The invention discloses a
license plate recognition method based on
sequence learning, and belongs to the technical field of intelligent traffic. A
license plate image is changed into a sequence composed of candidate characters through a deep neural network used for character detection and recognition, then the sequence is input into a sequence conversion model obtained through training, a new character index sequence is obtained, and finally original candidate characters are recombined according to indexes to obtain a
license plate recognition result. According to the license plate recognition method based on
sequence learning obtained through the technology, a
machine autonomously learns license plate character selection, rejection and permutation combination rules, the pressure of manually designing
heuristic rules is reduced, learning experience is achieved with the help of a large number of data samples, and a
sequence model can be more flexible in the aspect of the recognition problem of multi-standard license plates. Compared with a traditional
sequence processing method, such as hidden Markov, the method has the advantages that the connection of the sequence elements can be found in long-time dependence, and a better sequence conversion result is obtained.