基于改进Faster-RCNN的绝缘子检测算法

doi:10.11772/j.issn.1001-9081.2021020342

《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (2): 631-637.DOI: 10.11772/j.issn.1001-9081.2021020342

• 前沿与综合应用 • 上一篇

基于改进Faster-RCNN的绝缘子检测算法

马耀名(), 张雨

辽宁工程技术大学电气与控制工程学院，辽宁葫芦岛 125105

收稿日期:2021-03-08 修回日期:2021-05-24 接受日期:2021-05-25 发布日期:2021-06-02 出版日期:2022-02-10
通讯作者: 马耀名
作者简介:马耀名（1975—），男，辽宁阜新人，副教授，硕士，主要研究方向：智能电网、电力系统运行监测与控制；
张雨（1994—），男，辽宁葫芦岛人，硕士研究生，主要研究方向：电力系统运行监测与控制、目标检测。

Insulator detection algorithm based on improved Faster-RCNN

Yaoming MA(), Yu ZHANG

Faculty of Electrical and Control Engineering，Liaoning Technical University，Huludao Liaoning 125105，China

Received:2021-03-08 Revised:2021-05-24 Accepted:2021-05-25 Online:2021-06-02 Published:2022-02-10
Contact: Yaoming MA
About author:MA Yaoming， born in 1975， M. S.， associate professor. His research interests include smart grid， power system operation monitoring and control.
ZHANG Yu， born in 1994， M. S. candidate. His research interests include power system operation monitoring and control， object detection.

摘要/Abstract

摘要：

为了提高高压输电线路巡检效率，提出改进Faster-RCNN的绝缘子检测算法。首先，在特征提取网络中添加具有注意力机制动态选择机制网络（SKNet），从而使网络着重学习与绝缘子特征相关通道；其次，借助滤波器响应归一化（FRN）层替代原批归一化（BN）层，以避免模型陷入梯度饱和区域；最后，使用距离交并比（DIoU）代替原交并比（IoU）方法，以精确表达特征候选区域框位置。对开源航拍绝缘子数据集进行平移、旋转、Cutout和CutMix等操作来进行增强，将数据集扩充到3 000张并从中随机选择2 500张作为训练集，其余500张作为测试集。相较于原Faster-RCNN算法，所提算法平均准确率提高了3.46个百分点，平均召回率提高了2.76个百分点。实验结果表明：所提算法具有较高检测精度和稳定性，可满足电力巡线绝缘子检测应用场景需求。

关键词: 绝缘子检测, Faster-RCNN, 动态选择机制网络, 距离交并比, 滤波器响应归一化

Abstract:

In order to increase inspection efficiency of high-voltage transmission lines， an insulator detection algorithm based on improved Faster Region-based Convolutional Neural Network （Faster-RCNN） was proposed. Firstly， the Selective Kernel Neural Network （SKNet） with attention mechanism was added to feature extraction network to make the network focus on learning the insulator features related channels. Secondly， the Filter Response Normalization （FRN） layer was used to replace the original Batch Normalization （BN） layer to avoid the model falling into the gradient saturation region. Finally， the Distance Intersection Over Union （DIoU） was used to replace the original Intersection Over Union （IoU） to accurately express the positions of the feature candidate region boxs. The open source aerial insulator dataset was enhanced by the operations such as translation， rotation， Cutout and CutMix. The dataset was expanded to 3 000 images， and 2 500 images of them were randomly selected as the training set， and the remaining 500 images were selected as the test set. Compared with the original Faster-RCNN algorithm， the average accuracy of the proposed algorithm is improved by 3.46 percentage points， and the average recall is improved by 2.76 percentage points. Experimental results show that the proposed algorithm has high detection accuracy and stability， and can meet the requirements of the application scenarios of power line insulator detection.

