Hongru Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hongru Wang has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
Hu Cheng
Jiangsu University of Science and Technologyhttps://ror.org/00tyjp878, Zhenjiang, Jiangsu Province, China
3 shared publicationsJingtao Zhang
Jiangsu University of Science and Technologyhttps://ror.org/00tyjp878, Zhenjiang, Jiangsu Province, China
3 shared publications- Jingtao Zhang
Jiangsu University of Science and Technology, Zhenjiang, China
2 shared publications Chaolei Dai
School of Mechanical Engineering, Jiangsu University of Science and Technology, Zhenjiang, 212100, People’s Republic of China
1 shared publication- Chaolei Dai
School of Mechanical Engineering, Jiangsu University of Science and Technology, China
1 shared publication - Hu Cheng
School of Mechanical Engineering, Jiangsu University of Science and Technology, Zhenjiang 212100, China
1 shared publication
Latest Publications
- 14 August 2026
- Published by Optica Publishing Group in Applied Optics
- Vol. 65 (24) , 8300-8310
Multi-scale atmospheric adaptive fusion network for water surface image dehazing
Image dehazing is a critical task in computer vision, widely applied in environmental monitoring, autonomous driving, and surveillance. However, non-uniform haze distribution with complex water-surface reflections and refractions often suppresses image details and invalidates conventional atmospheric priors. To address these challenges, this paper proposes a novel, to our knowledge, U-Net-based architecture, the multi-scale atmospheric adaptive fusion U-Net (MSAAFU-Net). The framework incorporates three tailored modules: 1) an atmospheric scattering attention module (ASAM) that embeds learnable physical priors to model spatially varying haze and aquatic optical properties; 2) a multi-scale frequency enhancement module (MSFEM) designed to restore both low- and high-frequency structures; and 3) an adaptive bidirectional fusion module (ABFM) to strengthen feature interaction between the encoder and decoder. Extensive experiments on RESIDE-OTS, NH-Haze, and a self-made water-surface dataset demonstrate that MSAAFU-Net consistently outperforms recent state-of-the-art methods, including FFA-Net, AOD-Net, DehazeFormer, and several diffusion models. On average, it achieves a +1.24dB improvement in PSNR, with noticeable SSIM improvements on the self-made and NH-Haze datasets, while performing competitively on RESIDE-OTS. In terms of perceptual quality, our method reduces NIQE by 0.081 and LPIPS by 0.011 over the strongest baseline. These results confirm that MSAAFU-Net provides obvious advantages in both objective fidelity and subjective visual quality, underscoring its effectiveness for challenging water-surface dehazing tasks.
- 14 August 2026
- Published by Optica Publishing Group in Applied Optics
- Vol. 65 (24) , 8300-8310
Multi-scale atmospheric adaptive fusion network for water surface image dehazing
Image dehazing is a critical task in computer vision, widely applied in environmental monitoring, autonomous driving, and surveillance. However, non-uniform haze distribution with complex water-surface reflections and refractions often suppresses image details and invalidates conventional atmospheric priors. To address these challenges, this paper proposes a novel, to our knowledge, U-Net-based architecture, the multi-scale atmospheric adaptive fusion U-Net (MSAAFU-Net). The framework incorporates three tailored modules: 1) an atmospheric scattering attention module (ASAM) that embeds learnable physical priors to model spatially varying haze and aquatic optical properties; 2) a multi-scale frequency enhancement module (MSFEM) designed to restore both low- and high-frequency structures; and 3) an adaptive bidirectional fusion module (ABFM) to strengthen feature interaction between the encoder and decoder. Extensive experiments on RESIDE-OTS, NH-Haze, and a self-made water-surface dataset demonstrate that MSAAFU-Net consistently outperforms recent state-of-the-art methods, including FFA-Net, AOD-Net, DehazeFormer, and several diffusion models. On average, it achieves a +1.24dB improvement in PSNR, with noticeable SSIM improvements on the self-made and NH-Haze datasets, while performing competitively on RESIDE-OTS. In terms of perceptual quality, our method reduces NIQE by 0.081 and LPIPS by 0.011 over the strongest baseline. These results confirm that MSAAFU-Net provides obvious advantages in both objective fidelity and subjective visual quality, underscoring its effectiveness for challenging water-surface dehazing tasks.
