Jinyi Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Jinyi Liu has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Bin Chen
Yunnan University
1 shared publication - Dan Wu
School of Computer Science, Peking University, Beijing, China
1 shared publication - Daqing Zhang
Peking University
1 shared publication - Fusang Zhang
Beihang University, Beijing, China
1 shared publication Mujiangshan Wang
1 shared publication- Pingping Pan
Guangdong Laboratory of Artificial Intelligence and Digital Economy, Shenzhen, China
1 shared publication
Latest Publications
- Open Access
- 19 March 2026
- Published by Elsevier in SSRN Electronic Journal
DeepFT: Deep Unfolding Network for Super-Resolution Spectrum Estimation
Spectral analysis is fundamental to numerous industrial applications, including motor fault diagnosis, power quality monitoring, and vibration analysis. However, conventional fast Fourier transform (FFT) suffers from limited frequency resolution constrained by the Rayleigh limit, making it challenging to resolve closely spaced frequency components. This paper proposes DeepFT, a novel deep unfolding network that achieves super-resolution spectrum estimation while preserving computational efficiency. The key innovations include a learnable complex-valued transform matrix initialized from the discrete Fourier transform, enabling adaptive frequency analysis; full complex domain iterative processing that preserves phase information throughout the network, enhancing the ability to separate closely spaced frequencies; and an encoder-decoder-sparse module with learnable soft-thresholding for effective sparse signal recovery. Extensive experiments demonstrate that DeepFT achieves frequency resolution of 0.3/N, where N denotes the signal length, significantly surpassing the Rayleigh limit of 1/N. The proposed method shows robust performance under various signal-to-noise ratio (SNR) conditions and successfully resolves multiple frequency components in complex scenarios. The lightweight architecture with 649K parameters enables real time deployment in industrial embedded systems.
- 19 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDeepFT: Deep Unfolding Network for Super-Resolution Spectrum Estimation
Spectral analysis is fundamental to numerous industrial applications, including motor fault diagnosis, power quality monitoring, and vibration analysis. However, conventional fast Fourier transform (FFT) suffers from limited frequency resolution constrained by the Rayleigh limit, making it challenging to resolve closely spaced frequency components. This paper proposes DeepFT, a novel deep unfolding network that achieves super-resolution spectrum estimation while preserving computational efficiency. The key innovations include a learnable complex-valued transform matrix initialized from the discrete Fourier transform, enabling adaptive frequency analysis; full complex domain iterative processing that preserves phase information throughout the network, enhancing the ability to separate closely spaced frequencies; and an encoder-decoder-sparse module with learnable soft-thresholding for effective sparse signal recovery. Extensive experiments demonstrate that DeepFT achieves frequency resolution of 0.3/N, where N denotes the signal length, significantly surpassing the Rayleigh limit of 1/N. The proposed method shows robust performance under various signal-to-noise ratio (SNR) conditions and successfully resolves multiple frequency components in complex scenarios. The lightweight architecture with 649K parameters enables real time deployment in industrial embedded systems.
- Open Access
- 12 June 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 7 (2) , 1-24
Towards a Dynamic Fresnel Zone Model to WiFi-based Human Activity Recognition
The passive WiFi sensing research has largely centered on activity sensing using fixed-location WiFi transceivers, leading to the development of several theoretical models that aim to map received WiFi signals to human activity. Of these models, the Fresnel zone model has shown to be particularly noteworthy. However, the growing popularity of mobile WiFi receivers has not been matched by corresponding research on mobile receiver-based theoretical models. This paper fills this gap by presenting the first theoretical model to quantify the impact of moving a moving receiver for WiFi sensing. We propose a novel dynamic Fresnel zone model in the free space of an indoor environment, which takes the form of a cluster of concentric hyperbolas centered on the transmitter and reflection subject. We examine three properties of this model, i.e., relating the variation in RF signals received by the receiver to the position and orientation of the human, the movement of the receiver, and the presence of other objects in the environment. To validate this model, we develop a prototype system and conduct extensive experiments. The results are consistent with our theoretical analysis, and the system is able to detect the direction of the transmitter with an accuracy of 10° or better, measure the receiver's relative motion displacement within 1 cm a millimeter-level accuracy, and classify five receiver-side activities with an accuracy of 98%. Our work moves a significant step forward in WiFi sensing and may potentially open up new avenues for future research.
- 12 June 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 7 (2) , 1-24
Open AccessTowards a Dynamic Fresnel Zone Model to WiFi-based Human Activity Recognition
The passive WiFi sensing research has largely centered on activity sensing using fixed-location WiFi transceivers, leading to the development of several theoretical models that aim to map received WiFi signals to human activity. Of these models, the Fresnel zone model has shown to be particularly noteworthy. However, the growing popularity of mobile WiFi receivers has not been matched by corresponding research on mobile receiver-based theoretical models. This paper fills this gap by presenting the first theoretical model to quantify the impact of moving a moving receiver for WiFi sensing. We propose a novel dynamic Fresnel zone model in the free space of an indoor environment, which takes the form of a cluster of concentric hyperbolas centered on the transmitter and reflection subject. We examine three properties of this model, i.e., relating the variation in RF signals received by the receiver to the position and orientation of the human, the movement of the receiver, and the presence of other objects in the environment. To validate this model, we develop a prototype system and conduct extensive experiments. The results are consistent with our theoretical analysis, and the system is able to detect the direction of the transmitter with an accuracy of 10° or better, measure the receiver's relative motion displacement within 1 cm a millimeter-level accuracy, and classify five receiver-side activities with an accuracy of 98%. Our work moves a significant step forward in WiFi sensing and may potentially open up new avenues for future research.
- 12 June 2018
- Published by Springer Nature
- p. 141-152
Recognition and Classification of Rotorcraft by Micro-Doppler Signatures Using Deep Learning
- 12 June 2018
- Published by Springer Nature
- p. 141-152
Recognition and Classification of Rotorcraft by Micro-Doppler Signatures Using Deep Learning