Two-dimensional channel equalization model training method and two-dimensional channel equalization method

The invention discloses a two-dimensional channel equalization model training method and a two-dimensional channel equalization method, and belongs to the field of magnetic recording, and the method comprises the steps: building a two-dimensional channel equalization model based on a feedforward neural network, and carrying out the two-dimensional equalization of sub-data blocks in the two-dimensional channel equalization model according to a read-back data block, and obtaining balanced read-back information, wherein each hidden layer in the model adopts a nonlinear activation function; in a disk in which known data is written, obtaining bit sequences with the same length in a plurality of adjacent tracks to form read-back data blocks, obtaining corresponding write-back data blocks, takingthe read-back data blocks as feature information, taking sub-data blocks in the write-back data blocks as mark information, and forming a training sample by the mark information and the feature information; and after a train