Author name cluster

Guowei Yang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

1 author row

EAAI Journal 2026 Journal Article

Dual-domain parallel attention-driven underwater image enhancement method

Yinghao Zhang
Tianmeng Sun
Jiamin Hu
Haiyuan Cui
Jieru Chi
Guowei Yang
Chenglizhao Chen
Teng Yu

Due to the complex and variable nature of underwater environments, underwater images often suffer from severe degradation phenomena such as light absorption, scattering, and color distortion. Although convolutional neural networks (CNNs) have made progress in this field, their inherent limited receptive field hinders their ability to fully learn global image features, thereby affecting the generalization capability of the model. To address these challenges, this paper proposes an innovative Hybrid Parallel Multi-domain Enhancement Network (HPMENet). The network is based on an encoder–decoder architecture, into which a Mamba sub-network is embedded. Furthermore, a dual-domain hybrid attention module is designed to enhance the capability of the network in extracting and fusing both local and global features. Compared with existing state-of-the-art methods, the proposed approach fully leverages the advantages of the Mamba architecture and integrates it with a dual-domain, bidirectional attention mechanism. This effectively expands the receptive field of the network during the enhancement process, thereby improving the visual fidelity of underwater images. Extensive experimental results demonstrate that the proposed method significantly outperforms current mainstream underwater image enhancement methods in both quantitative and qualitative assessments. Our code and models are available at https: //github. com/Smm-ccd/HPMENet.

Details DOI

JBHI Journal 2025 Journal Article

A Cuffless Blood Pressure Estimation Method Using Dimensionality Increasing and Two-Dimensional Convolution

Shouyi Cui
Guowei Yang
Jingxuan Guan
Yuheng He
Xuefang Zhou
Meihua Bi
Hanghai Shen
Yuansheng Xu

Blood pressure (BP) monitoring is a basic way to evaluate hypertension and its related diseases. Since non-invasive measurement with cuff is not real-time and invasive measurement with vessel puncture is not practical in daily life, this paper proposes a cuffless BP estimation method using two-dimensional (2D) convolution. Dimensionality increasing algorithms including recurrence plot and Gramian angular field are firstly used to convert electrocardiography (ECG) and photoplethysmography (PPG) signals into 2D images. New fused Gramian angular field (FGAF) and combined Gramian angular field (CGAF) are proposed to reduce the input 2D images data and enhance the signals’ relevance. The converted images are used to train 2D convolutional models and estimate BP values. The 2D models effectively improved BP estimation accuracy, and the accuracy of the VGGNet 2D model using Gramian angular difference field (GADF) is improved by 38% compared with the corresponding 1D convolutional model. The proposed FGAF and CGAF can reduce input data by 50% while maintaining estimation accuracy, and the minimum mean absolute errors of the estimated BP values could reach 2. 71 and 1. 74 mmHg for systolic and diastolic blood pressures, respectively. To reduce model size, the VGGNet BP estimation model is pruned by reducing 60% of channel numbers while maintain the model performance. The pruned VGGNet model using the FGADF is then fine-tuned and validated by MIMIC-III dataset to show its generalization ability. Furthermore, a simple monitor system is built to show the feasibility of signal collection and BP estimation.

Details DOI

AAAI Conference 2025 Conference Paper

Enhancing the Adversarial Robustness via Manifold Projection

Zhiting Li
Shibai Yin
Tai-Xiang Jiang
Yexun Hu
Jia-Mian Wu
Guowei Yang
Guisong Liu

Deep learning has been widely applied to various aspects of computer vision, but the emergence of adversarial attacks raises concerns about its reliability. Adversarial training (AT) is one of the most effective defense methods, which incorporates adversarial examples into the training data. However, AT is typically employed in a discriminative learning manner, i.e., learning the mapping (conditional probability) from samples to labels, it essentially reinforces this mapping without considering the underlying data distribution. It is notable that adversarial examples often deviate from the distribution of normal (clean) samples. Therefore, building upon existing adversarial defense schemes, we propose to further exploit the distribution of normal samples, partly from the generative learning perspective, resulting in a novel robustness enhancement paradigm. We train a simple autoencoder (AE) autoregressively on normal samples to learn their prior distribution, effectively serving as an image manifold. This AE is then used as a manifold projection operator to incorporate the distribution information of normal samples. Specifically, we organically integrate the pretrained AE into the training process of both AT and adversarial distillation (AD), a method aiming at improving the robustness of small models with low capacity. Since the AE captures the distribution of normal samples, it can adaptively pull adversarial examples closer to the normal sample manifold, weakening the attack strength of adversarial samples and easing the learning of mappings from adversarial samples to correct labels. From the Pearson correlation coefficient (PCC) between the statistics on normal and adversarial examples, it’s validated that the AE indeed pulls adversarial samples closer to normal samples. Extensive experiments illustrate that our proposed adversarial defense paradigm significantly improves the robustness compared with previous state-of-the-art AT and AD methods.

