基于深度学习的指针式仪表自动识别的研究和设计

With the continuous development of technology and the continuous improvement of industrial informationization and digitization, it is especially important to carry out efficient and accurate data entry for traditional pointer instruments in industrial production. Aiming at the problem of the demanding environment and the single type of instrument recognition in the current automatic recognition system, this paper combines the related techniques of deep learning and computer vision to research and improve the existing process identification algorithm.By constructing the instrument training data set and learning and adjusting the target detection model MASKRCNN to realize the image segmentation and effective information extraction of the instrument panel in the natural scene.According to the design of image feature extraction, this paper use the Ostu threshold segmentation method and KNN to identify the instrument numbers, use the probabilistic Hough line method to fit and locate the pointer, and use the distance method to determine the indication at last in order to improving the robustness and generalization of the pointer type automatic identification system.