Paddle-OCR-Based Real-Time Online Recognition System for Steel Plate Slab Spray Marking Characters

In the process of steel plate slab production, it is necessary to identify the spray mark characters of the moving steel plates on the production line in real time. A real-time online recognition system based on paddle-OCR for steel slab spray mark characters is designed to address the security problems of manual recognition. The system adopts a high-sensitivity industrial camera to capture images, and its front end adds an optical light filter and a neutral density filter to solve the complex lighting environment and image sensor oversaturation problems in steel factories. The camera temperature control protection system is self-made to ensure that the camera can be used normally in the high-temperature environment of the steel factory. In terms of software, a four-thread synchronous online identification method is designed according to the system real-time requirements. Among them, the method of acquiring region of interest (ROI) images of the moving steel plate slab is designed, that is, for each piece of steel plate that enters the field of view, three frames of the complete character region ROI image are cropped. The Paddle-OCR character recognition algorithm is then used to recognize the three ROI images, and the result with the highest recognition rate is used as the output. The system has been operating stably in the steel factory for more than one year, recording its positive rate and detection rate on a weekly basis. Up to now, the minimum weekly positive rate is 91.2 % and the minimum weekly detection rate is 98.7 % , which is better than the requirements of the plant, and the software processing rate is 30 frames per second (fps), which meets the online identification requirements.