Abstract: In order to realize the control and management of steel structure and power equipment in the substation engineering construction, it is necessary to identify the relevant information from the title bar of a large number of steel structure drawings, and subsequently contrast them with the real structures. To deal with the blurriness of word, diversity of table and confusion of information, a deep learning method combining the CNN+RNN text detection model and the CRNN character recognition model is being proposed. Carrying out the detection and recognition experiments in the existing data set of steel structures, the detection precision reaches over 80% and the recognition accuracy reaches over 90%, which is superior to other detection and recognition methods. The results of the engineering application show that this method can effectively reduce the difficulty in feature extraction caused by the differences in arrangement, size, font and color of text, which can improve the accuracy of text recognition in title bar of steel structure drawings of substations.