Phase unwrapping in digital holography based on the Dv3p-IRA

The accurate acquisition of the phase information is crucial for the three-dimensional shape reconstruction of an object. However, the inverse tangent operation in the phase extraction process will inevitably cause the phase wrapping phenomenon, resulting in phase discontinuity. Therefore, a digital holographic phase unwrapping method based on Deeplabv3plus-Inverted-Residual-Attention (Dv3p-IRA) is proposed in this paper. This method takes Deeplabv3+ as the basic framework and adopts the encoder-decoder structure: the encoder achieves multi-scale feature extraction through dense block and convolutional block attention module-atrous spatial pyramid pooling (CBAM-ASPP), while the decoder achieves phase reconstruction through cross-layer feature fusion and up-sampling. Random matrix enlargement (RME) and Gaussian function superposition (GFS) methods are used to construct the dataset, cross-entropy loss and Dice loss are integrated as the loss function to optimize the network parameters, and a reflective off-axis digital holographic optical path system is constructed for experimental verification. The simulation and experimental results show that compared with other methods, the Dv3p-IRA achieves higher accuracy, with smaller fluctuations in the error value range and more stable model performance. In addition, it can effectively realize the separation of object information and background, and the reconstructed phase shape has good smoothness and continuity. Therefore, the proposed method not only realizes high-precision phase recovery, but also effectively deals with the holograms with speckle noise, and shows significant advantages in phase region segmentation and noise robustness.