Abstract: During the operation period of the distributed control system（DCS） of nuclear power plant the I/O module of the DCS needs to be retested and calibrated periodically due to the deterioration of the performance of electronic devices and the impact of time drift. In view of the traditional DCS I/O test coverage is insufficient, the test efficiency is low and is of human factors, this paper proposes an online testing method of the optical character recognition technology based on convolutional neural network applied in the DCS system. Through the simulation equipment and video acquisition equipment to complete the automatic switching of the picture, and the picture information is read, the specific picture is intercepted after the image preprocessing, and then the OCR recognition model is used to identify the picture content, the recognition results and the expected value of the comparison, so as to realize the automatic test. The test results show that after CNN training, the recognition rate of screen characters of display control equipment can reach 100%. This method can break through the barrier of equipment manufacturer’s proprietary communication protocol, which can effectively reduce operator’s human error and improve the test efficiency and the economy of nuclear power plant.