“Technology News Broadcast: In the field of point diffraction interferometry, an innovative phase unwrapping method has attracted widespread attention. This method is based on dilated spatial convolution, combining the autoencoder structure with dilated spatial convolution to achieve high-precision, efficient, and anti-interference detection of wrapped phase images. The research team has created a large and diverse dataset for training and optimizing this algorithm, enabling it to accurately identify the order of package phase and quickly process high-precision unpacking results. Compared with traditional unpacking algorithms, this method demonstrates significant advantages in both time and accuracy. The average time required to process an image is only 0.035 seconds, far lower than the traditional method's more than 1 second. Meanwhile, the experimental results show that the unpacking results of this method are similar to those of the branch cutting method used in professional interferometric image processing software, with a root mean square error value of 0.0222 rad. The peak valley difference of the surface fitting results is only 0.012 1 λ, and the root mean square difference is also only 0.004 2 λ. This breakthrough research provides a new feasible solution for high-precision phase unwrapping of point diffraction interference image processing, which is expected to play an important role in related fields.”