Zhixiong Yang

Zhixiong Yang

Auto-generated profile, based on publicly available information

Zhixiong Yang

Xinjiang University

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
15 November 2023 in IEEE Access
Verified
Journal Article
Open access
HXHuanyuan Xu
Huanyuan Xu
YZYajun Zhang
Yajun Zhang
ZYZhixiong Yang
Zhixiong Yang
HYHaoqiang Yan
Haoqiang Yan
XWXingqiang Wang
Xingqiang Wang
Deaf people use sign language for communication, which relies on the movement gestures of body parts and plays a vital role in human-computer interaction. Most wireless sensing-based gesture recognition studies have recognized simple gestures but overlooked the recognition of complex activities, such as sign language. In addition, cross-domain recognition often requires a large amount of data to train classifiers for each environment. Therefore, we propose RF-CSign, which aims to achieve high accuracy in sign language recognition and cross-domain recognition. First, we use Radio Frequency Identification (RFID) to collect signals and obtain denoised signals through data pre-processing, so that they can be processed in a neural network. Second, the RF-CSign network is proposed with the inclusion of large kernel convolution to reduce the complexity of the model and to make the model with long-range correlations, thereby enhancing recognition accuracy. Third, RF-CSign employs a pixel Normalization-based Attention Module (NAM) to enhance the stability of the model, thereby addressing the problem of model overfitting. Finally, RF-CSign achieves high accuracy in cross-domain environments through a migration learning approach. The experimental results showed that the average recognition accuracy of RF-CSign reached 99.17%, and the average recognition accuracy for new users and new environments recorded 96.67% and 97.50%, respectively.

6 October 2023 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
ZYZhixiong Yang
Zhixiong Yang
ZZZiyi Zhen
Ziyi Zhen
ZLZijian Li
Zijian Li
XLXu Liu
Xu Liu
BYBo Yuan
Bo Yuan
Yajun ZhangYajun Zhang
Yajun Zhang
Gesture recognition is the basis of human-computer interaction (HCI), and RFID has attracted wide attention for its advantages of lightweight, low cost, and universality. We propose RF-CGR, which implements cross-domain Chinese character gesture recognition intuitively and effectively. The fundamental concept of RF-CGR is to extract hand motion patterns from RFID phase information and display distinctive gesture features. Firstly, we build a signal sensing model to capture gesture information effectively. Secondly, we innovatively convert the Chinese character phase signal into an intuitive picture representation and use the gesture movement change pattern as input through data pre-processing operations. Finally, Modify-Visual Geometry Group Network (M-VGG) model is used for cross-domain gesture recognition. The M-VGG network learns high-level semantic features of gestures through spatiotemporal convolutions to extract cross-domain gesture features. Compared with other models, M-VGG has fewer parameters, and shorter training time, while still achieving high-accuracy gesture recognition. We evaluate its performance in three scenarios. The accuracy is 98.75% and 98% for new users and new scenarios respectively, significantly outperforming existing wireless signal-based gesture recognition methods.

15 November 2021 in IEEE Access
Verified
Journal Article
Open access
ZYZhixiong Yang
Zhixiong Yang
XLXu Liu
Xu Liu
ZLZijian Li
Zijian Li
BYBo Yuan
Bo Yuan
Yajun ZhangYajun Zhang
Yajun Zhang
As the basis of human-computer interaction (HCI), gesture recognition interprets user-performed gestures as commands, followed by the content execution expressed by users’ gestures. Gesture recognition through wireless signals denotes a novel branch of human perception. Despite the recent popularity of Radio Frequency Identification (RFID) following specific advantages (lightweight, low-cost, and universality), several intricacies remain unresolved in RFID sensing research. First, most studies performed simplified body movements assessments instead of identifying complex and fine-grained or subtle gestures. Second, users require extensive training in a novel discipline to collect training data in a specific pattern. Given the paucity of an intuitive and effective means of identifying user gestures, the RF-E-letter proposed in this study denotes an RFID recognition system for complex, fine-grained, and domain-independent gestures. A multi-label array was utilized to gather gesture signals. Fine-grained gesture data could be obtained pre-processing with a novel data-processing method. Seemingly irregular RFID phase data could be converted into intuitive images for the deep learning module input as convolutional neural networks (CNNs) encompass automatic extraction characteristics for complex space-time features. The average accuracy of new environments for novel users is 95.6% and 96.6%, respectively (significantly better than current RFID-based solutions), thus demonstrating effectiveness and versatility.

