可视化与非可视化特征融合超声3D目标识别研究

doi:10.16180/j.cnki.issn1007-7820.2019.05.002

摘要/Abstract

摘要：

目前常用的超声3D目标识别方法主要是利用传感器在空间一点或多点获取一维回波,通过信号处理得到目标体3D信息以实现3D目标体识别。这些方法普遍存在识别率低和鲁棒性差的问题,制约了该项技术的推广和应用。为此,文中提出了一种基于可视化和非可视化特征融合的超声3D目标体识别方法,该方法将目标体回波信号处理方法与合成孔径方法相结合,将提取的目标体信息在特征层进行了融合,然后经BP神经网络实现了分类识别,可使现有方法的不足得到显著改善。通过对3类人工靶标的实验表明,该方法可显著提高缺陷的3D识别率,能够保持在90%以上,且鲁棒性也得到明显改善。

关键词: 数据融合, 超声, 3D目标识别, 合成孔径, 可视化, 非可视化, 特征提取

Abstract:

Nowadays, the main method of ultrasonic 3D target recognition was using sensors to obtain one or more one-dimensional echo in space and getting the 3D information of the target body by signal processing to realize the 3D target recognition. These methods generally exist the problem of low recognition rate and poor robustness,which restrict the popularization and application of this technology. In this paper, an ultrasonic 3D target recognition method based on visual and non-visual feature fusion was proposed. This method combined the target body echo signal processing method with the synthetic aperture method, carried out data fusion of the extracted target information in the feature layer and realized the classification recognition by the BP neural network, by which the shortage of existing methods can be significantly improved. Experiments on three kinds of artificial targets showed that the method can significantly improve the 3D recognition rate of the defect which can be kept above 90%, and the robustness was also improved obviously.

Key words: data fusion, ultrasonic, 3D target recognition, synthetic aperture, visual, non-visual, feature extraction

中图分类号:

TP391

宋寿鹏,申静静,卢翠娥. 可视化与非可视化特征融合超声3D目标识别研究[J]. 电子科技, 2019, 32(5): 5-6.

SONG Shoupeng,SHEN Jingjing,LU Cuie. Research on Ultrasonic 3D Target Recognition Based on Visual and Non-Visual Feature Fusion[J]. Electronic Science and Technology, 2019, 32(5): 5-6.

