Research On Phase Unwrapping Algorithm Based On Deep Learning

Phase unwrapping is the key technology to obtain continuous phase information in optical phase measurement.The result of unwrapping will directly affect the accuracy of measurement,so phase unwrapping plays an important role in optical interference measurement.With the in-depth research of phase unwrapping algorithm,the unwrap algorithm has been improved,but there are still some problems in phase unwrapping demodulation with high noise.For example,the traditional phase unwrapping algorithm is not only low in precision,but also in large error,and it takes a long time.Even for some package phases,it can not be demodulated.In recent years,deep learning technology has been a great success in image processing,and it has been extended in the field of optical detection.Based on this,a method based on deep learning is proposed to study phase unwrapping,and the main contents are as follows:1.A semantic segmentation algorithm based on deep learning is proposed to demodulate the phase of the package.The method is based on the category of package phase,and then the corresponding model training is carried out,and then the demodulation can be realized by combining the phase and category of the package.The essence of this paper is to obtain the number of packages by the formula,which is the number of categories of semantic segmentation,which is generated by Zernike polynomial simulation.The core of the data is to train the phase of packages by using semantic segmentation model.2.Optimize network structure and improve ASPP module.The semantic segmentation model of deep learning is usually difficult to balance the accuracy and speed.Based on this paper,an improved lightweight neural network structure is proposed.On the basis of ensuring the speed,the accuracy is improved.At the same time,the expansion rate of ASPP module is fused and modified,which expands the sense field and further improves the segmentation accuracy of network structure.3.In view of the category imbalance of the phase of the package,a loss function which is commonly used in the detection of the target is proposed,which has a significant improvement in the accuracy of the phase segmentation.Then,the package phase and the number of categories are combined.According to the corresponding optimization algorithm,the high precision package phase can be obtained after demodulation.Finally,simulation and experimental results show that the deep learning method is not sensitive to high noise and has good robustness,simultaneous interpreting and demodulation with high accuracy and fast speed.