A pointer meter recognition method based on virtual sample generation technology

At present, the pointer meter recognition methods utilize the traditional image processing techniques. Such techniques are complex, unstable, and unable to meet the requirements of real-time recognition. In order to address these problems, a recognition model based on convolutional neural networks is proposed in this study. However, it is difficult to obtain a large number of real instrument images to train the recognition model. In this paper, a novel virtual sample generation technology is proposed to generate a large number of images from a small number of real instrument images to train the recognition model. The proposed method does not require to pre-process the original images that are used in the trained pointer meter recognition model, just like end-to-end recognition model. The simulation data, the testbed data, and the engineering application show that the proposed method performs better than the compared methods under the interference of illumination and other complex application scenarios.