Aiming at the problems of voucher preparation errors, numerous verification rules, lengthy audit cycle and low efficiency in the process of voucher preparation and audit of financial reimbursement in Colleges and universities, based on the implementation process of financial reimbursement process, this paper designs a framework of financial original voucher classification and verification system based on deep learning, and proposes a deep learning algorithm to identify the key contents of financial original vouchers, so as to support the voucher preparation process of financial reimbursement personnel Review and verification process of financial personnel. By testing the existing financial reimbursement process, the original voucher can be accurately classified according to the requirements of the national tax system for the classification of goods, for the identification of the type of daily reimbursement; based on the preprocessing of bills, FRCNN algorithms in deep learning can be applied to identify the key contents of bills.