In order to improve the accuracy of meter reading detection and recognition, an improved target detection network is proposed based on the lightweight and efficient YOLOv5s network. Firstly, in the feature extraction stage, CBAM attention mechanism is added to learn the important features of the image, and a feature fusion network D-BiFPN is designed to enhance the extraction of deep features; Secondly, the CIOU loss function is introduced to make the regression of the target box more stable. The backbone network of CRNN text recognition algorithm is improved. The model maintains the characteristics of lightweight, and has a good prospect in the deployment of mobile terminals. Finally, the test on the meter data set shows that compared with the original YOLOv5 algorithm, the average accuracy of the proposed algorithm is improved by 5.13%; Compared with the original CRNN algorithm, the accuracy of the proposed text recognition algorithm is improved by 7.4%. The experimental results show that the proposed text detection algorithm has high detection accuracy and stability, and the text recognition algorithm has high accuracy and speed, which can meet the application requirements of meter reading recognition.