Multimode fiber speckle image reconstruction method based on deep learning

The invention discloses a multimode optical fiber speckle image reconstruction method based on deep learning. The method comprises the following steps: S1, acquiring a speckle image of an MNIST handwritten body after the MNIST handwritten body is subjected to multimode optical fiber mode dispersion by using an optical experiment system; s2, the obtained speckle images are divided into a training set and a test set, 2000 speckles are used for training, and 200 speckles are used for testing; s3, replacing a convolutional layer in a U-net network framework with a multi-scale convolutional network module, and embedding a multi-dimensional attention mechanism module into a network model decoding layer; s4, putting the data set in a designed multi-scale convolution and multi-dimensional attention mechanism network for training to obtain a training model; and S5, performing image reconstruction on the verification set by using the trained model. According to the invention, the reconstruction quality of the multimode