Pengsen Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Pengsen Zhu has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Yonggang Zhang
Harbin Engineering University
4 shared publications - Yulong Huang
Harbin Engineering University
4 shared publications - Boqiang Lin
Department of Intelligent Science and Engineering, Harbin Engineering University, Harbin, China
2 shared publications - Fuheng Zhou
College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin, China
2 shared publications - Lina Gao
College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin, China
2 shared publications - Kunlong Zhao
College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin, China
1 shared publication
Latest Publications
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-22
RGDNet: Recognition-Guided Underwater Acoustic Signal Denoising via Mask Integration and Signal Decoupling
This paper proposes a novel recognition-guided denoising method for underwater acoustic signals, termed RGDNet, which is designed to bridge the gap where existing methods, despite achieving high denoising metrics, fail to effectively enhance recognition accuracy. Specifically, we first designed a task-specific mask integration module that minimizes phase and frequency distortions by converting signal fusion across tasks into a unified mask-based fusion, ensuring the fusion of signal features remains effective. We further developed a context-aware signal decoupling module to maintain the distinctiveness of sub-task signals through segmented context analysis. Following this, a recognition-guided branch equipped with an auxiliary classifier is proposed, which guides the optimization of the denoising process based on recognition loss. Additionally, we also implemented a complex domain multi-metric discriminator that assesses the quality of denoised signals from both metric-focused and task-specific perspectives, effectively linking denoising metrics with recognition performance. Extensive experiments show that RGDNet significantly outperforms existing methods, increasing the scale-invariant signal-to-noise ratio (SI-SNR) by an average of 6.33 dB and enhancing recognition accuracy by 11.26%.
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-22
RGDNet: Recognition-Guided Underwater Acoustic Signal Denoising via Mask Integration and Signal Decoupling
This paper proposes a novel recognition-guided denoising method for underwater acoustic signals, termed RGDNet, which is designed to bridge the gap where existing methods, despite achieving high denoising metrics, fail to effectively enhance recognition accuracy. Specifically, we first designed a task-specific mask integration module that minimizes phase and frequency distortions by converting signal fusion across tasks into a unified mask-based fusion, ensuring the fusion of signal features remains effective. We further developed a context-aware signal decoupling module to maintain the distinctiveness of sub-task signals through segmented context analysis. Following this, a recognition-guided branch equipped with an auxiliary classifier is proposed, which guides the optimization of the denoising process based on recognition loss. Additionally, we also implemented a complex domain multi-metric discriminator that assesses the quality of denoised signals from both metric-focused and task-specific perspectives, effectively linking denoising metrics with recognition performance. Extensive experiments show that RGDNet significantly outperforms existing methods, increasing the scale-invariant signal-to-noise ratio (SI-SNR) by an average of 6.33 dB and enhancing recognition accuracy by 11.26%.
- 20 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 50 (2) , 758-775
SCN: A Novel Underwater Images Enhancement Method Based on Single Channel Network Model
Light is absorbed, reflected, and refracted in an underwater environment due to the interaction between water and light. The red and blue channels in an image are attenuated due to these interactions. The red, green, and blue channels are typically employed as inputs for deep learning models, and the color casts, which result from different attenuation rates of the three channels, may affect the model's generalization performance. Besides, the color casts existing in the reference images will impact the deep-learning models. To address these challenges, a single channel network (SCN) model is introduced, which exclusively employs the green channel as its input, and is unaffected by the attenuations in the red and blue channels. An innovative feature processing module is presented, in which the characteristics of transformers and convolutional layers are fused to capture nonlinear relationships among the red, green, and blue channels. The public EUVP and LSUI data set experiments show that the proposed SCN model achieves competitive results with the existing best three channel models for the case of slight signal attenuation, and outperforms the existing state of arts three-channel models for the case of strong signal attenuation. Furthermore, the proposed model is trained on the self-built noncolor biased underwater image data set and is also tested on the public UCCS data set with three different types of color casts, whose experimental results exhibit balanced color distribution.
- 20 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 50 (2) , 758-775
SCN: A Novel Underwater Images Enhancement Method Based on Single Channel Network Model
Light is absorbed, reflected, and refracted in an underwater environment due to the interaction between water and light. The red and blue channels in an image are attenuated due to these interactions. The red, green, and blue channels are typically employed as inputs for deep learning models, and the color casts, which result from different attenuation rates of the three channels, may affect the model's generalization performance. Besides, the color casts existing in the reference images will impact the deep-learning models. To address these challenges, a single channel network (SCN) model is introduced, which exclusively employs the green channel as its input, and is unaffected by the attenuations in the red and blue channels. An innovative feature processing module is presented, in which the characteristics of transformers and convolutional layers are fused to capture nonlinear relationships among the red, green, and blue channels. The public EUVP and LSUI data set experiments show that the proposed SCN model achieves competitive results with the existing best three channel models for the case of slight signal attenuation, and outperforms the existing state of arts three-channel models for the case of strong signal attenuation. Furthermore, the proposed model is trained on the self-built noncolor biased underwater image data set and is also tested on the public UCCS data set with three different types of color casts, whose experimental results exhibit balanced color distribution.