Key words: insulator detection, Faster Region-based Convolutional Neural Network (Faster-RCNN), Selective Kernel Neural Network (SKNet), Distance Intersection Over Union (DIoU), Filter Response Normalization (FRN)

中图分类号:

TP391.4

马耀名, 张雨. 基于改进Faster-RCNN的绝缘子检测算法[J]. 计算机应用, 2022, 42(2): 631-637.

Yaoming MA, Yu ZHANG. Insulator detection algorithm based on improved Faster-RCNN[J]. Journal of Computer Applications, 2022, 42(2): 631-637.

图/表 13

图1 SKNet通道注意力模块

Fig. 1 SKNet channel attention module

图2 IoU与DIoU对比

Fig. 2 Comparison of IoU and DIoU

图3 FRN归一化层

Fig. 3 FRN normalization layer

图4 Faster-RCNN算法流程

Fig. 4 Flowchart of Faster-RCNN algorithm

图5 基于SKNet的残差结构

Fig. 5 Residual structure based on SKNet

表1 SKNet-ResNet50网络结构

Tab. 1 SKNet-ResNet50 network structure

层数	卷积层参数
卷积层1	$7 × 7$ 卷积核，64层，步长2
卷积层2_x	$3 × 3$ 最大池化层，步长2
卷积层3_x	$1 × 1,64 S K N e t 1 × 1,256 × 3$
卷积层4_x	$1 × 1,128 S K N e t 1 × 1,512 × 4$
卷积层5_x	$1 × 1,256 S K N e t 1 × 1,1024 × 6$
卷积层6_x	$1 × 1,512 S K N e t 3 × 3,512 × 3$

表1 SKNet-ResNet50网络结构

Tab. 1 SKNet-ResNet50 network structure

层数	卷积层参数
卷积层1	$7 × 7$ 卷积核，64层，步长2
卷积层2_x	$3 × 3$ 最大池化层，步长2
卷积层3_x	$1 × 1,64 S K N e t 1 × 1,256 × 3$
卷积层4_x	$1 × 1,128 S K N e t 1 × 1,512 × 4$
卷积层5_x	$1 × 1,256 S K N e t 1 × 1,1024 × 6$
卷积层6_x	$1 × 1,512 S K N e t 3 × 3,512 × 3$

图6 候选区域提取层

Fig. 6 Candidate region extraction layer

图7 Cutout操作

Fig. 7 Cutout operation

图8 CutMix操作

Fig. 8 CutMix operation

表2 绝缘子漏检、错检统计结果

Tab. 2 Statistical results of missing and wrong insulator detection

算法	绝缘子总数	正确数	遗漏数	错误数
Faster-RCNN	1 435	1 128	123	184
本文算法	1 435	1 311	35	89

表3 Faster-RCNN算法改变网络后数据统计 ( %)

Tab. 3 Data statistics after Faster-RCNN algorithm changing network

算法	AP	AR
Faster-RCNN	89.79	93.67
Faster-RCNN+DIoU	92.81	94.12
Faster-RCNN+FRN	91.32	93.74
Faster-RCNN+SKNet	90.10	93.90
本文算法	93.25	96.43

表4 不同算法实验结果对比 ( %)

Tab. 4 Experiment result comparison of different algorithms

算法	AP	AR
Faster-RCNN	89.79	93.67
YOLO-V2	85.98	89.15
YOLO-V3	90.51	93.97
本文算法	93.25	96.43

