The invention discloses a pedestrian refined identification method and device based on deep learning. The method comprises the steps that a pedestrian image database is constructed, a training dataset comprising pedestrian characteristics is created by means of the pedestrian image database, and the training dataset is input into an SSD network for training; a pedestrian refined image database is constructed according to training results, a training dataset comprising characteristics of all parts of pedestrians is created, and the training dataset is input into a convolutional neural network for training; the pedestrian characteristics in a real-time video stream coming from a streaming media server are extracted by means of the trained SSD network, and the characteristics of all the parts of the pedestrians are extracted by means of the trained convolutional neural network according to the extracted pedestrian characteristics; the extracted characteristics of all the parts of the pedestrians are matched with the characteristics of the corresponding parts of each pedestrian of the pedestrian refined image database, the matching results are combined to obtain the pedestrian refined identification result. By means of the pedestrian refined identification method and device based on deep learning, the accuracy of pedestrian refined identification is effectively improved.