低了光伏阵列的输出功率，更是给光伏发电系统安全带来严重威胁。因此，设计一种高效、快速的热斑故障检测方法，是确保光伏系统安全工作的关键。为此，本文以无人机巡检为背景，结合计算机视觉技术，开展了基于深度学习框架和注意力模型的光伏阵列热斑故障检测方法。该研究对于提高光伏阵列巡检效率，保障光伏电站安全运行具有重要意义。论文主要工作如下：

（1）针对红外光伏阵列图像中热斑故障受环境复杂干扰，导致故障特征无法有效体现，进而影响检测精度的问题，提出一种基于自适应动态特征优化策略和CBAM注意力模型的光伏阵列热斑故障检测算法。首先，为解决复杂环境中热斑目标特征难以有效表达的问题，通过引导滤波四尺度分解机制及海洋捕食者优化策略在增强输入图像局部对比度的同时抑制背景信息；同时，构造了基于动态参数优化的目标函数，以获得综合指标最优的增强图像；然后，在基于YOLOv5的目标检测框架基础上嵌入了CBAM注意力模块以提高网络对热斑故障目标的检测精度。通过将提出的方法与基线网络进行对比，网络的精确率与召回率分别能够达到86.4%和79%。

（2）针对传统目标检测网络难以有效兼顾热斑故障检测性能与训练复杂度的问题，提出一种基于轻量化Ghost模块和协同注意力模型的光伏阵列热斑故障检测算法。首先，在YOLOv5网络中采用Ghost模块替换原始卷积层提升网络推理速度；其次，针对巡检图像因角度畸变和背景复杂而难以有效表达热斑目标特征的问题，通过嵌入协同注意力块来增强热斑故障区域的特征表达能力；此外，对检测网络进行参数剪枝及模块优化，在提升检测精度的同时进一步加快网络检测速度；最后，通过引入SIoU损失函数使检测效果进一步增强。实验证明该方法不仅有效降低了网络模型的参数量，使单张图片检测速度达到了13.7ms，且mAP@0.5达到83.3%，相比基线网络提高了6.8%。

With the continuous promotion and application of solar photovoltaic power generation technology, the safe operation of photovoltaic power generation systems is also receiving more and more attention. Hot-spot fault not only affects the output power of photovoltaic array, but also poses a serious threat to the safety of photovoltaic power generation system itself.  Therefore, designing an efficient and fast hot-spot fault detection method is the key to ensure the safe working of photovoltaic systems. To this end, this paper carries out a photovoltaic array hot-spot fault detection method based on deep learning framework and attention model in the context of UAV inspection combined with computer vision technology. This research is important for improving the inspection efficiency of photovoltaic arrays and ensuring the safe operation of photovoltaic plants. The main work of this paper is as follows:

(1) To address the problem of complex environmental interference in infrared photovoltaic array images, where the hot-spot fault features cannot be effectively reflected leading to the difficulty of accurate detection by detection networks, a hot-spot fault detection algorithm for photovoltaic arrays based on adaptive dynamic feature optimization strategy and CBAM attention module is proposed. First, in order to solve the problem that the target features of hot spots in complex environments are difficult to be expressed effectively, a guided filtering four-scale decomposition mechanism and a marine predator optimization strategy are used to suppress the background information while enhancing the local contrast of the input image. Meanwhile, an objective function based on dynamic parameter optimization is constructed to obtain the enhanced image with the best comprehensive index. Then, a CBAM attention module is embedded in the YOLOv5-based target detection framework to improve the network's detection accuracy for hot-spot fault targets. By comparing the proposed method with the baseline network, the network was able to achieve an accuracy and recall of 86.4% and 79%, respectively.

(2) To address the problem that traditional target detection networks are difficult to effectively balance hot-spot fault detection performance with training complexity, a photovoltaic array hot-spot fault detection algorithm based on lightweight Ghost module and coordinate attention model is proposed. Firstly, the convolutional layers in the backbone network are replaced by Ghost modules, which effectively improves the network inference speed. To address the problem that the inspection images are difficult to effectively express the hot-spot features due to the angular distortion and complex background, the feature representation of the hot-spot fault region is enhanced by embedding the coordinate attention block. In addition, parameter pruning and module optimization are performed for the detection network to reduce the generation of redundant features while improving the detection accuracy. Finally, the effectiveness of the detection model is further enhanced by introducing the SIoU loss function. It is demonstrated that this method not only effectively reduces the number of parameters of the network model, but also achieves a detection speed of 13.7ms for a single image, and the detection mAP@0.5 value of the algorithm can reach 83.3%, improved by 6.8% compared to baseline network.

