WYWeili Yang

Weili Yang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Weili Yang has a total of 27 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Study of Full-View Finger Vein Biometrics on Redundancy-Compensated 3D Reconstruction and Frequency-Spatial Coupling Feature Extraction

    by Junduan Huang, Dacan Luo, Alexander Unnervik, Weili Yang, Massimo Tistarelli, Wenxiong Kang

    Full-view 3D finger vein (3DFV) biometrics is a significant advancement in vein biometrics. Nevertheless, its progress is fundamentally constrained by the limited robustness of 3D reconstruction. This challenge stems primarily from the inherent limitations of consumer-grade capture devices, which introduce imaging errors that are virtually impossible to fully circumvent. Meanwhile, how to adequately and efficiently extract features from the reconstructed 3DFV is also a direction that has not been thoroughly explored. To address these challenges, this paper introduces a robust 3DFV reconstruction method with a quality assessment mechanism and an adaptive feature extraction network. For reconstruction, we propose Redundancy-Compensated 3DFV (RC-3DFV), which incorporates a redundancy compensation term during reconstruction to mitigate imaging errors from consumer-grade devices. The reconstructed 3DFVs are unfolded into 3D finger texture (3DFT) and 3D finger shape (3DFS) maps to reduce dimensionality. A fast 3DFV Quality Assessment (3DFV-QA) mechanism is then applied to select the optimal reconstruction result. For feature extraction, we present 3DFVFSNet, an enhanced version of FVFSNet that effectively couples frequency-domain and spatial-domain convolutions. This design adequately captures both the global characteristics of 3DFS maps and the local details of 3DFT maps. We have conducted a series of rigorous experiments to evaluate the effectiveness of the proposed methods, including the 3DFV reconstruction experiment and 3DFV authentication experiment. Experimental results demonstrate the competent performance and efficiency of our proposed methods.

  • Open Access

    DFANet: A Lightweight Dual-Branch Feature-Aggregation Network for Metal Surface Defect Detection

    by Hongshan Gou, Yueji Wang, Fujian Feng, Fei Wang, Mian Tan, Weili Yang

    Object detection is pivotal for automated industrial surface defect detection. However, industrial imaging of metal surfaces is often hampered by adverse factors such as uneven illumination, strong specular reflection, low contrast, and blurring of tiny defects. These issues can significantly impair the feature extraction capability of mainstream detection networks and hinder their real-time deployment on edge imaging devices. To address the aforementioned issues, this paper proposes a lightweight Dual-branch Feature-Aggregation Network (DFANet). Aiming at excessive parameters and redundant computation of traditional backbones, we design a lightweight Dual-branch Feature Alignment Fusion (DFAF) backbone. The dual-stream structure separately extracts shallow texture imaging features and deep semantic imaging features of metal surfaces, and the embedded Feature Alignment Fusion (FAF) module eliminates spatial offsets of multi-scale imaging features caused by illumination interference. Furthermore, a Parallel Spatial Deformable Module (PSDM) is integrated into the backbone to dynamically adjust receptive fields via deformable attention, strengthening feature characterization for irregular defects under low-quality imaging conditions. In the neck part, a Reparameterized Context Interaction Block (RCIB) is constructed to filter redundant background texture information from multi-layer imaging feature maps, further reducing memory and computation consumption. Comprehensive experiments are carried out on three public metal surface imaging datasets: NEU-DET, X-SDD, and GC10-DET. The DFANet achieves mAP values of 83.8%, 76.6%, and 73.1%, respectively. Compared with the YOLO11m baseline, our DFANet reduces the number of parameters by 20.7% and the computational cost (GFLOPs) by 65.9%. Extensive comparisons with state-of-the-art detection methods and ablation studies demonstrate that DFANet achieves a favorable trade-off between detection precision and lightweight performance. The proposed network offers an efficient solution for real-time automatic defect inspection on industrial metal production lines.

  • Open Access

    Multisource Disentangled Transfer Learning for Generalizable Finger Verification

    by Weili Yang, Xing Li, Junduan Huang, Fei Zhong, Dacan Luo, Fujian Feng, Wenxiong Kang

    Finger biometrics has attracted widespread attention due to the convenience of finger traits and their potential wide application value. Most of the latest methods, especially those based on deep learning, tend to have high training data requirements to obtain an effective and robust recognition model. However, in real-world scenarios, collecting sufficient training data for each newly designed finger verification device is very time consuming. Therefore, obtaining an optimal model in the case of few-shot data is a recent research hotspot. Common approaches for solving this few-shot learning problem with multisource datasets use fine-tuning or joint training. Fine-tuning from multisource datasets suffers from catastrophic forgetting, and joint training discards the common features. To solve these problems, we propose a multisource disentangled transfer learning(MDTL) method to disentangle the common and domain-specific features by feature space normalization, distribution discrepancy measurement and feature transform network (FTN). In this way, we extract the common knowledge from multisource datasets and then transfer it to the target dataset to improve the performance. Experimental results show that the proposed MDTL method can obtain stable and efficient verification models with only a small number of samples.

