The clear license plate recognition algorithm has already become a mature technology, but in terms of fuzzy license plate unable to be identified by human eyes, the recognition rate of traditional license plate recognition algorithm is still low or could not be identified at all. In view of this, a kind of license plate character recognition algorithm based on convolution neural network was proposed. 9 720 fuzzy character training samples were made, and 8 748 images were trained for convolutional neural network. The blind segmentation of fuzzy license plate characters was realized, and a trained convoluted neural network was used to recognize the characters after blind segmentation. The experimental results showed that the recognition rate of the algorithm reached about 99.17% for the training set, and the recognition rate of the test set to be about 93.32%, indicating that the algorithm was robust for fuzzy license plate recognition and could be applied to various scenes.