基于强泛化卷积神经网络的输电线路图像覆冰厚度辨识
|
摘要
实际自然场景下的覆冰监测系统中,受天气、光线、摄像头老化和角度等问题的影响,覆冰图像具有低分辨率化和多形态化的特性,如何找到具有强泛化能力的覆冰图像识别方法成为关注的热点。该文提出一种基于强泛化卷积神经网络(India buffet process-convolutional neural network,IBP-CNN)的输电线路覆冰厚度识别方法。该方法首先通过增强与消减算法确定滤波器数目及滤波器参数,以减少模型的冗余度,然后基于输出损失函数,利用反向传播算法调整层间连接权值,最后根据更新的模型参数和网络结构推算逐层输出,得到强泛化性的IBP-CNN网络。实际场景数据集测试结果表明,相比力学模型监测方法、图像边缘检测和浅层机器学习方法,IBP-CNN能够在不同分辨率和不同位置角度的覆冰图像场景下保持较高的辨识精度和速度,具有较强的自然场景泛化能力和工程实用价值。
In the natural icing monitoring system, the icing images have the characteristics of low resolution and multi-morphological due to the problem of weather and light changing, camera aging and angle altering. How to find the image-based recognition method with strong generalization performance has become the research hotspots. In this paper, an India buffet process-convolutional neural network(IBP-CNN) method based on strong generalized convolutional neural network was proposed to identify the icing thickness of the transmission line. Firstly the number of each layer filter and the parameters of filters were determined by grow-and-prune(GAP) mechanism to reduce the redundancy of the model. Then the connected weights among layers were trained by back propagation algorithm based on loss function. Finally, the layer-wise outputs were inferred to minimize the reconstruction loss again, according to the updated parameters and network structure. Compared with mechanical model monitoring method, image edge detection method and shallow machine learning method, the experimental results show that the proposed method can maintain high accuracy and low computational complexity under different resolution and different position angles, it has strong natural scene generalization ability and can be put into practice in the future.
In the natural icing monitoring system, the icing images have the characteristics of low resolution and multi-morphological due to the problem of weather and light changing, camera aging and angle altering. How to find the image-based recognition method with strong generalization performance has become the research hotspots. In this paper, an India buffet process-convolutional neural network(IBP-CNN) method based on strong generalized convolutional neural network was proposed to identify the icing thickness of the transmission line. Firstly the number of each layer filter and the parameters of filters were determined by grow-and-prune(GAP) mechanism to reduce the redundancy of the model. Then the connected weights among layers were trained by back propagation algorithm based on loss function. Finally, the layer-wise outputs were inferred to minimize the reconstruction loss again, according to the updated parameters and network structure. Compared with mechanical model monitoring method, image edge detection method and shallow machine learning method, the experimental results show that the proposed method can maintain high accuracy and low computational complexity under different resolution and different position angles, it has strong natural scene generalization ability and can be put into practice in the future.
引文
[1]王黎明,李海东,梅红伟,等.输电线路覆冰在线监测系统国内外研究综述[J].高压电器,2013,49(6):48-56.Wang Liming,Li Haidong,Mei Hongwei,et al.Review of on-line monitoring system of ice coating on transmission lines[J].High Voltage Apparatus,2013,49(6):48-56(in Chinese).
[2]曹永兴,张昌华,黄琦,等.输电线路覆冰在线监测及预警技术的国内外研究现状[J].华东电力,2011,39(1):96-99.Cao Yongxing,Zhang Changhua,Huang Qi,et al.Research status of on-line ice coating monitoring&forewarning technology of overhead transmission lines in China and abroad[J].East China Electric Power,2011,39(1):96-99(in Chinese).
[3]李再华,白晓民,周子冠,等.电网覆冰防治方法和研究进展[J].电网技术,2008,32(4):7-13,22.Li Zaihua,Bai Xiaomin,Zhou Ziguan,et al.Prevention and treatment methods of ice coating in power networks and its recent study[J].Power System Technology,2008,32(4):7-13,22(in Chinese).
[4]姚陈果,张磊,李成祥,等.基于力学分析和弧垂测量的导线覆冰厚度测量方法[J].高电压技术,2013,39(5):1204-1209.Yao Chenguo,Zhang Lei,Li Chengxiang,et al.Measurement method of conductor ice covered thickness based on analysis of mechanical and sag measurement[J].High Voltage Engineering,2013,39(5):1204-1209(in Chinese).
