Yin Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yin Liu has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Tiecheng Song
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
7 shared publications - Chenqiang Gao
School of Intelligent Systems Engineering, Sun Yat-sen University, Shenzhen, Guangdong, China
6 shared publications - Feng Yang
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China
6 shared publications - Anyong Qin
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China
5 shared publications - Yi Zhou
Chongqing University of Posts and Telecommunications
5 shared publications - Yu Zhao
Chongqing University of Posts and Telecommunications
4 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 9054-9068
An Efficient Asymmetric Encoder-Decoder Network for High-Quality Monaural Speech Enhancement
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 9054-9068
An Efficient Asymmetric Encoder-Decoder Network for High-Quality Monaural Speech Enhancement
- Open Access
- 13 September 2026
- Published by Elsevier in SSRN Electronic Journal
CASTS: CLS-Guided Adaptive Sparse Token Selection with Multi-Layer Feature Fusion for Interpretable Auditory Brainstem Response Classification
The auditory brainstem response (ABR) is an important objective tool for hearing assessment, but its manual interpretation remains labor-intensive and subject to inter-observer variability. Although deep learning approaches applied to the actual ABR voltage measurements have shown promising performance, their clinical use may be constrained by limited access to raw data, as commercial ABR systems often store recordings in vendor-specific formats and require proprietary software or device-specific export procedures. In contrast, ABR report images are routinely generated in clinical practice and provide a more accessible source for automated analysis. However, existing image-based approaches often struggle to capture long-range waveform dependencies and may be distracted by non-informative backgrounds. To address these challenges, we propose CASTS, an interpretable framework for automated ABR report image classification. CASTS uses a pretrained ViT-B/16 backbone with multi-layer feature fusion to integrate hierarchical waveform-image representations. A classification (CLS)-guided top-k patch selection module is also introduced to identify and aggregate waveform-relevant patch tokens according to their similarity to the CLS token, thereby reducing the influence of background-dominant regions. Evaluated on a public dataset containing 186 ABR report images, CASTS achieved a mean classification accuracy of 94.88% and an AUC of 0.976, outperforming representative convolutional and Transformer-based feature extraction baselines. Ablation studies confirmed the effectiveness of the proposed components. Occlusion-based sensitivity analysis further showed that CASTS primarily focused on clinically meaningful waveform structure, especially wave V, rather than non-informative background areas. These results suggest that CASTS provides an effective and interpretable solution for automated ABR report image analysis.
- 13 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCASTS: CLS-Guided Adaptive Sparse Token Selection with Multi-Layer Feature Fusion for Interpretable Auditory Brainstem Response Classification
The auditory brainstem response (ABR) is an important objective tool for hearing assessment, but its manual interpretation remains labor-intensive and subject to inter-observer variability. Although deep learning approaches applied to the actual ABR voltage measurements have shown promising performance, their clinical use may be constrained by limited access to raw data, as commercial ABR systems often store recordings in vendor-specific formats and require proprietary software or device-specific export procedures. In contrast, ABR report images are routinely generated in clinical practice and provide a more accessible source for automated analysis. However, existing image-based approaches often struggle to capture long-range waveform dependencies and may be distracted by non-informative backgrounds. To address these challenges, we propose CASTS, an interpretable framework for automated ABR report image classification. CASTS uses a pretrained ViT-B/16 backbone with multi-layer feature fusion to integrate hierarchical waveform-image representations. A classification (CLS)-guided top-k patch selection module is also introduced to identify and aggregate waveform-relevant patch tokens according to their similarity to the CLS token, thereby reducing the influence of background-dominant regions. Evaluated on a public dataset containing 186 ABR report images, CASTS achieved a mean classification accuracy of 94.88% and an AUC of 0.976, outperforming representative convolutional and Transformer-based feature extraction baselines. Ablation studies confirmed the effectiveness of the proposed components. Occlusion-based sensitivity analysis further showed that CASTS primarily focused on clinically meaningful waveform structure, especially wave V, rather than non-informative background areas. These results suggest that CASTS provides an effective and interpretable solution for automated ABR report image analysis.