[1]杨扬, 武文佳. 基于双光相机的光伏组件缺陷自动检测[J]. 新型工业化, 2019, 9(5): 56-60.

[2]赵万明. 双碳愿景下中国光伏发电发展的困境及破解之策[J]. 北方经济, 2022(9): 41-44.

[3]蒋琳, 苏建徽, 李欣, 等. 基于可见光和红外热图像融合的光伏阵列热斑检测方法[J].太阳能学报, 2022, 043(001): 393-397.

[4]Wei S, Li X, Ding S, et al. Hotspots Infrared detection of photovoltaic modules based on Hough line transformation and Faster-RCNN approach[C]// 2019 6th International Conference on Control, Decision and Information Technologies (CoDIT), 2019: 1266-1271.

[5]赵玲玲, 汪烨, 刘俊. 基于无人机与HSV空间的光伏电池板检测分析[J]. 红外技术, 2020, 42(10): 978-982.

[6]范月圆. 无人机智能巡检在风电光伏故障检测中的应用 [J]. 电子元器件与信息技术 , 2021, 5(2): 75-76.

[7]Zou Z, Hu Y, Gao B, et al. Temperature recovery from degenerated infrared image based on the principle for temperature measurement using infrared sensor[J]. Journal of Applied Physics, 2014, 115(4): 1247-1253.

[8]曹阳, 孟茁. 基于无人机光伏巡检的视频图像传输系统设计[J]. 计算机与网络, 2020, 46(15): 62-65.

[9]Livera A, Theristis M, Makrides G, et al. Recent advances in failure diagnosis techniques based on performance data analysis for grid-connected photovoltaic systems[J]. Renewable Energy, 2019, 133(APR.): 126-143.

[10]Hossam B, Itako K. Real time hotspot detection using scan-method adopted with P&O MPPT for PV generation system[C]// 2016 IEEE 2nd Annual Southern Power Electronics Conference (SPEC). IEEE, 2016.

[11]尹春杰, 宋彦螟, 肖发达,等. 基于数据扩充的光伏组件故障诊断方法[J]. 软件导刊, 2021, 20(10): 31-36.

[12]Wang Y, Kazutaka I, Tsugutomo K, et al. Voltage-Based Hot-Spot Detection Method for Photovoltaic String Using a Projector[J]. Energies, 2017, 10(2): 230.

[13]张杰, 易辉, 王远鸣,等. 基于局部保持投影的光伏组件热斑故障诊断[J]. 计算机应用与软件, 2021, 38(11): 49-55.

[14]Harrou F, Taghezouit B, Sun Y. Improved kNN-Based Monitoring Schemes for Detecting Faults in PV Systems[J]. IEEE Journal of Photovoltaics, 2019, 9(3): 811-821.

[15]乔苏朋, 杨艳, 陈世群, 等. 光伏阵列故障检测方法综述[J]. 电气技术, 2021, 22(7): 1-6.

[16]王兴, 魏克湘, 杨文献. 光伏板红外成像缺陷检测技术研究进展[J]. 湖南工程学院学报：自然科学版, 2021, 31(1): 32-40.

[17]车曦. 基于红外图像识别的光伏组件热斑故障检测方法研究[D]. 重庆大学.

[18]Tsanakas J A, Chrysostomou D, Botsaris P N, et al. Fault diagnosis of photovoltaic modules through image processing and Canny edge detection on field thermographic measurements[J]. International Journal of Solar Energy, 2015, 34(6): 351-372.

[19]Dotenco S, Dalsass M, Winkler L, et al. Automatic detection and analysis of photovoltaic modules in aerial infrared imagery[C]// Applications of Computer Vision. IEEE, 2016.

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

[21]Girshick R. Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV), 2015: 1440-1448.

[22]Ren S, He K, Girshick R, et al. Faster R-CNN: towards real-time object detection with regionproposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137-1149.

[23]Liu W, Anguelov D, Erhan D, et al. SSD: singleshot MultiBox detector[C]//Proceedings of the 14thEuropean Conference on Computer Vision, 2016: 21-37.

[24]Redmon S, Divvala R, Girshick R, et al. You Only Look Once: Unified, Real-Time Object Detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016: 779-788.

[25]Redmon J , Farhadi A . YOLO9000: Better, Faster, Stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017: 6517-6525.