图9 不同算法检测效果对比

Fig. 9 Comparison of detection effect of different algorithms

参考文献 28

1	罗昊，苏盛，杨浩，等.基于FPGA的电力巡线无人机硬件加密通信方法［J］.中国电力， 2019， 52（7）： 11-16. 10.11930/j.issn.1004-9649.201903080
	LUO H， SU S， YANG H， et al. FPGA-based hardware encryption of power line patrol drones［J］. Electric Power， 2019， 52（7）： 11-16. 10.11930/j.issn.1004-9649.201903080
2	钱世豪.基于图像处理的输电线路异物检测技术研究与电力巡线管理系统设计［D］.南京：南京理工大学， 2019： 29-35.
	QIAN S H. Study on foreign body detection technology of transmission line based on image processing and design of power line inspection management system［D］. Nanjing： Nanjing University of Science and Technology， 2019： 29-35.
3	MARTINEZ C， SAMPEDRO C， CHAUHAN A， et al. The Power Line Inspection Software （PoLIS）： a versatile system for automating power line inspection［J］. Engineering Applications of Artificial Intelligence， 2018， 71： 293-314. 10.1016/j.engappai.2018.02.008
4	TAO X， ZHANG D P， WANG Z H， 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. 10.1109/tsmc.2018.2871750
5	郝艳捧，蒋晓蓝，阳林，等.基于图像分割评估运行绝缘子自然覆冰程度［J］.高电压技术， 2017， 43（1）： 285-292. 10.13336/j.1003-6520.hve.20161227037
	HAO Y P， JIANG X L， YANG L， et al. Evaluation of natural icing condition for in-service insulators based on image segmentation［J］. High Voltage Engineering， 2017， 43（1）： 285-292. 10.13336/j.1003-6520.hve.20161227037
6	黄新波，李菊清，张烨，等.复杂环境下覆冰绝缘子识别检测技术［J］.高电压技术， 2017， 43（3）： 891-899. 10.13336/j.1003-6520.hve.20170303028
	HUANG X B， LI J Q， ZHANG Y， et al. Recognition and detection technology of ice-covered insulators under complex environment［J］. High Voltage Engineering， 2017， 43（3）： 891-899. 10.13336/j.1003-6520.hve.20170303028
7	苑利，赵锐，谭孝元，等.基于红外成像技术的零值绝缘子检测［J］.高压电器， 2018， 54（2）： 97-102. 10.13296/j.1001-1609.hva.2018.02.016
	YUAN L， ZHAO R， TAN X Y， et al. Zero insulator detection based on the infrared imaging technology［J］. High Voltage Apparatus， 2018， 54（2）： 97-102. 10.13296/j.1001-1609.hva.2018.02.016
8	姚晓通，刘力，李致远.基于Canny边缘特征点的接触网绝缘子识别方法［J］.电瓷避雷器， 2020（1）： 142-148. 10.1088/1757-899x/394/3/032023
	YAO X T， LIU L， LI Z Y. Identification method of catenary insulator based on Canny edge feature point［J］. Insulators and Surge Arresters， 2020（1）： 142-148. 10.1088/1757-899x/394/3/032023
9	白洁音，赵瑞，谷丰强，等.多目标检测和故障识别图像处理方法［J］.高电压技术， 2019， 45（11）： 3504-3511.
	BAI J Y， ZHAO R， GU F Q， et al. Multi-target detection and fault recognition image processing method［J］. High Voltage Engineering， 2019， 45（11）： 3504-3511.
10	REN S Q， HE K M， GIRSHICK R， et al. Faster R-CNN： towards real-time object detection with region proposal networks［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2017， 39（6）： 1137-1149. 10.1109/tpami.2016.2577031
11	赵文清，程幸福，赵振兵，等.注意力机制和Faster RCNN相结合的绝缘子识别［J］.智能系统学报， 2020， 15（1）： 92-98. 10.11992/tis.201907023
	ZHAO W Q， CHENG X F， ZHAO Z B， et al. Insulator recognition based on attention mechanism and Faster RCNN［J］. CAAI Transactions on Intelligent Systems， 2020， 15（1）： 92-98. 10.11992/tis.201907023
12	HU J， SHEN L， SUN G. Squeeze-and-excitation networks ［C］// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2018： 7132-7141. 10.1109/cvpr.2018.00745
13	蒋姗，孙渊，严道森.基于深度学习算法的航拍巡检图像的绝缘子识别［J］.福州大学学报（自然科学版）， 2021， 49（1）： 58-64.
	JIANG S， SUN Y， YAN D S. Insulator recognition of aerial patrol image based on deep learning algorithm［J］. Journal of Fuzhou University （Natural Science Edition）， 2021， 49（1）： 58-64.
