Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Li Li has a total of 60 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • AHC-NeRF: A utonomous, H igh-Quality Neural Reconstruction of Two-Layer C omplex Nested Transparent Objects

    by Youcheng Cai, Fan Gao, Yibo Zhao, Li Li, Ligang Liu

    Reconstructing transparent objects with high fidelity presents significant challenges due to complex light refraction and reflection. Existing methods rely on intentionally designed patterns observed behind the transparent object to infer the correspondence between rays and the background, thereby improving the precision of the reconstruction. However, they are hindered by a refraction-tracing-based strategy that fails to reconstruct complex nested transparent objects and a tedious view-capture strategy relying on images captured from empirically determined viewpoints. To overcome these obstacles, we propose AHC-NeRF, an autonomous, high-quality neural SDF-based framework designed for reconstructing two-layer complex nested transparent objects. Firstly, our framework combines neural SDF with single-pixel imaging, a reflection-based method, which utilizes point-pair priors as guidance to achieve high-quality reconstruction of both the outer and inner surfaces. Secondly, we propose an adaptive single-pixel imaging method that achieves an acceleration of 1-2 orders of magnitude compared to vanilla single-pixel imaging for the acquisition of point-pair priors. Finally, we introduce a novel view-planning strategy that progressively identifies the viewpoints with the highest information gain throughout the optimization process, thereby achieving high-quality surface reconstruction. Extensive experimental results on both synthetic and real-world datasets demonstrate that AHC-NeRF outperforms state-of-the-art methods.

  • Open Access

    Reinforcement Learning-Based Collaborative Framework for Data Imputation and Fault Diagnosis with Online Segment Missing Data

    by Zhaojing Wang, Tianwei Xu, Bowen Song, Yang Wang, Li Li

    In Industrial Internet of Things environments, issues such as sensor failures often lead to missing data. Compared with random missing, segment missing causes severe temporal discontinuity and loss of critical features. However, most existing methods are designed for random missing patterns and cannot effectively handle such scenarios, resulting in significant decline in fault diagnosis performance.To address this issue, this paper proposes a reinforcement learning-based collaborative framework for data imputation and fault diagnosis to improve diagnostic reliability with segment missing data. First, based on generative adversarial imputation network, a dimension-wise random masking matrix is designed for segment missing, a hint mechanism is introduced to strengthen the discriminator’s constraint on missingness, and a joint loss function that balances observed-data preservation and missing-data reconstruction accuracy is constructed to enhance imputation quality. Second, a strategy based on deep deterministic policy gradient is introduced, in which fault diagnosis performance is used as the reward signal and dimension contribution scores are calculated to guide data imputation, enabling the model to preferentially recover feature dimensions that are more critical for diagnosis. In addition, a progressive two-stage training strategy, consisting of step-by-step pretraining and feedback-driven closed-loop optimization, is designed to reduce training difficulty and improve convergence stability. Finally, experiments conducted on the Huazhong University of Science and Technology bearing dataset and the Southeast University gearbox dataset demonstrate that, under segment missing conditions, the proposed method significantly outperforms existing methods in fault diagnosis accuracy and robustness, and that all proposed modules contribute positively to overall performance.

  • Shape–texture aware multi-source domain adaptation for industrial anomaly detection