[26]Redmon J, Farhadi A. YOLOv3:An incremental improvement [EB/OL].

[27]Bochkovskiy A , Wang C Y , Liao H . YOLOv4: Optimal Speed and Accuracy of Object Detection[J]. 2020.

[28]Wang C, Bochkovskiy A, Liao H. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2022.

[29]赖菲, 陈亚鹏, 单正涛,等. 深度学习算法在光伏电站无人机智能运维中应用[J]. 热力发电, 2019, 48(9): 139-144.

[30]Akram M W, Li G, Jin Y, et al. Automatic detection of photovoltaic module defects in infrared images with isolated and develop-model transfer deep learning[J]. Solar Energy, 2020, 198: 175-186.

[31]郭梦浩, 徐红伟. 基于Faster RCNN的红外热图像热斑缺陷检测研究[J]. 计算机系统应用, 2019, 28(11): 265-270.

[32]刘怀广, 丁晚成, 黄千稳. 基于轻量化卷积神经网络的光伏电池片缺陷检测方法研究[J]. 应用光学, 2021, 42: 1-8.

[33]郑发松, 叶昭龙, 王翔, 等. 基于多任务融合的光伏电站红外图像热斑检测[J].电力系统装备,2022,000(005): 164-169.

[34]吴湘宁, 贺鹏, 邓中港, 等. 一种基于注意力机制的小目标检测深度学习模型[J]. 计算机工程与科学, 2021, 43(01): 95-104.

[35]Jaderberg M , Simonyan K , Zisserman A , et al. Spatial Transformer Networks[J]. MIT Press, 2015.

[36]Hu J, Shen L, Albanie S, et al. Squeeze-and-Excitation Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(8): 2011-2023.

[37]Fu J, Liu J, Tian H, et al. Dual Attention Network for Scene Segmentation[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020: 3141-3149.

[38]Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision(ECCV), 2018: 3-19.

[39]曹城硕, 袁杰. 基于YOLO-Mask算法的口罩佩戴检测方法[J]. 激光与光电子学进展, 2021, 058(008): 203-210.

[40]郝帅, 马瑞泽, 赵新生, 等. 基于卷积块注意模型的YOLOv3输电线路故障检测方法[J]. 电网技术, 2021, 45(8): 2979-2987.

[41]邓杰, 万旺根. 基于改进YOLOv3的密集行人检测[J]. 电子测量技术, 2021, 44(11): 90-95.

[42]Ying Y, Qi Y, Rong L, et al. Anchor Points Based Accurate Fault Locating in Large-Scale Photovoltaic Plants via Aerial Infrared Videos[J]. IEEE Journal of Photovoltaics, 2021, 12(1): 437-443.

[43]孙海蓉, 李号. 基于深度迁移学习的小样本光伏热斑识别方法[J]. 太阳能学报, 2022, 043(001): 406-411.

[44]Liu L, OuyangW, Wang X, et al. Deep Learning for Generic Object Detection: A Survey[J]. International Journal of Computer Vision, 128 (2018): 261-318.

[45]Lee C, Xie S, Gallagher P, et al. Deeply-Supervised Nets[C]// Proceedings of AISTATS. 2015.

[46]Krizhevsky A, Sutskever I, Hinton G. ImageNet Classification with Deep Convolutional Neural Networks[J]. Communications of the Acm, 2012, 60(6): 84-90.

[47]Zou Z, Shi Z, Guo Y, et al. Object Detection in 20 Years: A Survey[J]. Proceedings of the IEEE, 111(3): 257-276.

[48]Neubeck A, Gool L. Efficient Non-Maximum Suppression[C]//18th International Conference on Pattern Recognition (ICPR'06), 2006: 850-855.

[49]Lin T, Dollar P, Girshick R, et al. Feature Pyramid Networks for Object Detection[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017: 936-944.

[50]Liu S, Qi L, Qin H, et al. Path Aggregation Network for Instance Segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 8759-8768.

[51]Guo M, Xu T, Liu J, et al. Attention mechanisms in computer vision: A survey[J]. Computational Visual Media, 2022, 8(3): 331-368.

[52]Wang Q, Wu B, Zhu P, et al. ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020: 11531-11539.

[53]Li X, Wang W, Hu X, et al. Selective kernel networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2019: 510-519.

[54]Zhang H, Wu C, Zhang Z, et al. ResNeSt: Split-Attention Networks[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2022: 2735-2745.

[55]He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016: 770-778.