[5]黄文焘,邰能灵,范春菊.基于杆塔结构力学测量的线路覆冰在线监测系统研究[J].电力系统保护与控制,2012,40(24):71-75,83.Huang Wentao,Tai Nengling,Fan Chunju.Study on icing monitoring system of different tower overhead transmission lines based on mechanics measurements[J].Power System Protection and Control,2012,40(24):71-75,83(in Chinese).
[6]王阳光,尹项根,游大海,等.基于无线传感器网络的电力设施冰灾实时监测与预警系统[J].电网技术,2009,33(7):14-19.Wang Yangguang,Yin Xianggen,You Dahai,et al.A real-time monitoring and warning system for electric power facilities icing disaster based on wireless sensor network[J].Power System Technology,2009,33(7):14-19(in Chinese).
[7]李昭廷,郝艳捧,李立浧,等.利用远程系统的输电线路覆冰厚度图像识别[J].高电压技术,2011,37(9):2288-2293.Li Zhaoting,Hao Yanpeng,Li Licheng,et al.Image recognition of ice thickness on transmission lines using remote system[J].High Voltage Engineering,2011,37(9):2288-2293(in Chinese).
[8]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.Hao Yanpeng,Jiang Xiaolan,Yang Lin,et al.Evaluation of natural icing condition for in-service insulators based on image segmentation[J].High Voltage Engineering,2017,43(1):285-292(in Chinese).
[9]史尊伟,阳林,李昊.基于改进Canny算子的覆冰厚度测量方法[J].电瓷避雷器,2013(6):24-29.Shi Zunwei,Yang Lin,Li Hao.Icing thickness measuring based on improved canny operator[J].Insulators and Surge Arresters,2013(6):24-29(in Chinese).
[10]陆佳政,张红先,方针,等.自适应分割阈值在覆冰厚度识别中的应用[J].高电压技术,2009,35(3):563-567.Luo Jiazheng,Zhang Hongxian,Fang Zhen,et al.Application of self-adaptive segmental threshold to ice thickness identification[j].High Voltage Engineering,2009,35(3):563-567(in Chinese).
[11]李小娟.基于数据挖掘的输电线路覆冰预测模型研究[D].太原:太原理工大学,2016.Li Xiaojuan.Research of the icing prediction model of transmission line based on data mining[D].Taiyuan:Taiyuan University of Technology,2016(in Chinese).
[12]欧阳丽莎,黄新波.基于灰关联分析微气象因素和导线温度对输电线路导线覆冰的影响[J].高压电器,2011,47(3):31-36.Ouyang Lisha,Huang Xinbo.Influences of meteorological conditions and conductor temperature on icing of transmission line based on grey relational analysis[J].High Voltage Apparatus,2011,47(3):31-36(in Chinese).
[13]黄新波,王玉鑫,朱永灿,等.基于遗传算法与模糊逻辑融合的线路覆冰预测[J].高电压技术,2016,42(4):1228-1235.Huang Xinbo,Wang Yuxin,Zhu Yongcan,et al.Icing forecast of transmission line based on genetic algorithm and fuzzy logic[J].High Voltage Engineering,2016,42(4):1228-1235(in Chinese).
[14]黄新波,李佳杰,欧阳丽莎,等.采用模糊逻辑理论的覆冰厚度预测模型[J].高电压技术,2011,37(5):1245-1252.Huang Xinbo,Li Jiajie,Ouyang Lisha,et al.Icing thickness prediction model using fuzzy logic theory[J].High Voltage Engineering,2011,37(5):1245-1252(in Chinese).
[15]戴栋,黄筱婷,代州,等.基于支持向量机的输电线路覆冰回归模型[J].高电压技术,2013,39(11):2822-2828.Dai Dong,Huang Xiaoting,Dai Zhou,et al.Regression model for transmission lines icing based on support vector machine[J].High Voltage Engineering,2013,39(11):2822-2828(in Chinese).
[16]黄宵宁,许家浩,杨成顺,等.基于数据驱动算法和LS-SVM的输电线路覆冰预测[J].电力系统自动化,2014,38(15):81-86.Huang Xiaoning,Xu Jiahao,Yang Chengshun,et al.Transmission line icing prediction based on data driven algorithm and LS-SVM[J].Automation of Electric Power Systems,2014,38(15):81-86(in Chinese).
[17]Hinton G E,Salakhutdinov R R.Reducing the dimensionality of data with neural networks[J].Science,2006,313(5786):504-507.
[18]Hinton G E.Learning multiple layers of representation[J].Trends in Cognitive Sciences,2007,11(10):428-434.