14	LIN T Y， 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. Piscataway： IEEE， 2017： 936-944. 10.1109/cvpr.2017.106
15	BODLA N， SINGH B， CHELLAPPA R， et al. Soft-NMS — improving object detection with one line of code ［C］// Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway： IEEE， 2017： 5562-5570. 10.1109/iccv.2017.593
16	赖秋频，杨军，谭本东，等.基于YOLOv2网络的绝缘子自动识别与缺陷诊断模型［J］.中国电力， 2019， 52（7）： 31-39. 10.11930/j.issn.1004-9649.201806102
	LAI Q P， YANG J， TAN B D， et al. An automatic recognition and defect diagnosis model of transmission line insulator based on YOLOv2 network［J］. Electric Power， 2019， 52（7）： 31-39. 10.11930/j.issn.1004-9649.201806102
17	REDMON J， FARHADI A. YOLO9000： better， faster， stronger ［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2017： 6517-6525. 10.1109/cvpr.2017.690
18	REDMON J， FARHADI A. YOLOv3： an incremental improvement［EB/OL］. （2018-04-08）［2020-12-10］. . 10.1109/cvpr.2018.00430
19	颜宏文，陈金鑫.基于改进YOLOv3的绝缘子串定位与状态识别方法［J］.高电压技术， 2020， 46（2）： 423-431. 10.13336/j.1003-6520.hve.20200131006
	YAN H W， CHEN J X. Insulator string positioning and state recognition method based on improved YOLOv3 algorithm［J］. High Voltage Engineering， 2020， 46（2）： 423-431. 10.13336/j.1003-6520.hve.20200131006
20	杨焰飞，曹阳.改进YOLOv3的无人机拍摄图玻璃绝缘子检测［J/OL］.计算机工程与应用［2021-01-24］. ，. 10.1109/icpics52425.2021.9524125
	CAO Y. Detection of glass insulators in images taken by drones based on improved YOLOv3［J/OL］. Computer Engineering and Applications. ［2021-01-24］. 10.1109/icpics52425.2021.9524125
21	SZEGEDY C， IOFFE S， VANHOUCKE V， et al. Inception-v4， Inception-ResNet and the impact of residual connections on learning［EB/OL］. （2016-08-23）［2021-01-24］. .
22	LI X， WANG W H， HU X L， et al. Selective kernel networks ［C］// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2019： 510-519. 10.1109/cvpr.2019.00060
23	SINGH S， KRISHNAN S. Filter response normalization layer： eliminating batch dependence in the training of deep neural networks ［C］// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2020： 11234-11243. 10.1109/cvpr42600.2020.01125
24	IOFFE S， SZEGEDY C. Batch normalization： accelerating deep network training by reducing internal covariate shift ［C］// Proceedings of the 32nd International Conference on Machine Learning. New York： JMLR.org， 2015： 448-456.
25	ZHENG Z H， WANG P， LIU W， et al. Distance-IoU loss： faster and better learning for bounding box regression ［C］// Proceedings of the 24th AAAI Conference on Artificial Intelligence. Palo Alto， CA： AAAI Press， 2020： 12993-13000. 10.1609/aaai.v34i07.6999
26	HE K M， ZHANG X Y， REN S Q. Deep residual learning for image recognition ［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2016： 27-30. 10.1109/cvpr.2016.90
27	DE VRIES T， TAYLOR G W. Improved regularization of convolutional neural networks with cutout［EB/OL］. （2017-11-29）［2021-01-24］. .
28	YUN S， HAN D， CHUN S， et al. CutMix： regularization strategy to train strong classifiers with localizable features ［C］// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision. Piscataway： IEEE， 2019： 6022-6031. 10.1109/iccv.2019.00612

基于改进Faster-RCNN的绝缘子检测算法

Insulator detection algorithm based on improved Faster-RCNN

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 28

相关文章 1

编辑推荐

Metrics