- Open Access
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 34 (15344320) , 2245-2255
A Dynamic Time Warping-Aware Series-Temporal Transformer for Automated Thresholding of Auditory Brainstem Responses
Auditory brainstem response (ABR) is widely used in clinical practice to assess hearing, with thresholds typically determined by clinicians via visual inspection—a process prone to human bias. Various statistical methods have therefore been developed to assist clinicians with this task and improve test accuracy and efficiency. However, existing methods primarily focus on determining the presence or absence of an ABR at a single stimulus level, thereby neglecting inter-level correlations that hold valuable information. This work proposes the Dynamic Time Warping (DTW)-Aware Series-Temporal Transformer (DTWA-STformer), a novel deep learning framework for automated ABR threshold estimation from a series of ABR waveforms. DTWA-STformer employs a DTW similarity-aware Series Transformer to capture inter-waveform dependencies, enhanced by stimulus level-informed positional encodings. A hierarchical multi-scale Temporal Transformer then extracts rich temporal representations, followed by a multi-class classifier for threshold prediction. The model was trained and evaluated on three pre-recorded datasets, including two large-scale human datasets (Dataset I: 8,350 subjects; Dataset II: 136 subjects) and a public mouse dataset (8,259 mice). The model achieved exact-match/±10 dB accuracies of 92.08%/99.31% (Dataset I), 90.35%/98.75% (Dataset II), 73.45%/99.27% (mouse click ABRs), and 60.85%/98.82% (mouse tone-pip ABRs), outperforming existing state-of-the-art methods for threshold estimation across both human and mouse data. Results show that DTWA-STformer provides accurate and objective threshold estimation from pre-recorded waveforms and may serve as a post hoc tool for verifying examiner-estimated thresholds. In future work, the approach could be integrated with active learning rules for dynamically selecting stimulus levels and more efficiently homing in on hearing threshold.
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 34 (15344320) , 2245-2255
Open AccessA Dynamic Time Warping-Aware Series-Temporal Transformer for Automated Thresholding of Auditory Brainstem Responses
Auditory brainstem response (ABR) is widely used in clinical practice to assess hearing, with thresholds typically determined by clinicians via visual inspection—a process prone to human bias. Various statistical methods have therefore been developed to assist clinicians with this task and improve test accuracy and efficiency. However, existing methods primarily focus on determining the presence or absence of an ABR at a single stimulus level, thereby neglecting inter-level correlations that hold valuable information. This work proposes the Dynamic Time Warping (DTW)-Aware Series-Temporal Transformer (DTWA-STformer), a novel deep learning framework for automated ABR threshold estimation from a series of ABR waveforms. DTWA-STformer employs a DTW similarity-aware Series Transformer to capture inter-waveform dependencies, enhanced by stimulus level-informed positional encodings. A hierarchical multi-scale Temporal Transformer then extracts rich temporal representations, followed by a multi-class classifier for threshold prediction. The model was trained and evaluated on three pre-recorded datasets, including two large-scale human datasets (Dataset I: 8,350 subjects; Dataset II: 136 subjects) and a public mouse dataset (8,259 mice). The model achieved exact-match/±10 dB accuracies of 92.08%/99.31% (Dataset I), 90.35%/98.75% (Dataset II), 73.45%/99.27% (mouse click ABRs), and 60.85%/98.82% (mouse tone-pip ABRs), outperforming existing state-of-the-art methods for threshold estimation across both human and mouse data. Results show that DTWA-STformer provides accurate and objective threshold estimation from pre-recorded waveforms and may serve as a post hoc tool for verifying examiner-estimated thresholds. In future work, the approach could be integrated with active learning rules for dynamically selecting stimulus levels and more efficiently homing in on hearing threshold.
- 25 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. PP (21682194) , 1-14
A Graph Convolutional Network with Pretrained Features and Iterative Polar Coordinate Attention for Cross-Subject EEG-based Neuropsychiatric Diagnosis
Electroencephalography (EEG)-based diagnosis of neuropsychiatric disorders offers a non-invasive and cost-effective solution for early detection. However, robust cross-subject generalization remains a major challenge due to substantial inter-individual variability in EEG signals. To address this, we propose PreIPCA-GCN, a novel graph convolutional network that integrates pretrained temporal features and Iterative Polar Coordinate Attention (IPCA)-based brain connectivity modeling. Specifically, we utilize a modified version of LaBraM, a large-scale pretrained EEG model, to extract subject-invariant node representations. Functional brain connectivity is then characterized using Pearson correlation and cosine similarity in polar space, capturing both connectivity strength (radius) and phase synchronization (angle). To fuse these complementary cues, we introduce a dual-path IPCA mechanism, refining the adjacency matrix across iterations. PreIPCA-GCN is evaluated on six public EEG datasets covering five neuropsychiatric disorders (e.g., attention-deficit/hyperactivity disorder, Alzheimer's disease, schizophrenia), consistently demonstrating strong cross-subject accuracies under both hold-out (86.77%-95.69%) and leave-one-subject-out cross-validation (88.81%-97.49%). Comprehensive comparative results show that PreIPCA-GCN outperforms several state-of-the-art methods. Ablation studies further confirm the effectiveness of both the pretrained node features and IPCA-based fused adjacency matrix in improving cross-subject generalization. These findings suggest PreIPCA-GCN as a robust and generalizable framework for cross-subject EEG-based neuropsychiatric diagnosis, offering strong potential for future clinical applications.
