Yanwei Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yanwei Wang has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- De Li
Yanbian University
4 shared publications - Xun Jin
Yanbian University
3 shared publications - Xuanyou Li
Information Engineering Major , College of Communication Engineering , Jilin University , Changchun , China , jlu.edu.cn
2 shared publications - Fang Gao
Smart Agriculture Research Institute, Jilin Agricultural University, Changchun 130118, China
1 shared publication - Helong Yu
Institute for the Smart Agriculture, Jilin Agricultural University, Changchun 130118, China
1 shared publication Jiang Gui
Department of Computer Science and Technology, Yanbian University, Yanji, China
1 shared publication
Latest Publications
- Open Access
MobileNet-PAS: A Lightweight Deep Learning Model for Non-Destructive Grade Identification of Pilose Antler Slices
Pilose antler is a valuable traditional Chinese medicinal herb. Slices of different grades show great disparities in medicinal potency and market price. Conventional visual discrimination of wax slices, powder slices, gauze slices and bone slices relies entirely on human inspectors. Such manual evaluation heavily depends on personal expertise and brings prominent subjective bias. Especially for gauze slices and powder slices with highly similar visual appearances, misclassification frequently occurs, making it hard to satisfy practical requirements for large-scale rapid inspection. Accordingly, this paper develops a modified lightweight convolutional network named MobileNet-PAS to realize intelligent grading of pilose antler slices. Built upon the MobileNetV4-small backbone, the proposed model reconstructs the dimensionality-reduction convolution inside UIB blocks using group convolution. The ConvBN layers following the last UIB block are replaced with partial convolution. A SimAM attention module is embedded at the rear of the UIB blocks to boost fine-grained feature representation. The original ReLU6 is further substituted with the PReLU activation function. Meanwhile, a Gap-Aware dynamic learning rate strategy is adopted to optimize parameter updating, so as to strengthen feature distinguishability while restraining model complexity. Experimental results demonstrate that MobileNet-PAS achieves a test-set accuracy of 96.03%, which outperforms the baseline by 3.31 percentage points with reduced parameters and computational cost. Grad-CAM visualization reveals that the improved network suppresses irrelevant background noise and focuses on intrinsic discriminative cues including osseous structure and texture within slices, yielding better classification performance for easily confused samples. Balancing prediction accuracy and lightweight inference performance, the presented approach offers a feasible algorithm for non-destructive rapid identification of pilose antler slices and delivers references for intelligent quality assessment of Chinese medicinal materials.
Open AccessMobileNet-PAS: A Lightweight Deep Learning Model for Non-Destructive Grade Identification of Pilose Antler Slices
Pilose antler is a valuable traditional Chinese medicinal herb. Slices of different grades show great disparities in medicinal potency and market price. Conventional visual discrimination of wax slices, powder slices, gauze slices and bone slices relies entirely on human inspectors. Such manual evaluation heavily depends on personal expertise and brings prominent subjective bias. Especially for gauze slices and powder slices with highly similar visual appearances, misclassification frequently occurs, making it hard to satisfy practical requirements for large-scale rapid inspection. Accordingly, this paper develops a modified lightweight convolutional network named MobileNet-PAS to realize intelligent grading of pilose antler slices. Built upon the MobileNetV4-small backbone, the proposed model reconstructs the dimensionality-reduction convolution inside UIB blocks using group convolution. The ConvBN layers following the last UIB block are replaced with partial convolution. A SimAM attention module is embedded at the rear of the UIB blocks to boost fine-grained feature representation. The original ReLU6 is further substituted with the PReLU activation function. Meanwhile, a Gap-Aware dynamic learning rate strategy is adopted to optimize parameter updating, so as to strengthen feature distinguishability while restraining model complexity. Experimental results demonstrate that MobileNet-PAS achieves a test-set accuracy of 96.03%, which outperforms the baseline by 3.31 percentage points with reduced parameters and computational cost. Grad-CAM visualization reveals that the improved network suppresses irrelevant background noise and focuses on intrinsic discriminative cues including osseous structure and texture within slices, yielding better classification performance for easily confused samples. Balancing prediction accuracy and lightweight inference performance, the presented approach offers a feasible algorithm for non-destructive rapid identification of pilose antler slices and delivers references for intelligent quality assessment of Chinese medicinal materials.
- Open Access
- 1 January 2026
- Published by Wiley in International Journal of Intelligent Systems
- Vol. 2026 (1) , 7678934
Panoramic Style Feature Transfer Method Based on CycleGAN and Attention Mechanism
Because of the high cost of panoramic photographs and the small amount of existing panoramic image datasets, research on the panorama‐based deep learning is restricted. In this paper, we propose a method to generate panoramic images based on a CycleGAN network. First, a selective transfer units module is added to carry out selective transmission and conversion of feature maps. Then, a convolutional block attention module is added to strengthen the image feature processing. Style conversion technology is combined to achieve spring and winter style conversion and night and day style conversion, so as to expand the dataset of panoramic images. The experimental results show that the proposed method is effective in panoramic image generation, and the use of panoramic images for style transfer is also effective in dataset expansion.
- 1 January 2026
- Published by Wiley in International Journal of Intelligent Systems
- Vol. 2026 (1) , 7678934
Open AccessPanoramic Style Feature Transfer Method Based on CycleGAN and Attention Mechanism
Because of the high cost of panoramic photographs and the small amount of existing panoramic image datasets, research on the panorama‐based deep learning is restricted. In this paper, we propose a method to generate panoramic images based on a CycleGAN network. First, a selective transfer units module is added to carry out selective transmission and conversion of feature maps. Then, a convolutional block attention module is added to strengthen the image feature processing. Style conversion technology is combined to achieve spring and winter style conversion and night and day style conversion, so as to expand the dataset of panoramic images. The experimental results show that the proposed method is effective in panoramic image generation, and the use of panoramic images for style transfer is also effective in dataset expansion.