- 1 January 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (1) , 015235
DMS-CycleGAN: an effective underwater image enhancement method
Underwater image enhancement is a critical component of underwater vision technology and has attracted growing attention from researchers worldwide. However, adverse factors such as excessive noise, severe haziness, and color distortion—primarily caused by the scattering and absorption of light in water as well as suspended particle interference—make it difficult to attain reliable enhancement of underwater images. Although conventional enhancement methods can improve image quality to some extent, they still face limitations in addressing these issues. To overcome such challenges, this article proposes an effective CycleGAN-based underwater image enhancement method, termed Dark-channel Multi-head Smooth CycleGAN (DMS-CycleGAN). Unlike previous studies that adopt attention mechanisms, activation functions, or loss terms individually, our method systematically integrates four complementary components:(1) a Dark Channel Prior (DCP) refined with guided filtering to deliver clearer and more stable input data; (2) multi-head attention modules integrated into the generator to effectively capture spatially variant degradation and enhance fine-grained details; (3) the Mish activation function in the discriminator to stabilize adversarial training and suppress artifacts; and (4) a texture loss function designed to preserve local structures and details detail realism beyond conventional perceptual or structural similarity losses. This unified design is developed to address challenges specific to underwater imagery, with the goal of systematically improving color fidelity, structural consistency, and texture preservation. Comprehensive experiments conducted on benchmark datasets demonstrate that the proposed DMS-CycleGAN outperforms existing approaches in both subjective and objective evaluations, particularly in terms of color cast correction and detail retention.
- 1 January 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (1) , 015235
DMS-CycleGAN: an effective underwater image enhancement method
Underwater image enhancement is a critical component of underwater vision technology and has attracted growing attention from researchers worldwide. However, adverse factors such as excessive noise, severe haziness, and color distortion—primarily caused by the scattering and absorption of light in water as well as suspended particle interference—make it difficult to attain reliable enhancement of underwater images. Although conventional enhancement methods can improve image quality to some extent, they still face limitations in addressing these issues. To overcome such challenges, this article proposes an effective CycleGAN-based underwater image enhancement method, termed Dark-channel Multi-head Smooth CycleGAN (DMS-CycleGAN). Unlike previous studies that adopt attention mechanisms, activation functions, or loss terms individually, our method systematically integrates four complementary components:(1) a Dark Channel Prior (DCP) refined with guided filtering to deliver clearer and more stable input data; (2) multi-head attention modules integrated into the generator to effectively capture spatially variant degradation and enhance fine-grained details; (3) the Mish activation function in the discriminator to stabilize adversarial training and suppress artifacts; and (4) a texture loss function designed to preserve local structures and details detail realism beyond conventional perceptual or structural similarity losses. This unified design is developed to address challenges specific to underwater imagery, with the goal of systematically improving color fidelity, structural consistency, and texture preservation. Comprehensive experiments conducted on benchmark datasets demonstrate that the proposed DMS-CycleGAN outperforms existing approaches in both subjective and objective evaluations, particularly in terms of color cast correction and detail retention.
- 16 January 2025
- Published by IOP Publishing in Engineering Research Express
- Vol. 7 (1) , 015222
An effective unsupervised image stitching method based on improved UDISNet
Addressing the challenges of artifacts, misalignment, and distortion in complex scenarios, which arise from difficulties in feature extraction from images, we proposed a novel unsupervised image stitching method based on Unsupervised Deep Image Stitching Net (UDISNet). This method improves the UDISNet-based unsupervised image stitching method from two aspects: image alignment and image reconstruction. In the image alignment stage, to solve the problem that the network cannot extract effective alignment information in large baseline scenarios, the Edge-Preserved Image Stitching Net(EPISNet) large baseline depth homography network is introduced and an AdaPool layer is used between each two convolution layers. In the image reconstruction stage, to address the problems of weak image edge information detection, artifacts, and misalignment, we employ the Edge-Preserved Deformation Net (EPDNet) image edge detection module and deepen the network. The comparative experimental results show that the RMSE, PSNR, and SSIM of our method are 1.93, 24.85, and 0.85, respectively, which are all superior to those of other state-of-the-art methods. Moreover, extensive experiments indicate that our method can effectively stitch both ground and water surface images, which means that it has a high generalization. Therefore, our work is effective and constructive.