- 25 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. PP (21682194) , 1-14
A Graph Convolutional Network with Pretrained Features and Iterative Polar Coordinate Attention for Cross-Subject EEG-based Neuropsychiatric Diagnosis
Electroencephalography (EEG)-based diagnosis of neuropsychiatric disorders offers a non-invasive and cost-effective solution for early detection. However, robust cross-subject generalization remains a major challenge due to substantial inter-individual variability in EEG signals. To address this, we propose PreIPCA-GCN, a novel graph convolutional network that integrates pretrained temporal features and Iterative Polar Coordinate Attention (IPCA)-based brain connectivity modeling. Specifically, we utilize a modified version of LaBraM, a large-scale pretrained EEG model, to extract subject-invariant node representations. Functional brain connectivity is then characterized using Pearson correlation and cosine similarity in polar space, capturing both connectivity strength (radius) and phase synchronization (angle). To fuse these complementary cues, we introduce a dual-path IPCA mechanism, refining the adjacency matrix across iterations. PreIPCA-GCN is evaluated on six public EEG datasets covering five neuropsychiatric disorders (e.g., attention-deficit/hyperactivity disorder, Alzheimer's disease, schizophrenia), consistently demonstrating strong cross-subject accuracies under both hold-out (86.77%-95.69%) and leave-one-subject-out cross-validation (88.81%-97.49%). Comprehensive comparative results show that PreIPCA-GCN outperforms several state-of-the-art methods. Ablation studies further confirm the effectiveness of both the pretrained node features and IPCA-based fused adjacency matrix in improving cross-subject generalization. These findings suggest PreIPCA-GCN as a robust and generalizable framework for cross-subject EEG-based neuropsychiatric diagnosis, offering strong potential for future clinical applications.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23, 15288-15301
SCIRCNet: Rotation-Invariant Point Cloud Completion via Self-Contour-Aware Representation
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23, 15288-15301
SCIRCNet: Rotation-Invariant Point Cloud Completion via Self-Contour-Aware Representation
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (1) , 957-970
Selective Fixed-Filter Active Noise Control Based on Parallel Structure and Bayesian Tracking
Active noise control (ANC) is an effective method for suppressing unwanted noise. While most commercial ANC products rely on pre-trained fixed-filters for noise attenuation due to their high computational efficiency and stability, the noise reduction (NR) performance is prone to variations in the direction of incident noise and noise spectral content mismatch. In this paper, a selective fixed-filter ANC algorithm based on parallel structure and Bayesian tracking (SFANC-PSBT) is proposed to mitigate these problems. Bayesian tracking is applied to deal with the noise source direction uncertainties and to select the grid-based fixed filter with the largest posterior probability. Moreover, the parallel broadband fixed-filter adaptive gain (PBFAG) approach is proposed to improve NR performance for off-grid directions, and the parallel narrowband fixed-filter adaptive gain (PNFAG) approach is proposed to further attenuate noise and to mitigate the noise spectral content mismatch issue. Besides, stochastic analysis of the proposed methods is provided. Experimental results demonstrate that the proposed methods achieve better NR performance, faster convergence rate and higher robustness against noise variations under changing primary path conditions. The PNFAG and PBFAG methods yield NR improvements of 23.71% and 18.67%, respectively, over the conventional fixed-filter method. Some before/after ANC sound samples under different noise conditions can be found in https://zhchengcq.github.io/ANCdemo/.
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (1) , 957-970
Selective Fixed-Filter Active Noise Control Based on Parallel Structure and Bayesian Tracking
Active noise control (ANC) is an effective method for suppressing unwanted noise. While most commercial ANC products rely on pre-trained fixed-filters for noise attenuation due to their high computational efficiency and stability, the noise reduction (NR) performance is prone to variations in the direction of incident noise and noise spectral content mismatch. In this paper, a selective fixed-filter ANC algorithm based on parallel structure and Bayesian tracking (SFANC-PSBT) is proposed to mitigate these problems. Bayesian tracking is applied to deal with the noise source direction uncertainties and to select the grid-based fixed filter with the largest posterior probability. Moreover, the parallel broadband fixed-filter adaptive gain (PBFAG) approach is proposed to improve NR performance for off-grid directions, and the parallel narrowband fixed-filter adaptive gain (PNFAG) approach is proposed to further attenuate noise and to mitigate the noise spectral content mismatch issue. Besides, stochastic analysis of the proposed methods is provided. Experimental results demonstrate that the proposed methods achieve better NR performance, faster convergence rate and higher robustness against noise variations under changing primary path conditions. The PNFAG and PBFAG methods yield NR improvements of 23.71% and 18.67%, respectively, over the conventional fixed-filter method. Some before/after ANC sound samples under different noise conditions can be found in https://zhchengcq.github.io/ANCdemo/.
