非扫描激光主动成像的频域特征分析  

Frequency domain feature analysis of non-scanning laser active imaging

作　　者：陈红 高原 谢勤岚[2] CHEN Hong;GAO Yuan;XIE Qinlan(Experimental Teaching And Laboratory Management Center,South-Central University for Nationalities,Wuhan 430074,China;School of Biomedical Engineering,South-Central University for Nationalities,Wuhan 430074,China)

机构地区：[1]中南民族大学实验教学与实验室管理中心,武汉430074 [2]中南民族大学生物医学工程学院,武汉430074

出　　处：《激光杂志》2022年第1期139-143,共5页Laser Journal

基　　金：国家民委高等教育研究项(No.19011)。

摘　　要：非扫描激光主动成像过程中的局部频域特征会出现噪声,为了提高非扫描激光主动成像的频域特征质量,提出了非扫描激光主动成像的频域特征分析。基于非扫描激光主动成像原理,引入神经网络—极限学习机,消除非扫描激光主动成像的频域特征样本训练误差,使得神经网络—极限学习机与最小二乘解的问题相同,识别出非扫描激光主动成像的频域特征,通过描述频域特征的尺度空间,定义了卷积处理的高斯核函数,利用梯度模值和主方向两个指标,提取非扫描激光主动成像的频域特征,通过设计非扫描激光主动成像的频域特征滤波去噪算法,提升非扫描激光主动成像的频域特征质量。实验结果表明,在非扫描激光主动成像的频域特征分析过程中发现,在30次迭代过程中,频域特征分析时间最长仅为2.45 min,频域特征信噪比一直在80 dB~93 dB之间波动,非扫描激光主动成像的频域特征滤波去噪算法可以缩短频域特征分析时间,还可以提高频域特征质量。In order to improve the quality of frequency domain features of non scanning laser active imaging, the frequency domain feature analysis of non scanning laser active imaging is proposed. Based on the principle of non scanning laser active imaging, the neural network extreme learning machine is introduced to eliminate the training error of frequency domain feature samples of non scanning laser active imaging, so that the problem of neural network extreme learning machine and least square solution is the same, and the frequency domain feature of non scanning laser active imaging is identified. By describing the scale space of frequency domain feature, the Gaussian function of convolution processing is defined Kernel function is used to extract the frequency domain features of non scanning laser active imaging by using gradient mode value and principal direction. The frequency domain feature filtering denoising algorithm of non scanning laser active imaging is designed to improve the frequency domain feature quality of non scanning laser active imaging. The experimental results show that in the process of frequency domain feature analysis of non scanning laser active imaging, the time of frequency domain feature analysis is only 2.45 min, and the signal-to-noise ratio of frequency domain feature fluctuates between 80~93 dB. The frequency domain feature filtering denoising algorithm of non scanning laser active imaging can shorten the time of frequency domain feature analysis and improve the quality of frequency domain feature.

关 键 词：非扫描激光 主动成像 频域特征 滤波去噪 

分 类 号：TN957[电子电信—信号与信息处理]