- 16 January 2025
- Published by IOP Publishing in Engineering Research Express
- Vol. 7 (1) , 015222
An effective unsupervised image stitching method based on improved UDISNet
Addressing the challenges of artifacts, misalignment, and distortion in complex scenarios, which arise from difficulties in feature extraction from images, we proposed a novel unsupervised image stitching method based on Unsupervised Deep Image Stitching Net (UDISNet). This method improves the UDISNet-based unsupervised image stitching method from two aspects: image alignment and image reconstruction. In the image alignment stage, to solve the problem that the network cannot extract effective alignment information in large baseline scenarios, the Edge-Preserved Image Stitching Net(EPISNet) large baseline depth homography network is introduced and an AdaPool layer is used between each two convolution layers. In the image reconstruction stage, to address the problems of weak image edge information detection, artifacts, and misalignment, we employ the Edge-Preserved Deformation Net (EPDNet) image edge detection module and deepen the network. The comparative experimental results show that the RMSE, PSNR, and SSIM of our method are 1.93, 24.85, and 0.85, respectively, which are all superior to those of other state-of-the-art methods. Moreover, extensive experiments indicate that our method can effectively stitch both ground and water surface images, which means that it has a high generalization. Therefore, our work is effective and constructive.
- 12 November 2024
- Published by Springer Nature in The Visual Computer
- Vol. 41 (7) , 5087-5104
Faster-PGYOLO: an efficient framework for floating debris detection in inland waters
- 12 November 2024
- Published by Springer Nature in The Visual Computer
- Vol. 41 (7) , 5087-5104
Faster-PGYOLO: an efficient framework for floating debris detection in inland waters
- Open Access
- 6 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (17) , 3547
HRA-YOLO: An Effective Detection Model for Underwater Fish
In intelligent fisheries, accurate fish detection is essential to monitor underwater ecosystems. By utilizing underwater cameras and computer vision technologies to detect fish distribution, timely feedback can be provided to staff, enabling effective fishery management. This paper proposes a lightweight underwater fish detection algorithm based on YOLOv8s, named HRA-YOLO, to meet the demand for a high-precision and lightweight object detection algorithm. Firstly, the lightweight network High-Performance GPU Net (HGNetV2) is used to substitute the backbone network of the YOLOv8s model to lower the computational cost and reduce the size of the model. Second, to enhance the capability of extracting fish feature information and reducing missed detections, we design a residual attention (RA) module, which is formulated by embedding the efficient multiscale attention (EMA) mechanism at the end of the Dilation-Wise Residual (DWR) module. Then, we adopt the RA module to replace the bottleneck of the YOLOv8s model to increase detection precision. Taking universality into account, we establish an underwater fish dataset for our subsequent experiments by collecting data in various waters. Comprehensive experiments are carried out on the self-constructed fish dataset. The results on the self-constructed dataset demonstrate that the precision of the HRA-YOLO model improved to 93.1%, surpassing the original YOLOv8s model, while the computational complexity was reduced by 19% (5.4 GFLOPs), and the model size was decreased by 25.3% (5.7 MB). And compared to other state-of-the-art detection models, the overall performance of our model shows its superiority. We also perform experiments on other datasets to verify the adaptability of our model. The experimental results on the Fish Market dataset indicate that our model has better overall performance than the original model and has good generality.
- 6 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (17) , 3547
Open AccessHRA-YOLO: An Effective Detection Model for Underwater Fish
In intelligent fisheries, accurate fish detection is essential to monitor underwater ecosystems. By utilizing underwater cameras and computer vision technologies to detect fish distribution, timely feedback can be provided to staff, enabling effective fishery management. This paper proposes a lightweight underwater fish detection algorithm based on YOLOv8s, named HRA-YOLO, to meet the demand for a high-precision and lightweight object detection algorithm. Firstly, the lightweight network High-Performance GPU Net (HGNetV2) is used to substitute the backbone network of the YOLOv8s model to lower the computational cost and reduce the size of the model. Second, to enhance the capability of extracting fish feature information and reducing missed detections, we design a residual attention (RA) module, which is formulated by embedding the efficient multiscale attention (EMA) mechanism at the end of the Dilation-Wise Residual (DWR) module. Then, we adopt the RA module to replace the bottleneck of the YOLOv8s model to increase detection precision. Taking universality into account, we establish an underwater fish dataset for our subsequent experiments by collecting data in various waters. Comprehensive experiments are carried out on the self-constructed fish dataset. The results on the self-constructed dataset demonstrate that the precision of the HRA-YOLO model improved to 93.1%, surpassing the original YOLOv8s model, while the computational complexity was reduced by 19% (5.4 GFLOPs), and the model size was decreased by 25.3% (5.7 MB). And compared to other state-of-the-art detection models, the overall performance of our model shows its superiority. We also perform experiments on other datasets to verify the adaptability of our model. The experimental results on the Fish Market dataset indicate that our model has better overall performance than the original model and has good generality.
