福州大学学报（自然科学版）2025，Vol.53Issue(4)：414-421,8.DOI:10.7631/issn.1000-2243.24269
结合轻量化网络和图像处理的非接触振动测量
Non-contact vibration measurement combining lightweight networks and image processing
摘要
Abstract
To address the challenges of vibration measurement in complex environments and resource-constrained conditions,a non-contact measurement method combining image processing technology and lightweight convolutional neural networks is proposed.A lightweight neural network model is constructed using an inverted residual block based on depthwise separable convolutions.To enhance prediction accuracy,the efficient channel attention(EC A)mechanism and Mish activation function are intro-duced.Additionally,Canny edge detection and the Hough transform are used to extract image features,replacing the spatial attention mechanism to optimize processing efficiency and improve accu-racy.Experimental results demonstrate that the proposed non-contact measurement method achieves a measurement error of less than 0.2％within the vibration frequency range from 5 to 50 Hz,verifying its effectiveness and reliability.关键词
图像处理/轻量化卷积神经网络/非接触式振动测量Key words
image processing/light weight convolutional neural network/non-contact vibration measurement分类
机械制造引用本文复制引用
周梓权,杨晓翔..结合轻量化网络和图像处理的非接触振动测量[J].福州大学学报（自然科学版）,2025,53(4):414-421,8.基金项目
国家自然科学基金资助项目(12272095) （12272095）
