Invoices, as vouchers for transaction activities and the basic basis for accounting by financial departments, play a significant role in China's economic life. In traditional economic life, the management process of invoices is also cumbersome and complex, usually relying on manual review, which is inefficient and wastes human resources. To effectively improve the efficiency of invoice processing, this paper designs an intelligent invoice flow recognition and real-time abnormal alarm system based on deep learning (DL) technology. The system integrates optical character recognition (OCR) methods and constructs a recognition model that combines dense connected convolutional networks (DenseNet) and bidirectional long short-term memory networks (BiLSTM). This model eliminates the tedious steps of character decomposition and classification in traditional OCR. The system utilizes DenseNet to perform content recognition and extraction on preprocessed bill images, and then uses BiLSTM to generate text long and short sentences, improving the accuracy of content recognition. The results indicate that the system can not only accurately identify various types of invoice information to a certain extent, but also issue real-time alerts when abnormal bills are detected.
Chen et al. (Sun,) studied this question.