In order to improve the water company's management of residential water consumption, a digital water meter reading detection and recognition method based on deep neural network model is proposed. The feature extraction network is improved on the basis of the original DBNet network model, the feature extraction network ResNet network is replaced by the DPN double-path feature extraction network, the perspective transformation algorithm is used to correct the inclined reading area image, the YOLOv5s network model is used to recognize a plurality of digits in the reading area; The reading results are displayed by the water meter recognition application system, and the function of real-time detection and recognition of water meter images is completed. The experimental results show that the YOLOv5s network model has good robustness, and can accurately identify the occlusion, blur and double-half digits, and the recognition accuracy is as high as 99.1%.