The work aims to study an improved algorithm to obtain higher digital recognition effect, improve the recognition efficiency of the bill digital and strengthen the equipment performance of the bank's intelligent handling business. Characters were extracted and divided according to the printed digital characteristics of bank notes. After image acquisition, noise reduction and binarization, the starting point histogram method was combined with the step size method for character segmentation, and then the improved LENET convolutional neural network was used to extract and classify digital features. Through experiments, the results showed that the proposed method can perform digital recognition in complex environments with an accuracy of more than 95%, and the recognition rate was 1.169 s/sheet. The new character segmentation algorithm combined with the improved LENET neural network can identify highly sensitive printed tickets with high accuracy.