Zhang Huiying

Dr. Zhang Huiying

orcid

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
8 February 2025 in Scientific Reports
Verified
Journal Article
Open access
PDF file
ZHZhang Huiying
Zhang Huiying
PXPan Xiao
Pan Xiao
FYFeifan Yao
Feifan Yao
QZQinghua Zhang
Qinghua Zhang
YGYifei Gong
Yifei Gong
In view of the complex environments and varying object scales in drone-captured imagery, a novel PARE-YOLO algorithm based on YOLOv8 for small object detection is proposed. This model enhances feature extraction and fusion across multiple scales through a restructured neck network. Additionally, it incorporates a lightweight detection head that is optimized for small objects, thereby significantly improving detection performance in cluttered and intricate backgrounds. To further enhance the extraction of small object features, the conventional C2f is replaced with a novel architecture. Moreover, the EMA-GIoU loss function is proposed to mitigate class imbalance and enhance robustness, particularly in scenarios characterized by skewed class distributions. Evaluation on the VisDrone2019 dataset indicates that PARE-YOLO achieves a 5.9% improvement in mean Average Precision (mAP) at a threshold of 0.5, compared to the original YOLOv8 model. In addition, the PARE-YOLO model exhibits significant robustness, achieving a mean Average Precision (mAP) at a threshold of 0.5 values on the HIT-UAV dataset that are 0.8%, 0.5%, and 1.2% higher than those of YOLOv8, YOLOv10, and RT-DETR. These results underscore the effectiveness of PARE-YOLO in addressing the challenges inherent in aerial scenarios. The code will be available online (https://github.com/Sunnyxiao69/PARE-YOLO).

23 October 2024 in IEEE Photonics Technology Letters
Verified
Journal Article
Closed access
FYFeifan Yao
Feifan Yao
ZHZhang Huiying
Zhang Huiying
YGYifei Gong
Yifei Gong
Addressing the unique optical properties of water in underwater images, this paper introduces the DifSG2-CCL model for generating images in complex underwater environments, aiming to mitigate the effects of water quality factors on the generated images. This paper proposes U-CCL (Underwater Cycle Consistency Loss) in the generator loss, allowing the generator to preserve real image information during conversion by reflecting the shot to prevent information loss. Consequently, the generated image is not only more realistic, but also highly consistent with the real image in content. Additionally, this paper utilizes the publicly available 9.235k Sea Anemone Dataset (SA Dataset) with a resolution of 256×256 for training. Experimental results indicate that assigning a weight of 1 to DiffSG2-CCL achieves the best training effect, reducing the FID value to 8.97, while significantly improving the detail and texture of the generated images, approaching aesthetic vision. Thus, this method effectively mitigates the special optical properties of water bodies and offers innovative approaches for generating images in complex underwater environments. The experimental code with pre-trained models will be published shortly at https://github.com/yff0428/DifSG2-CCL/tree/master.

25 March 2025 in Complex & Intelligent Systems
Verified
Journal Article
Open access
PDF file
FYFeifan Yao
Feifan Yao
ZHZhang Huiying
Zhang Huiying
YGYifei Gong
Yifei Gong
QZQinghua Zhang
Qinghua Zhang
PXPan Xiao
Pan Xiao
Aiming at the influence of factors such as the special optical characteristics of water bodies on the perceptual quality of generated images, this paper proposes the DifSG2-CCL model for reducing the special optical characteristics of water bodies and the DPL-SG2 model for introducing perceptual loss. Combining the ideas of cyclic consistency and style migration, this paper builds the Underwater Cycle Consistency Loss (U-CCL) module. The DifSG2-CCL model is based on the method of image reconstruction, which converts the underwater image into the style of the land image to reduce the influence of the water body factors. VGG16 is introduced as a perceptual loss into the DPL-SG2 to enhance the visual perception of the image by feature extraction with different layers and tonal weighting. Furthermore, in addition to the already disclosed SA dataset, a T dataset with a resolution of 256 × 256 in 9.366k sheets is provided in this paper. The experimental results show that DifSG2-CCL and DPL-SG2 can effectively enhance the perceptual quality of the images. The unique underwater image generation of DifSG2-CCL produces excellent results in qualitative analysis and reduces its FID value to 8.97. DPL-SG2 is more outstanding in the training of T dataset, and its FID value is reduced to 5.39. Therefore, the U-CCL and VGG16 can be applied as an innovative approach to enhance visual perception of underwater images. The experimental code with pre-trained models will be published shortly at https://github.com/yff0428/DPL-SG2/tree/main.