PDF Details DOI

EAAI Journal 2024 Journal Article

A novel bi-stream network for image dehazing

Qiaoyu Ma
Shijie Wang
Guowei Yang
Chenglizhao Chen
Teng Yu

The existing learning-based image dehazing methods usually adopt the encoder–decoder architecture with convolutional neural networks to estimate latent haze-free images from hazy images. However, the limited receptive field of convolutional neural networks may not effectively capture structure-level information, causing the model to be unable to the haze density. To solve this problem, this paper proposes a bi-decoder structure with a dense non-pooling encoder to enhance the structural features that are closely related to the haze density. Compared with conventional methods, the main advantage of our method is the integration of an additional coarse decoder in the encoder–decoder architecture, where a hybrid feature convolution (HFC) block is utilized to enlarge the receptive field to extract the structure of the image. Besides the difference in the network, the inputs of the fine and coarse decoders are non-pooling and pooling respectively. Moreover, a multi-scale feature attention (MSFA) module is proposed to selectively enhance the haze-relevant feature outputs of fine and coarse decoders. Experimental results on synthetic and real-world datasets demonstrate that the proposed method outperforms most state-of-the-art methods in terms of image quality and quantitative metrics. Especially in the NH-HAZE dataset, its PSNR exceeds other methods by more than 2. 13 dB. In the end, this paper applies this dehazing technology to object detection. The code of this paper and data are available online at https: //github. com/Qiaoyu-K/Bi-Decoder-Dehazing.

Details DOI

EAAI Journal 2024 Journal Article

Instance-aware image dehazing

Qingqing Chao
Jinqiang Yan
Tianmeng Sun
Silong Li
Jieru Chi
Guowei Yang
Chenglizhao Chen
Teng Yu

The existing deep learning-based image dehazing algorithms commonly employ an encoder–decoder structure to learn a direct mapping from hazy images to haze-free images. However, these state-of-the-art methods often fail to consider the varying contents of hazy images across different scenes, resulting in unsatisfactory dehazing outcomes. To address this issue, this paper attempts to integrate a novel instance-aware subnet into the classic encoder–decoder structure in order to achieve a clear separation between figure and background, conducting the selective incorporation of instance features into the dehazing network. Specifically, we introduce a novel architecture called the hybrid residual attention network, which is capable of separately extracting full-image features and instance-level features. This architecture incorporates attention mechanisms and a multi-scale dilated convolution structure, enabling adaptive perception of haze density in different scenes. Additionally, we introduce a global feature fusion subnet that employs a pixel attention structure to fuse features from the entire image and multiple individual instances, thus being aware of instance features. Compared to existing methods, our approach offers a major advantage in accurately estimating the haze density of individual instances, reducing color distortion, and mitigating noise amplification in the output images. Experimental results demonstrate that our method outperforms existing methods across different evaluation metrics and testing benchmarks. Therefore, we believe that our method will serve as a valuable addition to the current collection of artificial intelligence models and will benefit engineering applications in video surveillance and high-level computer vision tasks.

Details DOI

EAAI Journal 2023 Journal Article

RFA-Net: Residual feature attention network for fine-grained image inpainting

Min Chen
Shengrui Zang
Zhenhua Ai
Jieru Chi
Guowei Yang
Chenglizhao Chen
Teng Yu

Although most existing methods using Generative Adversarial Networks (GAN) generally produce plausible results, there is a significant amount of artifacts and less-than-ideal restoration of textures when large regions are missing or the background of missing regions is complex. To address this issue, in this paper, we propose a novel texture-aware backbone net named RFA-Net for finer texture image inpainting. Compared to conventional encoder–decoder methods, our main contribution is proposing a novel RFA-Net adopt a non-pooling residual CNN structure with three novel modules, which retains texture features from shallow layers and adaptively learn the importance of certain channels and locations of features that may potentially benefit image inpainting. In addition, we propose a hybrid loss optimization (HLO) module to enable the generator to focus on the semantic and texture details of the inpainted contents. Experimental results demonstrate that our RFA-Net is able to recover texture details and ground-truth consistent images, and outperforms the state-of-the-art methods both in terms of image quality and quantitative metrics. Our source code and data are available online at https: //github. com/Jamie-61/RFA-Net-Inpainting.

Details DOI

EAAI Journal 2023 Journal Article

Texture-aware gray-scale image colorization using a bistream generative adversarial network with multi scale attention structure

Shengrui Zang
Min Chen
Zhenhua Ai
Jieru Chi
Guowei Yang
Chenglizhao Chen
Teng Yu

Various methods based on deep neural networks have been proposed to generate color images from gray-scale images, meanwhile, Generative adversarial networks (GANs) are also gradually applied to image colorization. However, the existing methods are texture-unaware, resulting in dullish color and color bleeding artifacts in the output images. This paper attempt to integrate a novel texture-aware bistream GAN into the conventional encoder–decoder structure for image colorization. In this study, the proposed bistream feature extraction module (BSFEM) and the feature boosting module (FBM), extract the global and local features from two parallel encoders and fuse them via a novel hybrid attention structure, this novel structure could emphasize the importance of certain channels and locations of features that may potentially benefit image colorization. In addition, the texture colors can be better recovered though the proposed multi-scale feature attention module (MSFAM). The quantitative experiments demonstrate that, compared to the state-of-the-art approaches, the proposed method has improved the PSNR and SSIM metrics by 18% and 8% respectively. Moreover, the qualitative results show that this method is capable of producing visually pleasant color images especially in terms of recovering texture details and eliminating color bleeding along the edges. The source code and data are available online at https: //github. com/JarryZang/Image-Colorization-.

Details DOI