Yulin Cao

Prof. Dr. Yulin Cao


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
13 January 2025 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
YCYulin Cao
Yulin Cao
JMJianping Ma
Jianping Ma
ZHZaobo He
Zaobo He
YLYingshu Li
Yingshu Li
In the advancement of collaborative intelligence within the Internet of Things (IoT), Federated Learning (FL) enables clients to collaboratively train a global model without centralizing raw data. However, the non-independent and identically distributed (non-IID) nature of data among clients often leads to divergent local training objectives, deteriorating the performance of the aggregated global model. To address this challenge, we propose AP-CFL, a novel clustered federated learning algorithm that incorporates affinity propagation to dynamically discover the clustering structure of clients without the need to predefine the number of clusters. Specifically, AP-CFL calculates the mean of absolute differences of pairwise cosine similarity to effectively cluster clients based on similarities in their data distributions. Knowledge sharing is enhanced by decoupling each cluster model into a globally shared encoder and a cluster-specific classifier, and the local training objectives are modified to improve the generalization capacity of the shared encoder. Additionally, a robust strategy is introduced to manage partial client participation by employing a time and data importance index, which mitigates the adverse effects of model staleness and maintains the integrity of the clustering structure. Extensive experiments on diverse real-world datasets demonstrate that AP-CFL outperforms existing FL baselines in non-IID settings, effectively improving model quality and convergence stability.

31 March 2025 in IEEE/CAA Journal of Automatica Sinica
Verified
Journal Article
Closed access
JLJiexing Li
Jiexing Li
YCYulin Cao
Yulin Cao
ZXZhengtai Xie
Zhengtai Xie
Long JinLong Jin
Long Jin
Competition-based $k-\mathbf{winners}-\mathbf{take}-\mathbf{all}\ \ (k \mathbf{WTA})$ networks play a crucial role in multi-agent systems. However, existing $k \mathbf{WTA}$ networks either neglect the impact of noise or only consider simple forms, such as constant noise. In practice, noises often exhibit time-varying and nonlinear characteristics, which can be modeled using nonlinear functions and approximated by high-order polynomials. Such noises pose significant challenges for current $k \mathbf{WTA}$ networks, limiting their practical applications. To address this, a $k \mathbf{WTA}$. network with noise characteristics captured $(k \mathbf{WTA}-\mathbf{NCC})$ is proposed in this article. Theoretical analyses demonstrate that the residual error of the proposed $k\mathbf{WTA}- \mathbf{NCC}$ network converges to zero globally, while simulation results confirm its robustness against polynomial noises. Additionally, a $k \mathbf{WTA}$ coordination model is constructed by integrating the proposed network with a consensus estimator to achieve multi-agent tracking tasks. Finally, simulations and physical experiments are conducted further to demonstrate the validity and practicality of the $k$ WTA coordination model.

19 April 2024 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
ZHZaobo He
Zaobo He
FZFuping Zhang
Fuping Zhang
YLYusen Li
Yusen Li
YCYulin Cao
Yulin Cao
ZCZhipeng Cai
Zhipeng Cai
Federated learning (FL) enables multiple Internet of Things (IoT) devices to collaboratively train a global model without centralizing raw data. However, achieving optimal performance for each device is challenging due to the heterogeneity of IoT data. Moreover, IoT devices, constrained in computational power and energy, necessitate innovative privacy protection solutions in FL. In this article, we propose privacy-enhanced FL methods to address data heterogeneity in IoT. Our approach includes a dual-model framework for each device, combining a shared model to facilitate interclient communication with a personalized model. This personalized model is developed from local aggregation, utilizing layer-wise aggregation weights generated by local hypernetworks. This integration effectively balances collective intelligence with client-specific knowledge, addressing data heterogeneity for improved personalization. Additionally, we propose a local gradient shielding (LGS) method to protect training data. This method conceals gradients in locally trained personalized models, preventing data reconstruction and ensuring client privacy with minimal resource consumption. Extensive evaluations on benchmark data sets, showcasing various types and degrees of data heterogeneity, demonstrate the superiority of our approaches over current baselines.

