翟永杰,李海森,吴童桐,苑朝
ZHAI Yongjie,LI Haisen,WU Tongtong,YUAN Chao

• 1:华北电力大学自动化系

摘要(Abstract)：

对航拍图像进行自动判别是无人机巡线后期的主要工作。为此提出一种改进的RPN(区域候选网络),以提高航拍图像中绝缘子目标的检测准确率。在绝缘子样本不完备的情况下,通过截取、旋转、镜像以及人工合成等方法对绝缘子训练样本进行扩充和完善;对人工标注的绝缘子样本的标注框进行聚类统计,获得标注框的宽高比分布情况,用于锚点框尺寸的初始化;对特征提取网络VGG16进行逐层分析,融合其中第二、三、五层的特征图,用于绝缘子目标识别;更改损失函数,实现动态调整正负样本的比例,从而解决训练过程中正负样本不均衡的问题。实验结果表明,改进后的RPN能够更有效地检测出航拍图像中的绝缘子目标,显著提高了检测的准确性。
In the process of unmanned aerial vehicle(UAV) inspection of transmission lines, automatic identification of aerial images is the main work in the later period of inspection. An improved RPN(region proposal network) is proposed to improve the identification accuracy of insulator targets in aerial images. In the case of incomplete insulator samples, the training samples are expanded and improved by intercepting, rotating,mirroring and artificial synthesis. Clustering statistics of labelled boxes of manually labelled insulator subsamples is carried out to obtain the distribution of width-height ratio of labelled boxes, which is used to initialize the dimension of anchor box. The feature extraction network VGG16 is analyzed layer by layer, and the second, third and fifth feature maps are fused for insulator target identification. By changing the loss function,the ratio of positive and negative samples can be dynamically adjusted to solve the problem of imbalance between positive and negative samples in the training process. Experimental results show that the improved RPN can detect the insulator targets more effectively in aerial images, and the accuracy is improved.

关键词(KeyWords)： 绝缘子;航拍图像;目标检测;区域候选网络;卷积神经网络;训练样本;深度学习
insulator;aerial image;target detection;RPN;convolutional neural network;training sample;deep learning

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金资助项目(61773160)

作者(Author): 翟永杰,李海森,吴童桐,苑朝
ZHAI Yongjie,LI Haisen,WU Tongtong,YUAN Chao

DOI: 10.19585/j.zjdl.201812013

参考文献(References)：

• [1]姚春羽,金立军,闫书佳.电网巡检图像中绝缘子的识别[J].系统仿真学报,2012,24(9):1818-1822.
• [2]翟永杰,王迪.基于骨架提取的航拍绝缘子图像分步识别方法[J].华北电力大学学报(自然科学版),2015,42(3):105-110.
• [3]OBERWEGER M,WENDEL A,BISCHOF H.Visual recog nition and fault detection for power line insulators[C]//19th CVWW,2014:1-8.
• [4]赵振兵,金思新,刘亚春.基于NSCT的航拍绝缘子图像边缘提取方法[J].仪器仪表学报,2012,33(9):2045-2052.
• [5]陈芳,姚建刚,李佐胜,等.绝缘子串红外图像中单个绝缘子盘面的提取方法[J].电网技术,2010,34(5):220-224.
• [6]马帅营,安居白,陈舫明.基于区域定位的绝缘子图像分割[J].电力建设,2010,31(7):14-17.
• [7]黄宵宁,张真良.直升机巡检航拍图像中绝缘子图像的提取算法[J].电网技术,2010,34(1):194-197.
• [8]林聚财,韩军,陈舫明,等.基于彩色图像的玻璃绝缘子缺陷诊断[J].电网技术,2011,35(1):127-133.
• [9]张少平,杨忠,黄宵宁,等.航拍图像中玻璃绝缘子自爆缺陷的检测及定位[J].太赫兹科学与电子信息学报,2013,11(4):609-613.
• [10]WU Q G,AN J B,LIN B.A texture segmentation algorithm based on PCA and GMAC for aerial insulator im ages[J].IEEE J.Sel.Topics Appl.Earth Observ.2012,5(5):1509-1518.
• [11]李卫国,叶高生,黄锋,等.基于改进MPEG-7纹理特征的绝缘子图像识别[J].高压电器,2010,46(10):65-68.
• [12]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):1137-1149.
• [13]LONG J,SHELHAMER E,DARRELL T.Fully convolutional networks for semantic segmentation[J].IEEE Computer Society,2017,39(4):640.
• [14]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2014:580-587.
• [15]GIRSHICK R.Fast R-CNN[C]//IEEE International Conference on Computer Vision(ICCV),2015:1440-1448.
• [16]ALEXE B,DESELAERS T,FERRARI V.Measuring the objectness of image windows[J].IEEE Transactions on Pattern Analysis&Mach Intelligence(TPAMI),2012,32(11):2189-2202.
• [17]UIJLINGS J R,K.E.van de Sande,GEVERS T,et al.Segmentation as selective search for object recognition[C]//International Conference on Computer Vision,2011:1879-1886.
• [18]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[C]//International Conference on Learning Representations(ICLR),2015:730-734.
• [19]HE K M,ZHANG X Y,REN S Q,et al.Deep Residual Learning for Image Recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2016:770-778.
• [20]Christian Szegedy Google Inc,Vincent Vanhoucke,Sergey Ioffe,Jonathon Shlens.Rethinking the Inception Architecture for Computer Vision[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2016:2818-2826.
• [21]SZEGEDY C,LIU W,JIA Y,et al.Going deeper with con volutions[C]//In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2015:1-9.
• [22]Michael Nielsen,Neural Network and Deep Learning[OL].http://neuralnetworksanddeeplearning.com,2017.
• [23]VoronoiDiagram.k-means clustering[OL].https://zh.wikipedi-a.org/zh/K-%E5%B9%B3%E5%9D%87%E7%AE%97%E6%B3%95,2015.
• [24]Aaron Schumacher.Alternate explanation of Mean Average Precision[OL].https://www.kaggle.com/c/FacebookRecruiting/discussion/2002,2012.