  • Open Access

    Pose-Robust Finger Identification Based on Mutual Region Alignment and Minimum Convolution Point Feature

    by Weili Yang, Fujian Feng

    To address the performance degradation of finger identification systems caused by longitudinal rotation during contactless finger image acquisition, this paper proposes a novel pose-robust finger identification framework integrating mutual region extraction, minimum convolution point feature learning, and weighted score fusion. Specifically, longitudinal rotation of fingers easily leads to inconsistent imaging regions and non-rigid deformations, which severely undermine the stability of feature matching. To tackle these challenges, we propose a three-stage framework: First, a mutual region extraction strategy is proposed to adaptively align inconsistent imaging areas via vertical gradient calculation and minimum distance metric, effectively eliminating content discrepancies caused by longitudinal rotation. Second, a minimum convolution point feature learning module is designed to model non-linear correlations between cross-regional features, coupled with a deformation-aware matching algorithm to enhance robustness against finger deformation, mitigating the adverse impact of pose variations. Finally, a weighted score fusion method is introduced to integrate matching results from minimum convolution point features and Uniform Local Binary Pattern(LBP) features, fully exploiting the complementary information between structural and texture features. Extensive experiments are conducted on four public datasets, SDUMLA\_HMT, MMCBNU\_6000, FV-USM, and LFMB-3DFB, and the results demonstrate that the proposed method achieves state-of-the-art (SOTA) performance in terms of recognition accuracy and robustness against longitudinal rotation and finger deformation.

  • Open Access

    Pose-Robust Finger Identification based on Mutual Region Alignment and Minimum Convolution Point Feature

    by Weili Yang, Dewen Kong, Fujian Feng, Junduan Huang, Yonghuai Yang, Wenxiong Kang, Dacan Luo

    To address the performance degradation of finger identification systems caused by longitudinal rotation during contactless finger image acquisition, this paper proposes a novel pose-robust finger identification framework integrating mutual region extraction, minimum convolution point feature learning, and weighted score fusion. Specifically, longitudinal rotation of fingers easily leads to inconsistent imaging regions and non-rigid deformations, which severely undermine the stability of feature matching. To tackle these challenges, we propose a three-stage framework: First, a mutual region extraction strategy is proposed to adaptively align inconsistent imaging areas via vertical gradient calculation and minimum distance metric, effectively eliminating content discrepancies caused by longitudinal rotation. Second, a minimum convolution point feature learning module is designed to model non-linear correlations between cross-regional features, coupled with a deformation-aware matching algorithm to enhance robustness against finger deformation, mitigating the adverse impact of pose variations. Finally, a weighted score fusion method is introduced to integrate matching results from minimum convolution point features and Uniform Local Binary Pattern(LBP) features, fully exploiting the complementary information between structural and texture features. Extensive experiments are conducted on four public datasets, SDUMLA\_HMT, MMCBNU\_6000, FV-USM, and LFMB-3DFB, and the results demonstrate that the proposed method achieves state-of-the-art (SOTA) performance in terms of recognition accuracy and robustness against longitudinal rotation and finger deformation.

  • RSNet: Region-Specific Network for Contactless Palm Vein Authentication

    by Dacan Luo, Junduan Huang, Weili Yang, M. Saad Shakeel, Wenxiong Kang

    More palm features, such as veins and shapes obtained from an enlarged contactless palm vein region of interest (ROI), have been shown to improve recognition performance. However, a few efforts have been made to adequately utilize these features for mining identity information. To address this issue, we propose a Region-Specific Network (RSNet) for contactless palm vein authentication. Our RSNet is a dual-branch structure for global and local feature extraction. Firstly, a Region-based Local feature Enhancement Block (RLEB) is proposed at the local branch to extract region-specific features. In the RLEB, the intermediate feature maps are divided into three asymmetrical patches based on the physiological characteristics of palm vein and palm shape for extracting diversified features, enhancing the local feature representation. Then, a Multi-scale Aggregation Block (MAB) is proposed that efficiently aggregates multi-scale features at a more granular level. Furthermore, to guide the global and local branches in learning complementary feature aspects, a difference loss is introduced to apply a soft subspace orthogonality constraint between the global and local vectors during training. The global branch is designed to assist the learning process of local features, without being adopted for inference. Extensive experiments have demonstrated the effectiveness and superiority of our method, and the RSNet achieves new State-Of-The-Art (SOTA) authentication performance on seven public contactless palm vein databases in the open-set scenario.


