Blood stream imaging method and device based on U-net neural network

The invention discloses a blood stream imaging method and device based on a U-net neural network. The blood stream imaging device comprises a low-coherent light source, a CMOS camera, a telecentric lens and a to-be-tested sample, wherein the low-coherent light source and the to-be-tested sample are sequentially located in an illuminating light path, and the to-be-tested sample, the CMOS camera andthe telecentric lens are sequentially located in an imaging light path. The low-coherent light source is used to provide illumination, the CMOS camera is used to continuously acquire images, then, the acquired images are trained with the U-net neural network, a trained network model predicts each pixel of new original images, a predicted probability value is used as a gray value of a 2D blood steam distribution diagram for image reconstruction to realize contrast imaging of blood streams, imaging can be realized with a few original images only, the temporal resolution is improved, meanwhile,large and small blood vess