High-performance Embedded System Design for QR Code Recognition with Deep Learning

In the context of the Internet, QR code automatic identification technology is one of the key technologies. Aiming to address the issue that the typical QR code recognition system is sensitive to the influence of factors such as fuzzy and equipment, resulting in decreased recognition performance, we suggest a deep learning algorithm and Xilinx ZYNQ platform-based QR code recognition system solution in this research. The system created a recognition algorithm based on the YOLO v3 algorithm and an image deblurring algorithm based on the GAN model framework, which effectively removes Gaussian blur from QR code images by adding a relative entropy loss function. Additionally, hardware circuit design and Linux system porting are finished on ZYNQ side, and then the entire system installation and testing. The results demonstrate that the system outperforms the ARM platform of the experimental control, achieving a recognition rate of 85.9% on the complicated environment test set and a recognition speed of 93ms for a single image. The system also features cheap hardware costs and a great degree of flexibility to meet the performance standards for QR code recognition.