- 6 July 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 296
Progressive spectral-frequency-spatial guidance network for hyperspectral image dehazing
- 6 July 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 296
Progressive spectral-frequency-spatial guidance network for hyperspectral image dehazing
- 7 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 1670-1674
Improved Encoder-Decoder Architecture With Human-Like Perception Attention for Monaural Speech Enhancement
Speech enhancement (SE) models based on deep neural networks (DNNs) have shown excellent denoising performance. However, mainstream SE models often have high structural complexity and large parameter sizes, requiring substantial computational resources, which limits their practical application. In this paper, a high-efficiency encoder-decoder structure, inspired by the top-down attention mechanism in human brain perception and named human-like perception attention network (HPANet), is proposed for monaural speech enhancement, which is able to emulate brain perceptual attention in noise environments. In HPANet, the raw waveform is first encoded by using attention encoder to capture shallow global features. These features are then downsampled, and multi-scale information is aggregated through top attention module to prevent the loss of crucial information. Next, down attention module integrates features from neighboring layers to reconstruct signal in a top-down manner. Finally, the decoder reconstructs the denoised clean signal. Experiments show that the proposed method effectively reduces model complexity while maintaining competitive performance.
- 7 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 1670-1674
Improved Encoder-Decoder Architecture With Human-Like Perception Attention for Monaural Speech Enhancement
Speech enhancement (SE) models based on deep neural networks (DNNs) have shown excellent denoising performance. However, mainstream SE models often have high structural complexity and large parameter sizes, requiring substantial computational resources, which limits their practical application. In this paper, a high-efficiency encoder-decoder structure, inspired by the top-down attention mechanism in human brain perception and named human-like perception attention network (HPANet), is proposed for monaural speech enhancement, which is able to emulate brain perceptual attention in noise environments. In HPANet, the raw waveform is first encoded by using attention encoder to capture shallow global features. These features are then downsampled, and multi-scale information is aggregated through top attention module to prevent the loss of crucial information. Next, down attention module integrates features from neighboring layers to reconstruct signal in a top-down manner. Finally, the decoder reconstructs the denoised clean signal. Experiments show that the proposed method effectively reduces model complexity while maintaining competitive performance.
- Open Access
- 2 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 3989-4003
HDSA-Net: Haze Density and Semantic Awareness Network for Hyperspectral Image Dehazing
Hyperspectral image (HSI) dehazing is a challenging task due to the complex imaging conditions. Existing deep learning-based dehazing methods neither fully consider the physical characteristics of HSIs, nor take advantage of high-level semantic information to improve the dehazing performance. To remedy these, in this paper we propose a Haze Density and Semantic Awareness Network (HDSA-Net) for HSI dehazing. Our dual-awareness network not only provides low-level physical information guidance but also high-level semantic guidance for haze removal. Specifically, we estimate the haze density by considering both internal spectral characteristics and external dehazing effects. Based on this, we build a Haze Density Awareness (HDA) block, which enables the network to perceive and focus on difficult dehazing regions with high density. Moreover, we design a Semantic information Extraction Block (SEB) based on the pretrained Segment Anything Model (SAM), followed by several Semantic information Perception Blocks (SPBs), to provide semantic guidance for HSI dehazing. In particular, SEB adapts SAM for the special HSI data and SPBs enable the network to progressively recover semantic information via channel-level coarse guidance and pixel-level fine guidance. The experimental results on simulated and real datasets show the superiority of HDSA-Net over state-of-the-art methods.
- 2 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 3989-4003
Open AccessHDSA-Net: Haze Density and Semantic Awareness Network for Hyperspectral Image Dehazing
Hyperspectral image (HSI) dehazing is a challenging task due to the complex imaging conditions. Existing deep learning-based dehazing methods neither fully consider the physical characteristics of HSIs, nor take advantage of high-level semantic information to improve the dehazing performance. To remedy these, in this paper we propose a Haze Density and Semantic Awareness Network (HDSA-Net) for HSI dehazing. Our dual-awareness network not only provides low-level physical information guidance but also high-level semantic guidance for haze removal. Specifically, we estimate the haze density by considering both internal spectral characteristics and external dehazing effects. Based on this, we build a Haze Density Awareness (HDA) block, which enables the network to perceive and focus on difficult dehazing regions with high density. Moreover, we design a Semantic information Extraction Block (SEB) based on the pretrained Segment Anything Model (SAM), followed by several Semantic information Perception Blocks (SPBs), to provide semantic guidance for HSI dehazing. In particular, SEB adapts SAM for the special HSI data and SPBs enable the network to progressively recover semantic information via channel-level coarse guidance and pixel-level fine guidance. The experimental results on simulated and real datasets show the superiority of HDSA-Net over state-of-the-art methods.