摘要
数字全息术可以对粗糙物体的形变进行精确测量,但由于激光的高相干性,测量粗糙表面时会产生大量的散斑噪声,严重影响测量精度。为了提高测量精度,本文提出了一种基于多尺度空间注意力的卷积神经网络去噪方法,添加了不同尺度的空间注意力机制,对提取的相位信息进行去噪处理,有效降低了空间分布不一致的散斑噪声的干扰,并通过数值仿真和焊点热变形实验验证了该方法的有效性。
Digital holography can be used to measure the deformation of rough object. However, due to the high coherence of laser, a lot of speckle noise will be generated when measuring rough surface, which seriously affects the measurement accuracy.In this paper, a convolutional neural network based on multi-scale spatial attention denoising method is proposed to improve the accuracy of measurement. Through the multi-scale spatial attention mechanism, the noise of different scales and different regions in space is processed, and the interference of speckle noise with inconsistent spatial distribution is effectively reduced, and the effectiveness of the method is verified by simulation experiments and thermal deformation experiments of solder joints.
作者
周孟航
赵自新
杨兴宇
杜怡君
ZHOU Meng-hang;ZHAO Zi-xin;YANG Xing-yu;DU Yi-jun(State Key Laboratory for Manufacturing Systems Engineering,School of Mechanical Engineering,Xi’an Jiaotong University,Xi’an,Shaanxi,710049,China)
出处
《强度与环境》
CSCD
2022年第5期170-177,共8页
Structure & Environment Engineering
基金
国家自然科学基金(52175516)。
关键词
数字全息
去噪
多尺度空间注意力
卷积神经网络
Digital holography
Denoising
Multi-scale spatial attention
Convolutional neural network
