Insulator Defect Detection Based on Lightweight Improved RT-DETR Edge Deployment Algorithm
Jiang Xiangju1, Wang Ruitong1, Ma Yanhong2,3
1. School of Automation and Electrical Engineering Lanzhou Jiaotong University Lanzhou 730070 China; 2. School of Mechanical Engineering Lanzhou Jiaotong University Lanzhou 730070 China; 3. Gansu Province Special Equipment Inspection and Testing Institute Lanzhou 730050 China
Abstract：With the continuous advancement of the new generation power system construction and the continuous development of artificial intelligence technology, it has become an inevitable development trend to use intelligent methods to inspect the insulator condition of transmission lines. "Cloud-edge-end collaborative architecture" has the characteristics of cloud service center and edge computing equipment, and is suitable for insulator condition inspection, which can improve the inspection efficiency of insulators and avoid the risk of manual inspection. However, the computing power of edge devices in "cloud-edge-end collaborative architecture" is relatively poor. In order to facilitate its algorithm deployment, a lightweight rel-time detection transformer (RT-DETR) target detection algorithm is proposed to meet the requirements of edge devices for deploying lightweight algorithms, and at the same time, to ensure that the algorithm performance can meet the accuracy requirements of transmission line insulator condition inspection tasks. Firstly, RT-DETR is used as the baseline algorithm to reduce the difficulty of optimization and improve the robustness. Then, in order to reduce the parameters and calculation of the algorithm, the lightweight efficient model (EMO) is selected as the feature extraction backbone of the algorithm, and the long-distance feature interaction of insulator targets and the local feature interaction of defective small targets are fully learned. At the same time, a lightweight and efficient hybrid encoder is designed based on the lightweight attention-based intra-scale feature interaction module and lightweight cross-scale feature fusion module, which further lightens the algorithm. Finally, because the lightweight improvement will lead to a certain degree of algorithm performance degradation, the location information supplementary branch is introduced into the lightweight and efficient hybrid encoder to alleviate the problem of the loss of positioning information of defective small targets in deep features. Because the structure is lightweight, it will only bring a slight increase in the amount of algorithm parameters and calculation. DIoU loss function is used to measure the distance between the prediction frame and the real frame, which makes the regression process of the bounding box more stable. Combined with transfer learning skills, the loss of the algorithm decreases faster and the convergence is better. When constructing the insulator dataset, the pictures in it are processed to simulate rain, snow and fog weather. Using this dataset to train the algorithm can make it not be disturbed by these three kinds of weather to a certain extent, so that the algorithm can carry out the inspection of insulator condition in extreme cases. At the same time, the experimental results show that, compared with the baseline algorithm, the detection accuracy of the proposed algorithm reaches 97.2%, with a loss of only 0.7 percentage points, while the parameters and calculation amount decrease by 67.8% and 71.2% respectively, and the detection speed increases by 2.5 times, which can meet the requirements of the accuracy of the state inspection of transmission line insulators and the lightweight of edge deployment.
[1] 张血琴, 周志鹏, 郭裕钧, 等. 不同材质绝缘子污秽等级高光谱检测方法研究[J]. 电工技术学报, 2023, 38(7): 1946-1955. Zhang Xueqin, Zhou Zhipeng, Guo Yujun, et al.Detection method of contamination grades of insulators with different materials based on hyperspectral technique[J]. Transactions of China Electrotechnical Society, 2023, 38(7): 1946-1955. [2] 赵乐, 王先培, 姚鸿泰, 等. 基于可见光航拍图像的电力线提取算法综述[J]. 电网技术, 2021, 45(4): 1536-1546. Zhao Le, Wang Xianpei, Yao Hongtai, et al.Survey of power line extraction methods based on visible light aerial image[J]. Power System Technology, 2021, 45(4): 1536-1546. [3] 陈奎, 刘晓, 贾立娇, 等. 基于轻量化网络与增强多尺度特征融合的绝缘子缺陷检测[J]. 高电压技术, 2024, 50(3): 1289-1300. Chen Kui, Liu Xiao, Jia Lijiao, et al.Insulator defect detection based on lightweight network and enhanced multi-scale feature fusion[J]. High Voltage Engineering, 2024, 50(3): 1289-1300. [4] 苟军年, 杜愫愫, 王世铎, 等. 轻量化特征融合的CenterNet输电线路绝缘子自爆缺陷检测[J]. 北京航空航天大学学报, 2024, 50(7): 2161-2171. Gou Junnian, Du Susu, Wang Shiduo, et al.Insulator self-explosion detection in transmission line based on CenterNet fusing lightweight features[J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(7): 2161-2171. [5] 翟永杰, 王迪, 赵振兵, 等. 