Huanzun Zhang

Mr. Huanzun Zhang

Anhui University, Hefei, Anhui, China
orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
27 July 2025 in Electronics
Verified
Journal Article
Open access
PDF file
HLHaomin Li
Haomin Li
HZHuanzun Zhang
Huanzun Zhang
WZWenke Zang
Wenke Zang
Recent advances in precision manufacturing and high-end equipment technologies have imposed ever more stringent requirements on the accuracy, real-time performance, and lightweight design of online steel strip surface defect detection systems. To reconcile the persistent trade-off between detection precision and inference efficiency in complex industrial environments, this study proposes StripSurface–YOLO, a novel real-time defect detection framework built upon YOLOv8n. The core architecture integrates an Efficient Cross-Stage Local Perception module (ResGSCSP), which synergistically combines GSConv lightweight convolutions with a one-shot aggregation strategy, thereby markedly reducing both model parameters and computational complexity. To further enhance multi-scale feature representation, this study introduces an Efficient Multi-Scale Attention (EMA) mechanism at the feature-fusion stage, enabling the network to more effectively attend to critical defect regions. Moreover, conventional nearest-neighbor upsampling is replaced by DySample, which produces deeper, high-resolution feature maps enriched with semantic content, improving both inference speed and fusion quality. To heighten sensitivity to small-scale and low-contrast defects, the model adopts Focal Loss, dynamically adjusting to sample difficulty. Extensive evaluations on the NEU-DET dataset demonstrate that StripSurface–YOLO reduces FLOPs by 11.6% and parameter count by 7.4% relative to the baseline YOLOv8n, while achieving respective improvements of 1.4%, 3.1%, 4.1%, and 3.0% in precision, recall, mAP50, and mAP50:95. Under adverse conditions—including contrast variations, brightness fluctuations, and Gaussian noise—SteelSurface-YOLO outperforms the baseline model, delivering improvements of 5.0% in mAP50 and 4.7% in mAP50:95, attesting to the model’s robust interference resistance. These findings underscore the potential of StripSurface–YOLO to meet the rigorous performance demands of real-time surface defect detection in the metal forging industry.

27 July 2025 in Biomimetics
Verified
Journal Article
Open access
PDF file
TYTianzuo Yuan
Tianzuo Yuan
HZHuanzun Zhang
Huanzun Zhang
JJJie Jin
Jie Jin
ZCZhebo Chen
Zhebo Chen
SCShanshan Cai
Shanshan Cai
Superb Fairy-wren Optimization Algorithm (SFOA) is an animal-based meta-heuristic algorithm derived from Fairy-wren’s behavior of growing, feeding, and avoiding natural enemies. The SFOA has some shortcomings when facing complex environments. Its switching mechanism is not enough to adapt to complex optimization problems, and it faces a weakening of population diversity in the late stage of optimization, leading to a higher possibility of falling into local optima. In addition, its global search ability needs to be improved. To address the above deficiencies, this paper proposes an Adaptive Superb Fairy-wren Optimization Algorithm (ASFOA). To assess the ability of the proposed ASFOA, three groups of experiments are conducted in this paper. Firstly, the effectiveness of the proposed improved strategies is checked on the CEC2018 test set. Second, the ASFOA is compared with eight classical/highly cited/newly proposed metaheuristics on the CEC2018 test set, in which the ASFOA performed the best overall, with average rankings of 1.621, 1.138, 1.483, and 1.966 in the four-dimensional cases, respectively. Then the convergence and robustness of ASFOA is verified on the CEC2022 test set. The experimental results indicate that the proposed ASFOA is a competitive metaheuristic algorithm variant with excellent performance in terms of convergence and distribution of solutions. In addition, we further validate the ability of ASFOA to solve real optimization problems. The average ranking of the proposed ASFOA on 10 engineering constrained optimization problems is 1.500. In summary, ASFOA is a promising variant of metaheuristic algorithms.