[19]Feng Jiashi,Darrell T.Learning the structure of deep convolutional networks[C]//Proceedings of 2015 IEEE International Conference on Computer Vision.Santiago,Chile:IEEE,2015:2749-2757.
[20]陈先昌.基于卷积神经网络的深度学习算法与应用研究[D].杭州:浙江工商大学,2014.Chen Xianchang.Research on algorithm and application of deep learning based on convolutional neural network[D].Hangzhou:Zhejiang Gongshang University,2014(in Chinese).
[21]Hinton G.Where do features come from?[J].Cognitive Science A Multidisciplinary Journal,2014,38(6):1078-1101.
[22]Krizhevsky A,Sutskever I,Hinton G E.Image Net classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems.Lake Tahoe,Nevada:ACM,2012:1097-1105.
[23]Zhou Shusen,Chen Qingcai,Wang Xiaolong.Convolutional deep networks for visual data classification[J].Neural Processing Letters,2013,38(1):17-27.
[24]Le Cun Y,Bengio Y,Hinton G.Deep learning[J].Nature,2015,521(7553):436-444.
[25]Chen Yushi,Lin Zhouhan,Zhao Xing,et al.Deep learning-based classification of hyperspectral data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2014,7(6):2094-2107.
[26]郝艳捧,刘国特,薛艺为,等.输电线路覆冰厚度的小波分析图像识别[J].高电压技术,2014,40(2):368-373.Hao Yanpeng,Liu Guote,Xue Yiwei,et al.Wavelet image recognition of ice thickness on transmission lines[J].High Voltage Technology,2014,40(2):368-373(in Chinese).
[2]曹永兴,张昌华,黄琦,等.输电线路覆冰在线监测及预警技术的国内外研究现状[J].华东电力,2011,39(1):96-99.Cao Yongxing,Zhang Changhua,Huang Qi,et al.Research status of on-line ice coating monitoring&forewarning technology of overhead transmission lines in China and abroad[J].East China Electric Power,2011,39(1):96-99(in Chinese).
[3]李再华,白晓民,周子冠,等.电网覆冰防治方法和研究进展[J].电网技术,2008,32(4):7-13,22.Li Zaihua,Bai Xiaomin,Zhou Ziguan,et al.Prevention and treatment methods of ice coating in power networks and its recent study[J].Power System Technology,2008,32(4):7-13,22(in Chinese).
[4]姚陈果,张磊,李成祥,等.基于力学分析和弧垂测量的导线覆冰厚度测量方法[J].高电压技术,2013,39(5):1204-1209.Yao Chenguo,Zhang Lei,Li Chengxiang,et al.Measurement method of conductor ice covered thickness based on analysis of mechanical and sag measurement[J].High Voltage Engineering,2013,39(5):1204-1209(in Chinese).
[5]黄文焘,邰能灵,范春菊.基于杆塔结构力学测量的线路覆冰在线监测系统研究[J].电力系统保护与控制,2012,40(24):71-75,83.Huang Wentao,Tai Nengling,Fan Chunju.Study on icing monitoring system of different tower overhead transmission lines based on mechanics measurements[J].Power System Protection and Control,2012,40(24):71-75,83(in Chinese).
[6]王阳光,尹项根,游大海,等.基于无线传感器网络的电力设施冰灾实时监测与预警系统[J].电网技术,2009,33(7):14-19.Wang Yangguang,Yin Xianggen,You Dahai,et al.A real-time monitoring and warning system for electric power facilities icing disaster based on wireless sensor network[J].Power System Technology,2009,33(7):14-19(in Chinese).
[7]李昭廷,郝艳捧,李立浧,等.利用远程系统的输电线路覆冰厚度图像识别[J].高电压技术,2011,37(9):2288-2293.Li Zhaoting,Hao Yanpeng,Li Licheng,et al.Image recognition of ice thickness on transmission lines using remote system[J].High Voltage Engineering,2011,37(9):2288-2293(in Chinese).
[8]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.Hao Yanpeng,Jiang Xiaolan,Yang Lin,et al.Evaluation of natural icing condition for in-service insulators based on image segmentation[J].High Voltage Engineering,2017,43(1):285-292(in Chinese).
[9]史尊伟,阳林,李昊.基于改进Canny算子的覆冰厚度测量方法[J].电瓷避雷器,2013(6):24-29.Shi Zunwei,Yang Lin,Li Hao.Icing thickness measuring based on improved canny operator[J].Insulators and Surge Arresters,2013(6):24-29(in Chinese).
