Phase unwrapping via deep learning for surface shape measurement by using wavelength tuning interferometry.

Wavelength tuning interferometry is widely used in optical metrology in order to obtain the phase information of sample. The obtained wrapped phase usually unwraps in the range of [-π, π]. Therefore, phase unwrapping operation is required in order to obtain the right phase. But some factors, such as phase shift miscalibration, coupling error, and noise, always lower the precision of conventional phase shifting algorithm. To address such kind of problems, we proposed a deep learning method using deep convolutional neural network to perform phase unwrapping process by turning the task into a multiclass classification work and a 2N - 1 method for generating the training dataset. The results indicated that proposed method not only can compensate for phase shift miscalibration and coupling error, but also has strong robust and denoise ability, which means that proposed method is outperformed other conventional phase measurement algorithms.