In the study of detection and recognition of distorted, skewed, and other deformation multi-QR codes using the YOLO_CBAM algorithm, there were problems that the detection effect was poor in the case of stacked QR codes, and the small resolution QR codes couldn't be well recognized. In this paper, we used the Real-ESRGAN model for super-resolution reconstruction of polymorphic QR code images. Combining the characteristics of the multi-QR code dataset, we used 3 convolutional layers to reduce the complexity of image super-resolution processing and replaced the LEAKYRELU activation function in the discriminative network with the GELU function. Then we conducted experiments using the SR_ESAGAN algorithm. Experiments showed that after the improvement, the detection rate and recognition rate of multiple QR codes with deformations were enhanced by 3.71% and 5.27%, respectively, compared to those using only the YOLO_CBAM algorithm.