  • FVFSNet: Frequency-Spatial Coupling Network for Finger Vein Authentication

    by Junduan Huang, An Zheng, M. Saad Shakeel, Weili Yang, Wenxiong Kang

    Finger vein biometrics is becoming an important source of human authentication due to its advantages in terms of liveness detection, high security, and user convenience. Although there exist a lot of deep learning-based methods for finger vein authentication, they only extract features from finger vein images in the spatial domain and may lose some important information that is present in other domains, such as the frequency domain. Motivated by this conjecture and the remarkable performance of image feature extraction in the frequency domain, this work explores a method capable of extracting finger vein features in both the spatial and frequency domains. Therefore, the features extracted from different domains can complement each other. In addition, we propose a novel frequency-spatial coupling network (FVFSNet) for finger vein authentication. FVFSNet is mainly composed of three parts: (1) the frequency domain processing module (FDPM), (2) the spatial domain processing module (SDPM), and (3) the frequency-spatial coupling module (FSCM). The FDPM is used to extract the finger vein features present in the frequency domain, which is mainly composed of the frequency-spatial domain transformation and the frequency domain convolution layer. The SDPM is used to extract the finger vein features present in the spatial domain, which is mainly composed of convolution layers with an efficient design. The FSCM is used to couple the features extracted from the FDPM and SDPM, which is mainly composed of the channel and spatial attention mechanisms. To validate our conjecture and the performances of FVFSNet, extensive experiments are conducted on nine commonly used publicly available finger vein datasets. Experimental results show that the frequency domain constitutional neural network has a surprising effect on finger vein authentication, and the proposed FVFSNet achieves the state-of-the-art performance with the advantages of lightweight and low computational cost.


  • A Novel System and Experimental Study for 3D Finger Multibiometrics

    by Weili Yang, Zhuoming Chen, Junduan Huang, Wenxiong Kang

    Fingers contain various discriminative biometric traits, such as fingerprint, finger vein, finger knuckle, finger shape, and so on, which are complementary in identity information. However, only one or a few traits are used in most current research and practical applications, while others are ignored, resulting in degraded recognition performance and vulnerability to forgery. In this paper, we make the first attempt to collect and study all biometric traits on the finger. Firstly, a novel multi-view, multi-spectral 3D finger imaging system is proposed. To the best of our knowledge, it is the first biometric imaging system capable of capturing almost all finger-based traits. We scanned numerous fingers with this 3D finger imaging system, obtaining external skin images and internal vein images from 6 different views. The proposed 3D finger reconstruction and texture mapping algorithms are then used to generate 3D finger models with skin and vein textures. Second, we establish a benchmark dataset, namely the Large-scale Finger Multi-Biometric database and benchmark for 3D Finger Biometrics (LFMB-3DFB). The LFMB-3DFB contains 695 fingers, and each finger is acquired 10 times, yielding 6 finger skin images and 6 finger vein images for a total of 83,400 images and 6,950 3D finger models. Finally, we design a more scientific and comprehensive evaluation protocol to conduct extensive experimental research and analysis on this database for both subject-independent verification and subject-independent close-set identification tasks. Comprehensive and rigorous experiments for 2D finger traits recognition, multi-view finger traits recognition, 3D finger traits recognition, and score-level fusion on LFMB-3DFB are carried out, and excellent results are achieved. The LFMB-3DFB database will be released at https://github.com/SCUT-BIP-Lab/LFMB-3DFB to promote 3D finger multi-biometric research using cutting-edge imaging techniques.


  • FVT: Finger Vein Transformer for Authentication

    by Junduan Huang, Weijian Luo, Weili Yang, An Zheng, Fengzhao Lian, Wenxiong Kang

    In recent years, deep learning-based finger vein (FV) authentication has attracted the attention of biometric researchers and achieved breakthrough results. Previously, convolutional neural networks (CNNs) were the most commonly used deep learning-based methods for FV authentication. Recently, the vision Transformer (ViT)-based method has started getting attention from the research community due to its excellent performance in many computer vision tasks. In this article, we delve into ViTs and propose a novel model, FV Transformer (FVT), for FV authentication. The FVT consists of four key modules: 1) the conditional position embedding, which is capable of dynamically generating position codes according to the input FV tokens; 2) the weight-shared expanded multilayer perceptron (EMLP), which helps to extract richer and more robust token information; 3) the local information-enhanced feedforward network (FFN), which enhances the ability of local information extraction; and 4) the expansion-less mechanism (ELM) for aggregating adjacent FV tokens, which implements the pyramid structure, and hence, the multilevel feature extraction capability is introduced to the Transformer architecture, which originally focuses on global information. To fully validate the performance and generalization of FVT, experiments were conducted on nine publicly available FV datasets. The effectiveness of each key module of FVT is demonstrated in the ablation experiments. Also, the comparative experiments show that the FVT outperforms several baseline Transformer models and achieves competitive performance when compared with the state-of-the-art (SOTA) FV authentication methods.