The continuous development and intelligence of the power grid system have brought huge measurement needs and smart meters are widely laid as the main measurement equipment. However, the information of the meters carried by smart meters of different brands, models and batches is also very different. Non-intelligent artificial information collection has seriously hindered the upgrading and development of measurement infrastructure and collection security, and restricted the quality and level of power asset management. In this paper, the text recognition technology is applied to the information collection process of smart meters. Design a two-stage system to detect and identify the text information in the photos of the meters, which realizes intelligent collection of meter information and improves the efficiency and safety. The two-stage system in this paper includes a text detection module and a text recognition module. The text detection module detects the text position in the meter picture through the improved PSENet network, and the text recognition module recognizes the detected text box through the CRNN network. The algorithm itself is not constrained by the quality of the input image and the scene, and it has strong anti-interference ability to the problems of different font sizes, too high or too low exposure for text detection and recognition in smart meters. English and numbers have high recognition accuracy. And the recognition accuracy for Chinese characters, English and numbers in the meter picture is very high.