Phase unwrapping method based on compression and excitation neural network

The invention discloses a digital unwrapping method based on a compression and excitation neural network, and the method is characterized in that on the basis of a conventional U-Net deep learning phase unwrapping method, a residual block containing an attention mechanism (SE) is added behind a convolution block of each layer, and the digital unwrapping method based on the compression and excitation neural network is formed. The specific process is as follows: two branches are output after passing through a residual block, one branch is not operated, the other branch firstly compresses the size into 1 * 1 through an average pooling layer, then each channel value is normalized to be between [0, 1] through a Sig-moid activation function after passing through a full connection layer FC - 1, a Leaky Rule activation function and a full connection layer FC - 2, and finally the two branches are added. The full connection layer can emphasize the global features of the image, so that the compression and excitation mod