2 March 2026 in Optics & Laser Technology
Verified
Journal Article
Closed access
ZHZhang Huiying
Zhang Huiying
QZQipeng Zhong
Qipeng Zhong
JLJiahui Liu
Jiahui Liu
RCRuibo Chen
Ruibo Chen
ZWZhenyu Wang
Zhenyu Wang

18 December 2023 in IEEE Access
Verified
Journal Article
Open access
ZHZhang Huiying
Zhang Huiying
YGYifei Gong
Yifei Gong
FYFeifan Yao
Feifan Yao
QZQinghua Zhang
Qinghua Zhang
A novel XM-YOLOViT real-time detection algorithm for pedestrians and vehicles in foggy days based on YOLOV5 framework is proposed, which effectively solves the problems of dense target interference and obscuration by haze, and improves the detection effect in complex foggy environments. Firstly, Inverted Residual Block and MobileViTV3 Block are introduced to construct XM-net feature extraction network, secondly, EIOU is used as a location loss function and a high resolution detection layer is added in the Neck region. In terms of data, a nebulization method is designed to map images from fogless space to foggy space based on the atmospheric scattering model and the dark channel prior. Finally, the validity of verified on four datasets with different foggy environments, respectively. The experimental results show that the accuracy, recall, and mAP of the XM-YOLOViT model are respectively 54.95%, 41.93%, and 43.15%, with an F1-Score of 0.474, and the accuracy improved by 3.42%, recall improved by 7.08%, mAP improved by 7.52%, and F1 score improved by 13.94% compared to the baseline model, while model parameters are decreased by 41.7% to 4.09M, FLOPs of 25.2G, and detection speed of 70.93 FPS. The XM-YOLOViT model performs better than the advanced YOLO detectors, the F1-Score and mAP are improved by 5.57% and 3.65% compared with YOLOv7-tiny, and 2.38%, 2.37% respectively compared with YOLOv8s. Therefore, the XM-YOLOViT algorithm proposed in this article has high detection accuracy and an extremely lightweight structure, which can effectively improve the efficiency and quality of UAV foggy weather detection tasks, especially for extremely small targets. Our source code is available at: https://github.com/AFeiV8/XM-YOLOViT.

23 February 2024 in IEEE Access
Verified
Journal Article
Open access
ZHZhang Huiying
Zhang Huiying
FYFeifan Yao
Feifan Yao
YGYifei Gong
Yifei Gong
QZQinghua Zhang
Qinghua Zhang
Given the complexity and uncertainty of the underwater environment, it is of great importance to generate realistic and high-quality images. In this paper, we propose six unconditional generative models based on the Diffusion-Styegan2 generative model, incorporating Wasserstein, R2 regularization terms, and other techniques for marine biology image generation. The Wasserstein distance technique is used in the loss part of Diffusion-Styegan2, combined with the back propagation algorithm to compute the gradient in the neural network while retaining the computational map to improve the training efficiency and training stability; the R2 regularization term is used to introduce the r2 hyperparameter, and the L2 regularization technique is used based on the original R1 regularization term to regularize the gradient of the discriminator to improve the training and generation performance of the model; the ADA technique is used based on DWBG-Stylegan2 to further improve the quality and stability of the generated images. In addition, a set of SA datasets (sea anemone datasets) with a resolution of 256*256 is proposed in this paper. The experimental results show that the FID value of Diffusuion-Stylegan2 is 10.31, the value of FID of DWBG-Stylegan2 is 8.32, the value of FID of Diffusion-Stylegan2-R2 is 9.58, and the optimal FID value of this experiment is achieved by DWBG-Stylegan2-ADA with a value of 5.67, which is considerably lower compared to the FID value of Diffusion-Stylegan2. Therefore, techniques such as Wasserstein and R2 regularization terms can effectively generate more realistic images of marine organisms. Meanwhile, this experiment provides new ideas and methods for the construction of the unconditional generative model.