- Open Access
- 30 December 2025
- Published by Springer Nature in EURASIP Journal on Audio, Speech, and Music Processing
- Vol. 2026 (1)
MIDI music plagiarism detection method based on feature similarity learning
- 30 December 2025
- Published by Springer Nature in EURASIP Journal on Audio, Speech, and Music Processing
- Vol. 2026 (1)
Open AccessMIDI music plagiarism detection method based on feature similarity learning
- Open Access
- 13 October 2025
- Published by MDPI AG in Electronics
- Vol. 14 (20) , 4011
Video Content Plagiarism Detection Using Region-Based Feature Learning
Due to the continuous increase in copyright infringement cases of video content, the economic losses of copyright owners continue to rise. To improve the efficiency of plagiarism detection in video content, in this paper, we propose region-based video feature learning. The first innovation of this paper lies in the combination of temporal positional encoding and attention mechanisms to extract global features for weakly supervised model training. Self- and cross-attention mechanisms are combined to enhance similar features within and between videos by incorporating position coding to capture timing relationships between video frames. Global classification description is embedded for capturing global spatiotemporal information and combined with a weak supervised loss for model training. The second innovation is the frame sequence similarity calculation, which is composed of Chamfer similarity, coordinate attention mechanism, and residual connection, to aggregate similarity scores between videos. Experimental results show that the proposed method can achieve the mAP of 0.907 on the short video dataset from Douyin. The proposed method outperforms frame-level and video-level features in achieving higher detection accuracy, and further contributes to the improvement of video content plagiarism detection performance.
- 13 October 2025
- Published by MDPI AG in Electronics
- Vol. 14 (20) , 4011
Open AccessVideo Content Plagiarism Detection Using Region-Based Feature Learning
Due to the continuous increase in copyright infringement cases of video content, the economic losses of copyright owners continue to rise. To improve the efficiency of plagiarism detection in video content, in this paper, we propose region-based video feature learning. The first innovation of this paper lies in the combination of temporal positional encoding and attention mechanisms to extract global features for weakly supervised model training. Self- and cross-attention mechanisms are combined to enhance similar features within and between videos by incorporating position coding to capture timing relationships between video frames. Global classification description is embedded for capturing global spatiotemporal information and combined with a weak supervised loss for model training. The second innovation is the frame sequence similarity calculation, which is composed of Chamfer similarity, coordinate attention mechanism, and residual connection, to aggregate similarity scores between videos. Experimental results show that the proposed method can achieve the mAP of 0.907 on the short video dataset from Douyin. The proposed method outperforms frame-level and video-level features in achieving higher detection accuracy, and further contributes to the improvement of video content plagiarism detection performance.
- Open Access
- 1 January 2025
- Published by Wiley in International Journal of Intelligent Systems
- Vol. 2025 (1)
A Robust Watermarking Method for Hyperspectral Images Based on Hybrid Attention Mechanism
Because of the copyright issues of hyperspectral images continue to rise, in this paper, we propose to use a neural network–based watermarking model to protect the copyright. By applying normalization‐based attention module (NAM) to deep dispersed watermarking with synchronization and fusion (DWSF), a NDWSF model is proposed for robust hyperspectral image watermarking. It consists of encoding, decoding, discrimination, and attack modules. The encoding and decoding modules are used for embedding and extracting watermarks. Discrimination module is proposed for improving the quality of watermarked image. The discrimination module and the encoding module are in an adversarial relationship to motivate the encoder to generate watermarks with stronger invisibility. Attack module is employed between embedding and extraction to improve robustness against compression and noise and geometric attacks. In order to more effectively utilize image features for watermarking, a kind of hybrid attention mechanism is employed in embedding and extraction by adding NAM. Experimental results show that the loss convergence and stability in training is improved. The peak signal‐to‐noise ratio of the proposed method is 48.08 dB, higher than other methods about 2.5 dB. The bit error rate of the proposed method is less than 2.5% for various hybrid attacks, showing good robustness.
- 1 January 2025
- Published by Wiley in International Journal of Intelligent Systems
- Vol. 2025 (1)
Open AccessA Robust Watermarking Method for Hyperspectral Images Based on Hybrid Attention Mechanism
Because of the copyright issues of hyperspectral images continue to rise, in this paper, we propose to use a neural network–based watermarking model to protect the copyright. By applying normalization‐based attention module (NAM) to deep dispersed watermarking with synchronization and fusion (DWSF), a NDWSF model is proposed for robust hyperspectral image watermarking. It consists of encoding, decoding, discrimination, and attack modules. The encoding and decoding modules are used for embedding and extracting watermarks. Discrimination module is proposed for improving the quality of watermarked image. The discrimination module and the encoding module are in an adversarial relationship to motivate the encoder to generate watermarks with stronger invisibility. Attack module is employed between embedding and extraction to improve robustness against compression and noise and geometric attacks. In order to more effectively utilize image features for watermarking, a kind of hybrid attention mechanism is employed in embedding and extraction by adding NAM. Experimental results show that the loss convergence and stability in training is improved. The peak signal‐to‐noise ratio of the proposed method is 48.08 dB, higher than other methods about 2.5 dB. The bit error rate of the proposed method is less than 2.5% for various hybrid attacks, showing good robustness.