The invention is suitable for the field of vehicle detection, and provides a vehicle detection method based on deep learning, and the method comprises the following steps: S1, obtaining a plurality ofimages containing vehicles through a video stream, carrying out the manual marking of the images, and carrying out the dividing of the images into a training set and a verification set of a detectionmodel according to a proportion; S2, constructing a target detection PDN network based on the DDN; S3, performing model training on the target detection PDN network by using a training set, and selecting a model with an optimal performance of a verification set; And S4, on the basis of the optimal model, performing prediction on the GPU server to realize target detection on the video stream. Thetechnical problems that in the prior art, in actual target detection, a video is usually processed, an algorithm is needed to carry out target detection in real time, a small part of researchers' methods meet the requirements on speed, however, with the cost of sacrificing precision, the precision is low, an application cannot fall to the ground, and a large amount of missed detection and false detection cannot be accepted are solved.