Deep Learning-based Intelligent Reading Recognition Method of the Digital Multimeter

Abstract

Digital multimeter, as a multi-purpose electronic measuring instrument, has basic fault diagnosis function and is widely used in laboratories and industries. In order to solve the problems of manual calibration wasting manpower, manual reading is error-prone, and to improve the accuracy and robustness of automatic reading recognition of digital meters. In this paper, a deep learning-based reading recognition method of the digital multimeter is proposed. Firstly, the YOLO target detection method is used to extract the digital display area of multimeter. Then, the CNN classification method is used to identify 0~9 numbers, negative signs and blanks in the reading area. Simultaneously, the connected domain method is adopted to locate the decimal point based on its location characteristics. Finally, the complete reading information can be obtained by merging the reading recognition result and the decimal point positioning result. Experiments were conducted and the results shown that the overall characters recognition accuracy reaches 99.85%. Besides, only one error occurred in the final test with an accuracy of 98%. Meanwhile, this proposed method can adapt to various complex environments, and meet the real-time requirements. Specifically, compared with the traditional method, the proposed modified deep learning method has high robustness and practicability.