License plate recognition method based on improved U-Net in complex environment

The invention belongs to the technical field of traffic management systems, and particularly relates to a license plate recognition method based on improved U-Net in a complex environment. The method comprises the following steps: a neural network preprocessing part and a license plate detection and recognition part. The neural network preprocessing part is divided into four steps: A1, making a data set; a2, designing a neural network; step A3, iterative optimization is carried out; and A4, generating a model. The license plate detection and recognition part comprises the following six steps: step B1, acquiring an image; b2, carrying out binarization processing; step B3, license plate positioning; step B4, character segmentation; step B5, carrying out improved U-Net neural network processing; and B6, obtaining a license plate detection result. The method aims at a license plate recognition scheme in a complex environment under special weather such as haze, rain and sand blowing. A traditional image segmentati