Fourier transform profilometry based on convolution neural network

Fourier transform profilometry method has great value in high-speed three-dimensional shape measurement. In the method of Fourier transform profilometry, it is necessary to obtain the phase of the deformation fringe containing the height information of the object through Fourier transform, frequency domain filtering and inverse Fourier transform. Filtering in frequency domain is a very important and essential process. Filtering window is usually selected manually, which is inefficient and subjective. Too large filtering window can not filter useless information, and too small filtering window will lose the height information of the object. In this paper, an adaptive spectrum extraction method is used. In order to be more convenient and simple, this paper presents a method of frequency domain filtering based on convolution neural network. Convolution neural network can realize image recognition and image feature extraction. The proposed method uses convolution neural network to identify the carrier frequency components carrying the details of the object in the spectrum image. This paper introduces the theoretical analysis and the training process of convolution neural network. The adaptive spectrum extraction method and the convolution neural network method are compared. The method of spectrum extraction based on convolution neural network is feasible.