基于空域形态一致性特征的绝缘子串定位方法[J]. 中国电机工程学报, 2017, 37(5): 1568-1578. Zhai Yongjie, Wang Di, Zhao Zhenbing, et al.Insulator string location method based on spatial configuration consistency feature[J]. Proceedings of the CSEE, 2017, 37(5): 1568-1578. [6] 刘洋, 陆倚鹏, 高嵩, 等. 边缘检测在盘形悬式瓷绝缘子串红外图像上的应用[J]. 电瓷避雷器, 2020(1): 198-203. Liu Yang, Lu Yipeng, Gao Song, et al.Edge detection on infrared image of high voltage porcelain disc type suspension insulator strings[J]. Insulators and Surge Arresters, 2020(1): 198-203. [7] 蒲天骄, 乔骥, 韩笑, 等. 人工智能技术在电力设备运维检修中的研究及应用[J]. 高电压技术, 2020, 46(2): 369-383. Pu Tianjiao, Qiao Ji, Han Xiao, et al.Research and application of artificial intelligence in operation and maintenance for power equipment[J]. High Voltage Engineering, 2020, 46(2): 369-383. [8] 王卓, 王玉静, 王庆岩, 等. 基于协同深度学习的二阶段绝缘子故障检测方法[J]. 电工技术学报, 2021, 36(17): 3594-3604. Wang Zhuo, Wang Yujing, Wang Qingyan, et al.Two stage insulator fault detection method based on collaborative deep learning[J]. Transactions of China Electrotechnical Society, 2021, 36(17): 3594-3604. [9] 张欣, 王红星, 陈玉权, 等. 基于改进Cascade R-CNN算法的多类型绝缘子缺陷图像联合检测[J]. 电瓷避雷器, 2022(1): 189-196. Zhang Xin, Wang Hongxing, Chen Yuquan, et al.Multi-type insulator defect joint detection based on improved Cascade R-CNN algorithm[J]. Insulators and Surge Arresters, 2022(1): 189-196. [10] Ling Zenan, Zhang Dongxia, Qiu R C, et al.An accurate and real-time self-blast glass insulator location method based on faster R-CNN and U-net with aerial images[J]. CSEE Journal of Power and Energy Systems, 2019, 5(4): 474-482. [11] 苟军年, 杜愫愫, 刘力. 基于改进掩膜区域卷积神经网络的输电线路绝缘子自爆检测[J]. 电工技术学报, 2023, 38(1): 47-59. Gou Junnian, Du Susu, Liu Li.Transmission line insulator self-explosion detection based on improved mask region-convolutional neural network[J]. Transactions of China Electrotechnical Society, 2023, 38(1): 47-59. [12] 李斌, 屈璐瑶, 朱新山, 等. 基于多尺度特征融合的绝缘子缺陷检测[J]. 电工技术学报, 2023, 38(1): 60-70. Li Bin, Qu Luyao, Zhu Xinshan, et al.Insulator defect detection based on multi-scale feature fusion[J]. Transactions of China Electrotechnical Society, 2023, 38(1): 60-70. [13] 宋立业, 刘帅, 王凯, 等. 基于改进EfficientDet的电网元件及缺陷识别方法[J]. 电工技术学报, 2022, 37(9): 2241-2251. Song Liye, Liu Shuai, Wang Kai, et al.Identification method of power grid components and defects based on improved EfficientDet[J]. Transactions of China Electrotechnical Society, 2022, 37(9): 2241-2251. [14] 王韵琳, 冯天波, 孙宁, 等. 融合注意力与多尺度特征的电力绝缘子缺陷检测方法[J]. 高电压技术, 2024, 50(5): 1933-1942. Wang Yunlin, Feng Tianbo, Sun Ning, et al.Defect detection method for power insulators based on attention and multi-scale context information[J]. High Voltage Engineering, 2024, 50(5): 1933-1942. [15] 汤璐, 王淑青, 王年涛, 等. 基于改进YOLOX网络的雾天绝缘子缺陷检测[J]. 高压电器, 2024, 60(3): 223-228. Tang Lu, Wang Shuqing, Wang Niantao, et al.Insulator defect detection in foggy condition based on improved YOLOX network[J]. High Voltage Apparatus, 2024, 60(3): 223-228. [16] 张烨, 李博涛, 尚景浩, 等. 基于多尺度卷积注意力机制的输电线路防振锤缺陷检测[J]. 电工技术学报, 2024, 39(11): 3522-3537. Zhang Ye, Li Botao, Shang Jinghao, et al.Defect detection of transmission line damper based on multi-scale convolutional attention mechanism[J]. Transactions of China Electrotechnical Society, 2024, 39(11): 3522-3537. [17] Zhao Yian, Lü Wenyu, Xu Shangliang, et al. DETRs beat YOLOs on real-time object detection[J/OL]. ArXiv, 2024: 2304.08069v3. https://doi.org/10.48550/arXiv.2304.08069. [18] Zhang Jiangning, Li Xiangtai, Li Jian, et al.Rethinking mobile block for efficient attention-based models[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, 2023: 1389-1400. [19] Szegedy C, Vanhoucke V, Ioffe S, et al.Rethinking the inception architecture for computer vision[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016: 2818-2826. [20] Howard A G, Zhu Menglong, Chen Bo, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J/OL]. ArXiv, 2017: 1704.04861. http://arxiv.org/abs/1704.04861v1. [21] He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016: 770-778. [22] Sandler M, Howard A, Zhu Menglong, et al.MobileNetV2: inverted residuals and linear bottlenecks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 2018: 4510-4520. [23] Ma Ningning, Zhang Xiangyu, Zheng Haitao, et al.ShuffleNet V2: practical guidelines for efficient CNN architecture design[C]//Computer Vision-ECCV 2018: 15th European Conference, Munich, Germany, 2018: 122-138. [24] Zheng Zhaohui, Wang Ping, Liu Wei, et al.Distance-IoU loss: faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000. [25] Tao Xian, Zhang Dapeng, Wang Zihao, et al.Detection of power line insulator defects using aerial images analyzed with convolutional neural networks[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2020, 50(4): 1486-1498. [26] Everingham M, Ali Eslami S M, Van Gool L, et al. The pascal visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 111(1): 98-136.
2025, Vol. 40 