24 October 2024 in IEEE Transactions on Cognitive Communications and Networking
Verified
Journal Article
Closed access
ZYZhixiong Yang
Zhixiong Yang
ZZZiyi Zhen
Ziyi Zhen
HXHui Xu
Hui Xu
Yajun ZhangYajun Zhang
Yajun Zhang
XFXinlong Feng
Xinlong Feng
Continuous user identification could facilitate large-scale identity-based services, potentially including access control, security management, personalized services, and beyond. Although current RFID-based user identification systems demonstrate effective performance in single-user scenarios, they exhibit a lack of robustness and accuracy when extended to multi-user environments. In this paper, we propose RF-UI, a low-cost continuous user identification system that can tolerate different interference factors (e.g., appearance changes, inconsistent walking paths). The intuition underlying our design is that when multiple users traverse the radio gate sequentially, the received signal is dominated by the user traversing the radio gate. We develop an algorithm that utilizes phase energy fluctuation to separate signals from different users and extract valid gait-related patterns by applying neighborhood energy sliding windows. Then, we construct a Joint Similarity Matrix (JSM) for characterizing gait features that are robust against various interference factors. Finally, RF-UI achieves cost-effective data augmentation through the deployment of only a few additional tags. Extensive experiments show that RF-UI achieved an accuracy of 94% under various interference factors and maintained a high accuracy of 92.8% in continuous user identification.

5 March 2025 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
ZYZhixiong Yang
Zhixiong Yang
ZZZiyi Zhen
Ziyi Zhen
HXHui Xu
Hui Xu
XFXinlong Feng
Xinlong Feng
Camera-based air handwriting recognition can enable a new form of user-computer interface. For a practically viable air handwriting recognition system, real-time responsiveness, robustness against environmental and human-induced interferences, and coherent handwriting recognition are essential. However, existing systems fall short of fulfilling these requirements simultaneously. To this end, we propose K-AirWrite, a real-time, robust system designed for continuous air handwriting recognition. The key enabler is to simplify complex video stream recognition into a single-frame trajectory image recognition problem by mapping hand motion in the video stream to hand trajectory coordinate images and enhancing real-time performance through a lightweight detection network. Furthermore, we develop a segmentation method that leverages coordinate changes induced by the inherent pattern during the character-switching process to segment continuous characters. These segmented characters are then sequentially input into the M-RepVGG network for recognition. Extensive experiments demonstrate that K-AirWrite is remarkably efficient and exhibits robust generalization capabilities, achieving an accuracy rate of 98.2% for individual characters and 96.4% for continuous characters.

15 April 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
FLFengrong Liu
Fengrong Liu
SCShuli Cheng
Shuli Cheng
ZYZhixiong Yang
Zhixiong Yang
ADAnyu Du
Anyu Du
HHHaoxiang Huang
Haoxiang Huang
RHRuimin Han
Ruimin Han
Hyperspectral image classification (HSIC) serves as a fundamental task in remote sensing, enabling fine-grained and detailed interpretation of land-cover characteristics. However, substantial spectral redundancy and spatial heterogeneity make it difficult to learn discriminative spectral--spatial representations. Most existing methods exploit first-order spectral information and isotropic spatial interactions, which are insufficient for characterizing high-order inter-channel dependencies and direction-sensitive spatial structures. To overcome these issues, we propose a Manifold-Aware Spectral Covariance and Directional Anchor Network (MSCANet), which integrates second-order Riemannian-aware spectral modeling with anisotropic global dependency learning. Specifically, we first propose the Cascaded Riemannian Covariance Enhancement (CRCE) module to construct compact symmetric positive definite (SPD) descriptors from grouped spectral features. Through low-rank modeling and geometry-consistent refinement on the SPD manifold, CRCE effectively exploits intrinsic second-order correlations to amplify spectral separability, while mitigating the cubic channel-wise complexity. Second, we introduce a Spatial-Spectral Anisotropic Gaussian Anchor Attention (S2AGA), which extracts content-adaptive anchors via spatial elliptical kernels and channel-wise anisotropy to reconstruct global dependencies through degree-calibrated Markov restoration, scaling efficiently with O(NK) complexity. Experimental results on four benchmark datasets (IP, SA, WHU-HC, and WHU-HH) reveal that MSCANet achieves state-of-the-art performance, yielding overall accuracies of 97.81%, 98.50%, 98.12%, and 98.71%, respectively, while maintaining fewer parameters and lower computational cost.