A novel STRSC-UNet image denoising method for computer vision enhanced RFID multi-tag measurement and localization

Abstract Most of the traditional radio frequency identification (RFID) multi-tag realizes identification by using electromagnetic communication. However, with the increasing number of RFID multi-tags and the increasing complexity of electromagnetic interference in RFID application environments, only using traditional electromagnetic approach to realize RFID multi-tag localization cannot meet RFID application requirements. Therefore, in this paper, a novel Swin Transformer fusing residual shrinkage network and convolution integrated in U-Shaped network (STRSC-UNet) image denoising method for computer vision enhanced RFID multi-tag measurement and localization is proposed. Firstly, an RFID multi-tag localization system embedded with computer vision is designed for RFID tag image acquisition. Secondly, a novel STRSC-UNet denoising method, which introduces the Transformer method in high-level vision into the field of low-level vision, is proposed for tag image denoising. Subsequently, based on the denoised images, scale-invariant feature transform matching algorithm is used for tag recognition to obtain homonymous image points. Finally, the spatial forward intersection method is used to realize RFID multi-tag localization. Four different baseline models are used to compare with the proposed denoising model. The experimental results show that compared with other benchmark methods, the image quality of STRSC-UNet is improved by at least 0.17 dB. Further localization experimental results demonstrate that, by using the proposed STRSC-UNet denoising method, the localization accuracy of RFID multi-tag system can be effectively enhanced.