[10]陆佳政,张红先,方针,等.自适应分割阈值在覆冰厚度识别中的应用[J].高电压技术,2009,35(3):563-567.Luo Jiazheng,Zhang Hongxian,Fang Zhen,et al.Application of self-adaptive segmental threshold to ice thickness identification[j].High Voltage Engineering,2009,35(3):563-567(in Chinese).
[11]李小娟.基于数据挖掘的输电线路覆冰预测模型研究[D].太原:太原理工大学,2016.Li Xiaojuan.Research of the icing prediction model of transmission line based on data mining[D].Taiyuan:Taiyuan University of Technology,2016(in Chinese).
[12]欧阳丽莎,黄新波.基于灰关联分析微气象因素和导线温度对输电线路导线覆冰的影响[J].高压电器,2011,47(3):31-36.Ouyang Lisha,Huang Xinbo.Influences of meteorological conditions and conductor temperature on icing of transmission line based on grey relational analysis[J].High Voltage Apparatus,2011,47(3):31-36(in Chinese).
[13]黄新波,王玉鑫,朱永灿,等.基于遗传算法与模糊逻辑融合的线路覆冰预测[J].高电压技术,2016,42(4):1228-1235.Huang Xinbo,Wang Yuxin,Zhu Yongcan,et al.Icing forecast of transmission line based on genetic algorithm and fuzzy logic[J].High Voltage Engineering,2016,42(4):1228-1235(in Chinese).
[14]黄新波,李佳杰,欧阳丽莎,等.采用模糊逻辑理论的覆冰厚度预测模型[J].高电压技术,2011,37(5):1245-1252.Huang Xinbo,Li Jiajie,Ouyang Lisha,et al.Icing thickness prediction model using fuzzy logic theory[J].High Voltage Engineering,2011,37(5):1245-1252(in Chinese).
[15]戴栋,黄筱婷,代州,等.基于支持向量机的输电线路覆冰回归模型[J].高电压技术,2013,39(11):2822-2828.Dai Dong,Huang Xiaoting,Dai Zhou,et al.Regression model for transmission lines icing based on support vector machine[J].High Voltage Engineering,2013,39(11):2822-2828(in Chinese).
[16]黄宵宁,许家浩,杨成顺,等.基于数据驱动算法和LS-SVM的输电线路覆冰预测[J].电力系统自动化,2014,38(15):81-86.Huang Xiaoning,Xu Jiahao,Yang Chengshun,et al.Transmission line icing prediction based on data driven algorithm and LS-SVM[J].Automation of Electric Power Systems,2014,38(15):81-86(in Chinese).
[17]Hinton G E,Salakhutdinov R R.Reducing the dimensionality of data with neural networks[J].Science,2006,313(5786):504-507.
[18]Hinton G E.Learning multiple layers of representation[J].Trends in Cognitive Sciences,2007,11(10):428-434.
[19]Feng Jiashi,Darrell T.Learning the structure of deep convolutional networks[C]//Proceedings of 2015 IEEE International Conference on Computer Vision.Santiago,Chile:IEEE,2015:2749-2757.
[20]陈先昌.基于卷积神经网络的深度学习算法与应用研究[D].杭州:浙江工商大学,2014.Chen Xianchang.Research on algorithm and application of deep learning based on convolutional neural network[D].Hangzhou:Zhejiang Gongshang University,2014(in Chinese).
[21]Hinton G.Where do features come from?[J].Cognitive Science A Multidisciplinary Journal,2014,38(6):1078-1101.
[22]Krizhevsky A,Sutskever I,Hinton G E.Image Net classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems.Lake Tahoe,Nevada:ACM,2012:1097-1105.
[23]Zhou Shusen,Chen Qingcai,Wang Xiaolong.Convolutional deep networks for visual data classification[J].Neural Processing Letters,2013,38(1):17-27.
[24]Le Cun Y,Bengio Y,Hinton G.Deep learning[J].Nature,2015,521(7553):436-444.
[25]Chen Yushi,Lin Zhouhan,Zhao Xing,et al.Deep learning-based classification of hyperspectral data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2014,7(6):2094-2107.
[26]郝艳捧,刘国特,薛艺为,等.输电线路覆冰厚度的小波分析图像识别[J].高电压技术,2014,40(2):368-373.Hao Yanpeng,Liu Guote,Xue Yiwei,et al.Wavelet image recognition of ice thickness on transmission lines[J].High Voltage Technology,2014,40(2):368-373(in Chinese).