Research on Reading Recognition Algorithm of Industrial Instruments Based on Faster-RCNN

Abstract

Industrial instruments are widely used in military, aerospace, industry and other fields, especially in the harsh environment of high temperature, high voltage and high radiation such as substation. According to the classification of counting mode, industrial instruments can be divided into pointer instruments and digital instruments. This paper proposes a simultaneous reading recognition algorithm of pointer and digital multi and multi type instruments based on Faster-RCNN network model. Aiming at the problems of complex background, insensitive to small targets and low detection accuracy of industrial instrument reading recognition system, a detection method of industrial instrument based on Feature Pyramid Network (FPN) and Faster-RCNN network is proposed in this paper. In addition, in order to improve the recognition accuracy, an adaptive training data sampling algorithm is introduced to improve the effectiveness of positive and negative training sample extraction. Finally, the experimental results of industrial instrument reading recognition algorithm are systematically analyzed. The results show that the algorithm can be well applied to multi type industrial instrument reading recognition.