图/表 7

图1

图2

图3

图4

表1

表2

表3

参考文献 20

[1]	Chen S, Huang Y, Zhang H , et al. Application of fractal theory on identification of near-surface defects in ultrasonic A-Scan detection[J]. Journal of Computer Applications, 2014,5(3):1978-1989.
[2]	王丽莎, 李梦洁, 汪路明 , 等. 基于多特征优化的超声波缺陷分类识别方法研究[J]. 浙江树人大学学报, 2017(3):12-16.
	Wang Lisha, Li Mengjie, Wang Luming . Research on defect classification and recognition based on multi-feature optimization in ultrasonic detection[J]. Journal of Zhejiang Shuren University, 2017(3):12-16.
[3]	李迎雪 . 基于稀疏表示的管道缺陷超声复合阵列成像研究[D]. 镇江:江苏大学, 2017.
	Li Yingxue . Research on Sparse-Representation based ultrasonic composite array imaging for pipeline defect[D]. Zhenjiang:Jiangsu University, 2017.
[4]	Fan Z, Ma S Y, Wu Z J . Experimental investigation of underwater pipeline inspection using ultrasonic guided waves[J]. Piezoelectrics and Acoustooptics, 2014,36(3):480-483.
[5]	姜岩, 涂骏 . 超声波合成孔径聚焦3D成像技术在点焊质量检测中的应用[J]. 金属加工(冷加工), 2016(s1):361-364.
	Jiang Yan, Tu Jun . Application of ultrasonic synthetic aperture focusing 3D imaging technology in spot welding quality inspection[J]. Machinist Metal Cutting, 2016(s1):361-364.
[6]	Trots I, Nowicki A, Lewandowski M . Synthetic transmit aperture method in medical ultrasonic imaging[J]. World Academy of Science,Engineering and Technology, 2010(40):294-297.
[7]	李运才, 王泽勇 . 一种超声合成孔径成像优化方法研究[J]. 信息技术, 2016(2):139-141.
	Li Yuncai, Wang Zeyong . Study on optimization approach of ultrasonic synthetic aperture imaging[J]. Information Technology, 2016(2):139-141.
[8]	陈志伟, 陈凯歌, 蒋俊俊 . 基于超声导波波谱法管道缺陷检测仿真研究[J]. 电子科技, 2016,29(9):48-51.
	Chen Zhiwei, Chen Kaige, Jiang Junjun . Simulation of pipeline defect detection by the ultrasonic guided wave spectrum method[J]. Electronic Science and Technology, 2016,29(9):48-51.
[9]	Bareinboim E, Pearl J . Causal inference and the data-fusion problem[J]. Proceedings of the National Academy of Science, 2016,113(27):7345-7352. doi: 10.1073/pnas.1510507113
[10]	张延龙, 王俊勇 . 多传感器数据融合技术概述[J]. 舰船电子工程, 2013,33(2):41-44.
	Zhang Yanlong, Wang Junyong . Overview of the multi-sensor data fusion technology[J]. Ship Electronic Engineering, 2013,33(2):41-44.
[11]	周晨航 . 基于数据融合的运动目标识别方法研究[D]. 沈阳:沈阳大学, 2016.
	Zhou Chenhang . Research on moving target recognition method based on data fusion[D]. Shenyang:Shenyang University, 2016.
[12]	杨晓凤 . 掌纹和静脉特征融合算法的研究[D]. 北京:北京交通大学, 2017.
	Yang Xiaofeng . The research of palmprint and palm vein fusion algorithm based on feature-level[D]. Beijing:Beijing Jiaotong University 2017.
[13]	王凤姣 . 图像语义多特征的融合提取方法[J]. 电子技术与软件工程, 2016(19):89-92.
	Wang Fengjiao . Fusion extraction method of image semantic multiple features[J]. Electronic Technology and Software Engineering, 2016(19):89-92.
[14]	Hossain M S. On finding appropriate reject region in serial fusion based biometric verification[C]. Boston:IEEE International Multi-Disciplinary Conference on Cognitive Methods in Situation Awareness and Decision Support, 2016.
[15]	Gorai A K, Mitra G . A comparative study of the feed forward back propagation(FFBP) and layer recurrent(LR) neural network model for forecasting ground level ozone concentration[J]. Air Quality Atmosphere and Health, 2016(3):1-11.
[16]	Liang Z H, Wei L U . Ultrasound application of ultrasonic combined with back-propagation neural network in the diagnosis of central precocious puberty[J]. Journal of Medical Imaging, 2015(7):312-327.
[17]	陈进, 徐凯, 王学磊 , 等. 基于BPNN与DS理论的联合收割机监控系统设计[J]. 电子科技, 2016(12):152-155.
	Chen Jin, Xu Kai, Wang Xuelei . Monitoring system of combine harvester based on BPNN and DS theory[J]. Electronic Science and Technology, 2016(12):152-155.
[18]	宋寿鹏, 堵莹 . 基于MSMC管道缺陷超声检测信号压缩采样方法[J]. 电子科技, 2016,29(12):134-137.
	Song Shoupeng, Du Ying . Compressive sampling of pipeline defect ultrasonic testing signal based on MSMC[J]. Electronic Science and Technology, 2016,29(12):134-137.
[19]	王智军, 王建华 . 多特征融合的图像目标跟踪方法[J]. 电光与控制, 2017(11):49-52.
	Wang Zhijun, Wang Jianhua . A multi-feature fusion algorithm for moving target tracking of image sequences[J]. Electronics Optics and Control, 2017(11):49-52.
[20]	Dalai S, Jena M . Object recognition using higher order moments and comparative study between shape space and moment invariant method[J]. Esrsa Publications, 2014(3):89-95.

	圆度	偏心率	边界能量/×10⁷	梯度均值	矩形度
A	3.3	1.9	4.631 4	2.112 2	0.89
B	1.6	2.1	4.120 9	3.235 1	0.61
C	2.7	1.3	5.048 3	2.000 4	0.72

	1	2	3	4	5	6	7
A	1.744 8	5.987 2	22.734 3	21.146 3	43.980 9	26.631 3	43.087 0
B	1.770 7	7.127 7	22.590 6	19.756 4	41.930 1	23.631 3	41.486 1
C	1.774 7	7.135 8	24.359 5	21.473 3	45.403 4	25.345 7	44.945 1

	非可视化特征提取识别率	可视化特征提取识别率	串行融合算法识别率	并行融合算法识别率
A	66.7%	75%	91.7%	91.7%
B	66.7%	83.3%	91.7%	100%
C	75%	66.7%	83.3%	91.7%