[56]Wang F, Jiang M, Qian C, et al. Residual Attention Network for Image Classification[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017: 6450-6458.

[57]Hou Q, Zhou D, Feng J. Coordinate Attention for Efficient Mobile Network Design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021: 13713-13722.

[58]Howard A G, Zhu M, Chen B, et al. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications//2017 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017.

[59]Sandler M, Howard A, Zhu M, et al. Mobilenetv2: Inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2018: 4510-4520.

[60]Howard A, Sandler M, Chen B, et al. Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE, 2020: 1314-1324.

[61]Zhang X, Zhou X, Lin M, et al. ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). IEEE, 2018: 6848-6856.

[62]Ma N, Zhang X, Zheng H, et al. ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design[C]//Proceedings of the European Conference on Computer Vision(ECCV), 2018: 122-138.

[63]Han K, Wang Y, Tian Q, et al. GhostNet: More Features From Cheap Operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 1577-1586.

[64]He K, Sun J, and Tang X. Guided Image Filtering[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(6): 1397-1409.

[65]Bavirisetti D P, Xiao G, Zhao J, et al. Multi-scale Guided Image and Video Fusion: A Fast and Efficient Approach[J]. Circuits, systems, and signal processing: CSSP,2019,38(12): 5576-5605.

[66]Zhang L, Zhang L, Mou X, et al. FSIM: A Feature Similarity Index for Image Quality Assessment[J]. IEEE Transactions on Image Processing, 2011, 20(8): 2378-2386.

[67]Ma J, Ma Y, Li C. Infrared and visible image fusion methods and applications: A survey[J]. Information Fusion, 2019: 153-178.

[68]David A. Van Veldhuizen,Gary B. Lamont. Multiobjective Evolutionary Algorithms: Analyzing the State-of-the-Art[J].Evolutionary computation,2000,8(2): 125-147.

[69]Branke J, Deb K, Dierolf H, et al. Finding knees in multi-objective optimization[C]//8th International Conference on Parallel Problem Solving from Nature DBLP, 2004: 722-731.

[70]Marler R T, Arora J S. Survey of multi-objective optimization methods for engineering[J]. Structural & Multidisciplinary Optimization, 2004, 26(6): 369-395.

[71]郑新宇. 基于多目标优化模型的最优上车点位置研究[J]. 科学与信息化, 2020, 000(020): 25-26.

[72]陈锐智, 李析鸿, 陈思羽, 等. 基于EPSILON约束法的配电自动化设备多目标优化布点模型[J]. 电力系统保护与控制, 2021, 049(024): 51-58.

[73]Tseng C, Lu T. Minimax multiobjective optimization in structural design[J]. International Journal for Numerical Methods in Engineering, 2010, 30(6): 1213-1228.

[74]Schaffer J D. Multiple Objective Optimization with Vector Evaluated Genetic Algorithms[C]// Proceedings of the 1st International Conference on Genetic Algorithms, Lawrence Erlbaum Associates Publishers,1985: 93-100.

[75]Wang S, Chen T. Using NSGA-II to solve multi-objective competitive location problem with cooperative coverage for distance-based attractiveness[J]. Journal of Intelligent and Fuzzy Systems, 2021(153): 1-12.

[76]李婕, 杨淑英, 谢震,等. 基于有效信息迭代快速粒子群优化算法的永磁同步电机参数在线辨识[J]. 电工技术学报, 2022, 37(18): 4604-4613.

[77]王子琪, 张慧媛, 许军, 等. 基于改进人工蜂群算法的区域电网储能系统能量管理优化策略[J]. 中国电力, 2022, 55(9): 16-22.

[78]Faramarzi A, Heidarinejad M, Mirjalili S, et al. Marine Predators Algorithm: A Nature-inspired Metaheuristic[J]. Expert Systems with Applications, 2020, 152: 113377.

[79]Ma J, Ma Y, Li C. Infrared and visible image fusion methods and applications: A survey[J]. Information Fusion, 2019: 153-178.

[80]Cui G, Feng H, Xu Z, et al. Detail preserved fusion of visible and infrared images using regional saliency extraction and multi-scale image decomposition[J]. Optics Communications, 2015, 341: 199-209.

[81]Eskicioglu A M, Fisher P S. Image quality measures and their performance[J]. IEEE Trans Commun, 1995, 43(12): 2959-2965.

[82]Gevorgyan Z. SIoU Loss: More Powerful Learning for Bounding Box Regression[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022: 1-12.