    by Yaochong Xie, Li Li, Zhaojing Wang, Yaxi Zhou, Tao Peng, Xinrong Hu

  • Le-Radio: Toward High-Gloss Leather Defect Detection Using Wireless Signal

    by Qing Shu, Zhiyuan Guan, Li Li, Lei Ding, Yongmei Wang, Yuan Wu, Yanjiao Chen

    Leather is among the most widely traded textile materials worldwide, and detecting surface defects is crucial to ensure its quality. The conventional manual inspection of leather surfaces is labor-intensive, time-consuming, and prone to human error. Advances in computer vision have facilitated the development of automated leather surface defect detection. However, vision-based methods remain sensitive to lighting conditions, particularly on high-gloss leather surfaces. In this article, we introduce Le-Radio, an Industrial Internet of Things (IIoT) oriented RF-based system for automated leather defect detection. Le-Radio provides a gloss-robust, ubiquitous, and continuous inspection solution, which is crucial for integrating quality control into smart factory pipelines and advancing the digital transformation of the traditional leather industry. To extract informative representations of leather defects from wireless signals, we employ continuous radio snapshots (CRSs) and refined signal features to differentiate defect patterns. Specifically, we design an imaging algorithm that continuously visualizes defects on moving leather samples using a fixed radar. To enhance Le-Radio’s robustness in diverse environments, we train a transferable model that maintains consistent detection performance across different scenarios. We conduct extensive experiments in three rooms using 120 leather samples to evaluate Le-Radio. Experimental results demonstrate that Le-Radio accurately detects diverse defects in high-gloss leather. Beyond leather inspection, this work establishes Le-Radio as a practical IoT edge sensing node for material surface inspection in the IIoT. By integrating RF-based defect scanning capabilities and a robust radar imaging framework into a unified system, this work establishes RF sensing as a viable and novel paradigm for material surface inspection in the IIoT, particularly in scenarios where optical methods are fundamentally limited.

  • 2CLIP-SAM: Dual-Mask-Guided CLIP and SAM Collaboration for Coating Adhesion Detection on Oriented Silicon Steel

    by Yaxi Zhou, Li Li, Qian Xiang, Guobin Zhang, Zhaojing Wang, Tao Peng, Yongkang Ma, Xinrong Hu

    The quality of the insulating coating on oriented silicon steel directly impacts its magnetic performance and service life. Currently, coating adhesion is assessed manually by measuring the detachment area after a bending test. To improve efficiency and reduce costs, companies plan to use machine vision for coating adhesion testing and evaluation. However, traditional machine vision methods face challenges like few-shot learning and highlight interference. Therefore, we propose a few-shot cross-domain anomaly detection framework, 2CLIP-SAM, based on the collaboration of CLIP and SAM. This framework efficiently locates and segments coating detachment areas under highlight interference with minimal samples. First, we introduce multi-source cross-domain learning and domain adversarial training to build a dual-CLIP localization network for coarse localization of both coating detachment and light reflection areas. Next, we design a dual-mask highlight suppression (MDS) mechanism to mitigate light interference, and use the SAM model for precise segmentation of the detachment areas. We also create a dataset with 165 images of reflective oriented silicon steel surfaces, annotated for highlight and detachment areas at the pixel level. We conducted 1,000 coating adhesion tests using real data from the production line, following the GB/T 2522-2007 standard. The results validate the method and establish qualitative standards.


  • GadNet: Garment Adaptive Network Optimization Dressing Human Animation

    by Jiazhe Miao, Tao Peng, Fei Fang, Xinrong Hu, Li Li

    In virtual try-on, film, and gaming applications, the ability to depict realistic and convincing clothing behavior is crucial. Physics-based simulation demands high computational costs, while learning-based methods are limited to specific types of clothing. Although methods that directly reconstruct from images or monocular videos are more in line with consumer demands, existing approaches focus on capturing static or dynamic geometric features of clothing, neglecting the physical properties of clothing, resulting in unrealistic animations. Therefore, we propose GadNet. We separately reconstruct the body and clothing from the video, and then use our proposed motion sequence optimization module to learn plausible motion sequences, alleviating the frame jitter issue common in existing methods. Subsequently, we design a garment adaptive network. We employ a frequency-domain separation optimization method to process different frequency information of the clothing, thereby better subdividing the clothing mesh in high-frequency folding areas, enhancing the processing performance of clothing with different degrees of looseness. In summary, by incorporating physical constraints, we generate dynamic clothing models that both adhere to the animation effects of real video data and comply with physical constraints. Using video as input not only reduces costs but also better meets consumer demands. Additionally, it can be applied to body clothing animation in other datasets (e.g., CLOTH3D [1]). Our approach demonstrates reliable performance in unknown poses and clothing types, producing more realistic generated animations and satisfactory results, the advantages of GadNet are shown in Figure 1.

  • Orhlr-net: one-stage residual learning network for joint single-image specular highlight detection and removal

    by Wenzhe Shi, Ziqi Hu, Hao Chen, Hengjia Zhang, Jiale Yang, Li Li