25 March 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
YCYulin Cao
Yulin Cao
SDShihao Dong
Shihao Dong
JZJiawei Zhang
Jiawei Zhang
HXHan Xu
Han Xu
YZYan Zhang
Yan Zhang
Yuhui ZhengYuhui Zheng
Yuhui Zheng
As a tool for near-earth remote sensing, unmanned aerial vehicle (UAV) can be used to acquire images and data of the earth's surface. This provides a powerful support for Earth observation and resource management. Object tracking in UAV videos has been a topic of much interest in recent years. A large number of algorithms have been proposed. Among these algorithms, deep learning has achieved a high accuracy rate. However, it is difficult to carry hardware devices for UAV, which makes it difficult to be practically applied. The correlation filter does not require a graphic processing unit (GPU) to accelerate the computation, but it uses only manual features, which makes it difficult to achieve satisfactory performance. In order to solve the above problems, we proposed adaptive spatial regularization correlation filters, called DTSRT. Specifically, we first introduce deep features in the correlation filter instead of the original manual features, which can greatly improve the discriminative ability of the model. At the same time, in order to prevent affecting the real-time performance of the algorithm, we use histogram of oriented gradients (HOG) features to determine the target scale and deep features to determine the target location. In addition, considering the large inter-frame distance between targets in UAV videos, we use a saliency detection method to dynamically generate spatially constrained templates. The proposed DTSRT outperforms other state-of-the-art algorithms with area under curve (AUC) of 0.481, 0.431, and 0.474 on UAV123@10FPS, UAVDT, and DTB70 datasets, respectively.

12 January 2026 in IEEE Transactions on Geoscience and Remote Sensing
Verified
Journal Article
Closed access
Le SunLe Sun
Le Sun
BSBingcheng Shi
Bingcheng Shi
Yuhui ZhengYuhui Zheng
Yuhui Zheng
LXLiang Xiao
Liang Xiao
YCYulin Cao
Yulin Cao
Deep learning methods have achieved remarkable performance in hyperspectral image (HSI) change detection. However, they still face two primary limitations. First, most approaches treat all spectral channels equally, overlooking the varying importance of spectral information. Second, they often struggle to clearly delineate edges in low-resolution images, resulting in suboptimal feature extraction and increased false or missed detections. To address these challenges, we propose a novel Spectral Frequency-Induced Edge Enhancement Transformer (SFIEET) network to enhance the accuracy and robustness of HSI change detection. The proposed SFIEET framework comprises three key modules. The Spectral Frequency and Spatial Domain Fusion (SFSDF) module implements an adaptive frequency-domain attention mechanism based on grouped DCT, dynamically focusing on channels rich in weak semantic information. Simultaneously, it integrates multi-scale contextual spatial information to capture morphological changes such as texture fragmentation, effectively enhancing recognition of structural changes. The Sharpening and Laplace-Guided Edge Enhancement (SLGEE) module integrates multi-scale pooling with the Laplacian operator to fuse edge features across spectral bands, enabling precise boundary localization and more reliable detection in blurred regions. Finally, the Depthwise Convolution-Embedded Lightweight Transformer (DW-ELT) module incorporates depthwise separable convolutions into the Transformer architecture to enhance local feature modeling, while gated linear units (GLUs) strengthen nonlinear representations, making change-region features more discriminative. Extensive experiments on three benchmark datasets demonstrate that the proposed SFIEET consistently outperforms existing state-of-the-art methods. The source code is available at https://github.com/bcshi83/SFIEET.git for the sake of reproducibility.

19 December 2025 in IEEE Transactions on Industrial Electronics
Verified
Journal Article
Closed access
JLJiexing Li
Jiexing Li
YGYongji Guan
Yongji Guan
YCYulin Cao
Yulin Cao
Long JinLong Jin
Long Jin
In multi-robot systems, the $\boldsymbol{k}$-winners-take-all ($\boldsymbol{k}$-WTA) problem serves as a competitive selection mechanism to identify the $\boldsymbol{k}$ most suitable robots from a group of $\boldsymbol{n}$ ($\boldsymbol{n} > \boldsymbol{k}$) ones for a given task. Existing $\boldsymbol{k}$-WTA networks either fail to eliminate lagging errors or involve a complex modeling process due to the introduction of additional Lagrangian variables to deal with bounded constraints. To address these challenges, a primal-dual-based differential $\boldsymbol{k}$-WTA (PDD-$\boldsymbol{k}$WTA) network is proposed in this article. This network incorporates time-derivative information to mitigate lagging errors and leverages linear variational inequalities (LVIs) to streamline the network structure. Theoretical analyses and comparative simulations validate the effectiveness and superiority of the PDD-$\boldsymbol{k}$WTA network. Furthermore, a multi-robot competitive coordination (MCC) model, which integrates the PDD-$\boldsymbol{k}$WTA network, a consensus estimator, and a robot control law, is developed to achieve the tracking of a mobile target.