14 July 2025 in IEEE Geoscience and Remote Sensing Letters
Verified
Journal Article
Closed access
SHShuting Huang
Shuting Huang
GZGe Zhang
Ge Zhang
HZHuanzun Zhang
Huanzun Zhang
HXHui Xu
Hui Xu
GYGuangzhen Yao
Guangzhen Yao
JKJun Kong
Jun Kong
In recent years, hybrid convolutional neural networks (CNNs) and Transformer-based object detection technologies have achieved remarkable success. In the field of remote sensing image detection, since remote sensing systems rely on large-scale deployment of edge devices, detection models need to be lightweight with low parameter complexity to adapt to resource-constrained environments. However, existing lightweight models often struggle with an imbalance in extracting low-frequency global and high-frequency local information. In particular, when processing high-frequency local information (such as edges, textures, and fine structures), these models often lack in-depth analysis, leading to insufficient extraction of local features and reduced detection accuracy. To address the imbalance between low-frequency global information and high-frequency local information in lightweight remote sensing models, we propose an efficient and lightweight hybrid network detection framework, which mainly consists of the Global-Local Balance (GLB) module and the Detail-Aware Feature Fusion (DAFF) module. The GLB module adopts dynamic weight adjustment and context-aware mechanisms to effectively aggregate high-frequency local information in the image. The DAFF module further enhances feature fusion and detail refinement, improving the model’s performance and generalization ability. Experimental results on remote sensing datasets, including RSOD, NWPU VHR-10, and LEVIR, demonstrate that our proposed method achieves a well-balanced trade-off between model size and detection accuracy, reaching state-of-the-art performance.

20 June 2025
Verified
Conference Paper
Closed access
HZHuanzun Zhang
Huanzun Zhang
Vehicle recognition plays a vital role in intelligent transportation systems (ITS), contributing to traffic flow monitoring, vehicle tracking, and smart city infrastructure. With the rise of deep learning, Convolutional Neural Networks (CNNs) and Transformer-based architectures have become dominant solutions for image classification tasks. In this paper, we conduct a comprehensive comparison of seven representative models—five CNN-based models (VGG16, ResNet50, MobileNetV2, ConvNeXt-Tiny, InceptionNeXt-Tiny) and two Transformer-based models (ViT-Tiny, Swin-Tiny)—on a $\mathbf{1 0}$-class vehicle image dataset. All models are fine-tuned using ImageNet-pretrained weights and trained under a consistent experimental protocol. We evaluate their performance using class-wise metrics and analyze model complexity. Furthermore, we investigate misclassification patterns by generating Grad-CAM visualizations on the Top-5 misclassified images per model. Our findings offer insights into model selection trade-offs between accuracy and efficiency, and highlight key differences in decision-making mechanisms between CNN and Transformer architectures. The results provide a solid reference for future deployment of vision models in real-world ITS applications.

11 April 2025
Verified
Conference Paper
Closed access
MWMingtao Wang
Mingtao Wang
HZHuanzun Zhang
Huanzun Zhang
Existing research methods have not been able to accurately predict the aging failure of power electronic equipment, we propose an accuracy improvement method, the goal of this study is to predict the age group and RUL of power electronic equipment more accurately. To fill the gap of existing models, this paper proposes a remaining life prediction model based on Long Short-Term Memory (LSTM) network, which is chosen as the central element of time-series neural network because of its ability to effectively capture long-term dependencies in long-term series. Using the optimization of traditional LSTM model, we determine and propose adaptive gating mechanism improved model which dynamically adjusts learning strategy of the network by device state change, and enhances the model fitting ability for nonlinear recession process. Furthermore, we develop a multilayer LSTM network and use a hybrid loss function by integrating mean square error (MSE) with absolute error (MAE) to enhance the predictive performance of the model in different failure periods. The experimental results demonstrate that the improved LSTM model has higher accuracy and stronger robustness than the traditional LSTM and other benchmark models in the RUL prediction task of multiple power electronic devices.