- Open Access
- 24 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 43-46
Video stitching method utilizing mesh segmentation
In this paper, we propose a video stitching method based on mesh segmentation to solve the problems of temporal and spatial artifacts caused by video shake and insufficient motion estimation at enlarged views. The proposed method constructs a unified framework for video stitching and stabilization, which aims to perform simultaneous video stitching and stabilization optimization. Firstly, the optical flow grid method was used to optimize the feature points, so that the effective feature points could also be extracted from the low texture area, and the feature points were optimized to concentrate the feature points in the overlapping area. Then, the camera motion model is estimated by local matrix homography estimation, camera path estimation, and camera path smoothing, and the cumulative error of video frames is reduced by adding motion estimation compensation, to realize the high-quality video stitching task. By designing a comparative experiment, the proposed method is qualitatively and quantitatively evaluated from three aspects: stability score, stitching score, and video stitching quality.
- 24 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 43-46
Open AccessVideo stitching method utilizing mesh segmentation
In this paper, we propose a video stitching method based on mesh segmentation to solve the problems of temporal and spatial artifacts caused by video shake and insufficient motion estimation at enlarged views. The proposed method constructs a unified framework for video stitching and stabilization, which aims to perform simultaneous video stitching and stabilization optimization. Firstly, the optical flow grid method was used to optimize the feature points, so that the effective feature points could also be extracted from the low texture area, and the feature points were optimized to concentrate the feature points in the overlapping area. Then, the camera motion model is estimated by local matrix homography estimation, camera path estimation, and camera path smoothing, and the cumulative error of video frames is reduced by adding motion estimation compensation, to realize the high-quality video stitching task. By designing a comparative experiment, the proposed method is qualitatively and quantitatively evaluated from three aspects: stability score, stitching score, and video stitching quality.
- Open Access
- 25 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-7
A Binocular Vision Stereo Matching Algorithm for Depth Map Estimation
A new binocular stereo matching algorithm is proposed to solve the problem that the existing stereo matching algorithm is susceptible to noise and pixel mismatch and wrong matching in weak texture areas. First, set the search window and the neighborhood window, the neighborhood window slides inside the search window, compare the similarity between the neighborhood window and the Census transformation window, and determine the degree of influence of the center pixel of the neighborhood window relative to the center pixel of the transformation window according to the similarity, and assign the weight to process the gray value of the pixel. Secondly, the Sobel operator is introduced to calculate the gradient value of the image, and the texture level of the image is divided according to the gradient value. Three thresholds are set according to the texture level, and the parallax is aggregated with the cross-domain algorithm. Finally, the disparity map is optimized by the least squares fitting plane method. Experiments show that the average mismatch rate of the disparity map processed by the algorithm is 5.39% in the test of randomly selected images. Compared with other algorithms, it is more robust to noise, smoother in low-texture areas, has higher matching accuracy of disparity maps, and better optimization of disparity maps.
- 25 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessA Binocular Vision Stereo Matching Algorithm for Depth Map Estimation
A new binocular stereo matching algorithm is proposed to solve the problem that the existing stereo matching algorithm is susceptible to noise and pixel mismatch and wrong matching in weak texture areas. First, set the search window and the neighborhood window, the neighborhood window slides inside the search window, compare the similarity between the neighborhood window and the Census transformation window, and determine the degree of influence of the center pixel of the neighborhood window relative to the center pixel of the transformation window according to the similarity, and assign the weight to process the gray value of the pixel. Secondly, the Sobel operator is introduced to calculate the gradient value of the image, and the texture level of the image is divided according to the gradient value. Three thresholds are set according to the texture level, and the parallax is aggregated with the cross-domain algorithm. Finally, the disparity map is optimized by the least squares fitting plane method. Experiments show that the average mismatch rate of the disparity map processed by the algorithm is 5.39% in the test of randomly selected images. Compared with other algorithms, it is more robust to noise, smoother in low-texture areas, has higher matching accuracy of disparity maps, and better optimization of disparity maps.