Congyuan Xu

Dr. Congyuan Xu


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
16 January 2025 in PLOS ONE
Verified
Journal Article
Open access
PDF file
CXCongyuan Xu
Congyuan Xu
YZYong Zhan
Yong Zhan
GCGuanghui Chen
Guanghui Chen
ZWZhiqiang Wang
Zhiqiang Wang
SLSiqing Liu
Siqing Liu
WHWeichen Hu
Weichen Hu
The network intrusion detection system (NIDS) plays a critical role in maintaining network security. However, traditional NIDS relies on a large volume of samples for training, which exhibits insufficient adaptability in rapidly changing network environments and complex attack methods, especially when facing novel and rare attacks. As attack strategies evolve, there is often a lack of sufficient samples to train models, making it difficult for traditional methods to respond quickly and effectively to new threats. Although existing few-shot network intrusion detection systems have begun to address sample scarcity, these systems often fail to effectively capture long-range dependencies within the network environment due to limited observational scope. To overcome these challenges, this paper proposes a novel elevated few-shot network intrusion detection method based on self-attention mechanisms and iterative refinement. This approach leverages the advantages of self-attention to effectively extract key features from network traffic and capture long-range dependencies. Additionally, the introduction of positional encoding ensures the temporal sequence of traffic is preserved during processing, enhancing the model’s ability to capture temporal dynamics. By combining multiple update strategies in meta-learning, the model is initially trained on a general foundation during the training phase, followed by fine-tuning with few-shot data during the testing phase, significantly reducing sample dependency while improving the model’s adaptability and prediction accuracy. Experimental results indicate that this method achieved detection rates of 99.90% and 98.23% on the CICIDS2017 and CICIDS2018 datasets, respectively, using only 10 samples.

1 July 2025 in Scientific Reports
Verified
Journal Article
Open access
PDF file
CXCongyuan Xu
Congyuan Xu
YZYong Zhan
Yong Zhan
ZWZhiqiang Wang
Zhiqiang Wang
JYJun Yang
Jun Yang
As network environments become increasingly complex and new attack methods emerge more frequently, the diversity of network attacks continues to grow. Particularly with new or rare attacks, gathering a large number of labeled samples is extremely difficult, resulting in limited training data. Existing few-shot learning methods, while reducing reliance on large datasets, mostly handle single-modality data and fail to fully exploit complementary information across different modalities, limiting detection performance. To address this challenge, we introduce a multimodal fusion based few-shot network intrusion detection method that merges traffic feature graphs and network feature sets. Tailored to these modal characteristics, we develop two models: the G-Model and the S-Model. The G-Model employs convolutional neural networks to capture spatial connections in traffic feature graphs, while the S-Model uses the Transformer architecture to process and fuse network feature sets with long-range dependencies. Furthermore, we extensively study the fusion effects of these two modalities at various interaction depths to enhance detection performance. Experimental validation on the CICIDS2017 and CICIDS2018 datasets demonstrates that our method achieves multi-class accuracy rates of 93.40% and 98.50%, respectively, surpassing existing few-shot network intrusion detection methods.

2 July 2025 in PLOS ONE
Verified
Journal Article
Open access
PDF file
CXCongyuan Xu
Congyuan Xu
DLDonghui Li
Donghui Li
ZLZihao Liu
Zihao Liu
JYJun Yang
Jun Yang
QSQinfeng Shen
Qinfeng Shen
NTNingbing Tong
Ningbing Tong
Deep learning methods have achieved remarkable progress in network intrusion detection. However, their performance often deteriorates significantly in real-world scenarios characterized by limited attack samples and substantial domain shifts. To address this challenge, we propose a novel few-shot intrusion detection method that integrates multi-domain feature fusion with a bidirectional cross-attention mechanism. Specifically, the method adopts a dual-branch feature extractor to jointly capture spatial and frequency domain characteristics of network traffic. The frequency domain features are obtained via two-dimensional discrete cosine transform (2D-DCT), which helps to highlight the spectral structure and improve feature discriminability. To bridge the semantic gap between support and query samples under few-shot conditions, we design a dual-domain bidirectional cross-attention module that enables deep, task-specific alignment across spatial and frequency domains. Additionally, we introduce a hierarchical feature encoding module based on a modified Mamba architecture, which leverages state space modeling to capture long-range dependencies and temporal patterns in traffic sequences. Extensive experiments on two benchmark datasets, CICIDS2017 and CICIDS2018, demonstrate that the proposed method achieves accuracy of 99.03% and 98.64% under the 10-shot setting, outperforming state-of-the-art methods. Moreover, the method exhibits strong cross-domain generalization, achieving over 95.13% accuracy in cross-domain scenarios, thereby proving its robustness and practical applicability in real-world, dynamic network environments.