- 16 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (16) , 26199-26210
An Underwater Acoustic Target Recognition Method Based on Iterative Short-Time Fourier Transform
In the multifaceted marine environment, numerous factors affect the noise radiated by ships, thereby weakening the traditional spectral characteristics and diminishing the spectrum’s ability to express identity information. Consequently, these changes result in reduced accuracy in recognition tasks. To address this issue, we propose a novel method for underwater acoustic target recognition aimed at extracting the intrinsic frequency distribution features of different frequency band energies in ship radiated noise to supplement traditional time-frequency features, named ISNet. This method iteratively extracts the real and imaginary parts of the frequency features using Short-Time Fourier Transform (STFT) and combines them with time-frequency features to form a feature matrix. After enhancement through fuzzy mixed features, the features are input into a dimensionality-reduced RSNet-18 network for training, and final predictions are obtained through a step-wise voting strategy. Experimental results show that this method surpasses existing approaches, achieving a recognition accuracy of 84.24% on the DeepShip dataset.
- 16 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (16) , 26199-26210
An Underwater Acoustic Target Recognition Method Based on Iterative Short-Time Fourier Transform
In the multifaceted marine environment, numerous factors affect the noise radiated by ships, thereby weakening the traditional spectral characteristics and diminishing the spectrum’s ability to express identity information. Consequently, these changes result in reduced accuracy in recognition tasks. To address this issue, we propose a novel method for underwater acoustic target recognition aimed at extracting the intrinsic frequency distribution features of different frequency band energies in ship radiated noise to supplement traditional time-frequency features, named ISNet. This method iteratively extracts the real and imaginary parts of the frequency features using Short-Time Fourier Transform (STFT) and combines them with time-frequency features to form a feature matrix. After enhancement through fuzzy mixed features, the features are input into a dimensionality-reduced RSNet-18 network for training, and final predictions are obtained through a step-wise voting strategy. Experimental results show that this method surpasses existing approaches, achieving a recognition accuracy of 84.24% on the DeepShip dataset.
- 2 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 61 (01962892) , 1-23
SFC-Sup: Robust Two-Stage Underwater Acoustic Target Recognition Method Based on Supervised Contrastive Learning
This paper presents an underwater acoustic target recognition method to reduce recognition errors in continuous recordings caused by variations in ship operating conditions. The proposed method comprises two-stages: the spectral feature classification and the supervised contrastive learning, and it is called as SFC-Sup as a result in this paper. In the first stage, a new spectral feature classification strategy is designed to choose appropriate feature sets for contrastive learning, based on which an instance discrimination pretext task is created by utilizing different spectral features to capture invariant features across segments under different operating conditions. In the second stage, a dynamic weighted loss function is introduced to guide the joint optimization process in the framework of contrastive learning. Different to existing methods which focus on improving the recognition accuracy by designing features for individual segments, the proposed two-stage method SFC-Sup considers consistent features across diverse segments, which is expected to improve recognition accuracy in a continuous recording. Experimental results demonstrate that in the presence of complex operating conditions, SFC-Sup exhibits superior stability and enhances recognition accuracy by 2.06% compared to state-of-the-art methods.
- 2 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 61 (01962892) , 1-23
SFC-Sup: Robust Two-Stage Underwater Acoustic Target Recognition Method Based on Supervised Contrastive Learning
This paper presents an underwater acoustic target recognition method to reduce recognition errors in continuous recordings caused by variations in ship operating conditions. The proposed method comprises two-stages: the spectral feature classification and the supervised contrastive learning, and it is called as SFC-Sup as a result in this paper. In the first stage, a new spectral feature classification strategy is designed to choose appropriate feature sets for contrastive learning, based on which an instance discrimination pretext task is created by utilizing different spectral features to capture invariant features across segments under different operating conditions. In the second stage, a dynamic weighted loss function is introduced to guide the joint optimization process in the framework of contrastive learning. Different to existing methods which focus on improving the recognition accuracy by designing features for individual segments, the proposed two-stage method SFC-Sup considers consistent features across diverse segments, which is expected to improve recognition accuracy in a continuous recording. Experimental results demonstrate that in the presence of complex operating conditions, SFC-Sup exhibits superior stability and enhances recognition accuracy by 2.06% compared to state-of-the-art methods.