Abstract: There are many manufacturers and models of dismantled electric meters, so manual sorting and data entry is faced with the problem of low efficiency and difficult to guarantee accuracy. To save this problem, an extraction method for electric meters’ heterogeneous information based on improved correlation coefficient neural network is proposed. Firstly, the connectionist text proposal network (CTPN) is used to locate the model text. Then, the densely connected convolutional network (DenseNet) and the connectionist temporal classification (CTC) are used to recognize the model text image and get model’s preliminary recognition results. Secondly, the you only look once v4-tiny (YOLOv4-Tiny) is used to detect the logo and get manufacturer’s preliminary recognition results. Thirdly, testing on the validation set to get appropriate adaptive correlation coefficient. Finally, based on the information correlation between electric meters’ model and manufacturer and their preliminary recognition results, the correlative identification is carried out to extract the electric meters’ heterogeneous information. The results of the experiments show that the improved correlation coefficient neural network can effectively extract the dismantled electric meters’ model and manufacturer information. The accuracy of the method could reach 98.71%, and the average time cost to extract a single electric meter’s information is 0.406s。Compared with other mainstream text recognition and object detection methods, the proposed method improves the accuracy of information extraction for dismantled electric meters, and helps realize automatic entry and filing of dismantled electric meters’ information.