License plate character recognition method based on double-circulation transcription network

The invention discloses a license plate character recognition method based on a double-circulation transcription network. The method is used for recognizing unbalanced Chinese characters, English characters and numeric characters in a license plate image. The method comprises the following steps: license plate image features are extracted through a convolutional neural network; then, two Bi-LSTMswhich are arranged in parallel are built in the double-circulation transcription network; two Bi-LSTMs arranged in parallel respectively perform feature calculation on Chinese characters, English characters and numeric characters to obtain a license plate character confidence estimation sequence, and finally, the license plate character confidence estimation sequence is mapped through a transcription layer to obtain a prediction label. The license plate character recognition method successfully solves the problem of training sample imbalance in license plate character recognition, and improvesthe accuracy of the licen