29 August 2025 in PLOS ONE
Verified
Journal Article
Open access
PDF file
CXCongyuan Xu
Congyuan Xu
JYJun Yang
Jun Yang
PLPanpan Li
Panpan Li
The widespread deployment of Internet of Things (IoT) devices has made them prime targets for cyberattacks. Existing intrusion detection systems (IDSs) heavily rely on large-scale labeled datasets, which limits their effectiveness in detecting novel attacks under few-shot scenarios. To address this challenge, we propose a meta-learning-based intrusion detection method called MACML (Marrying Attention and Convolution-based Meta-Learning). It integrates a self-attention mechanism to capture global dependencies and a convolutional neural network to extract local features, thereby enhancing the model’s overall perception of traffic characteristics. MACML adopts an optimization-based meta-learning framework that enables rapid adaptation to new tasks using only a small number of training samples, improving detection performance and generalization capability. We evaluate MACML on the CICIDS2018 and CICIoT2023 datasets. Experimental results show that, with only 10 training samples, MACML achieves an average accuracy of 98.75% and a detection rate of 99.17% on the CICIDS2018 dataset. On the CICIoT2023 dataset, it reaches 94.47% accuracy and a 95.32% detection rate, outperforming existing state-of-the-art methods.

2 March 2025 in The European Journal on Artificial Intelligence
Verified
Journal Article
Closed access
CXCongyuan Xu
Congyuan Xu
DLDonghui Li
Donghui Li
ZLZihao Liu
Zihao Liu
JYJun Yang
Jun Yang
With the advancement of technology, the demand for healthy eating has increased, making food classification a research hotspot. Existing deep learning-based food image classification models demonstrate high accuracy but require substantial computational resources, limiting their use on resource-constrained devices. In this study, a lightweight convolutional neural network model named MSNet for food classification is proposed. MSNet mainly consists of M Blocks and S Blocks. The M Block uses improved depthwise convolution to reduce the computational cost of conventional convolutions, and the S Block uses channel shuffle techniques to enhance feature information flow between channels without increasing additional computation, effectively capturing relationships between different channel features. Experimental results on three benchmark datasets (ETHZ Food-101, Vireo Food-172, and ISIA Food-500) show that MSNet achieves top-1 accuracies of 86.24%, 87.98%, and 65.70%, with model sizes of 13.8 MB, 15.9 MB, and 25.4 MB, respectively, outperforming mainstream models in terms of computational efficiency. Further quantization produces two MSNet-Lite variants with competitive model size, while maintaining high accuracy and significantly improving inference speed. Additionally, visualization analysis indicates that MSNet effectively extracts essential features of food images, offering good interpretability and generalization across datasets of varying complexity. The proposed MSNet model provides a feasible solution for practical deployment in food classification tasks on mobile and embedded devices.

20 September 2024 in Electronics
Verified
Journal Article
Open access
PDF file
RGRongbin Guo
Rongbin Guo
JYJiantao Yuan
Jiantao Yuan
GWGuan Wang
Guan Wang
CXCongyuan Xu
Congyuan Xu
RYRui Yin
Rui Yin
The next generation network requires not only an ultra-high data rate, global coverage, and connectivity, but also a reduction in network deployment costs and energy consumption. The emergence of reconfigurable intelligent surface (RIS) technology provides an effective way to improve efficiency and reduce cost, while the passive elements bring new challenges of channel estimation (CE) and beam tracking. For an RIS-aided multiple-input and single-output (MISO) system, in this paper, to obtain the channel state information (CSI), we propose a principle component analysis (PCA)-based staged channel estimation method. Based on the estimated channel, we propose a deep learning (DL)-based beam tracking scheme to realize low-complexity RIS reflection coefficient design, which effectively improves the signal-to-noise ratio (SNR) on the user side. The simulation results verified our proposed channel estimation scheme based on PCA, and the beam tracking scheme based on a deep neural network (DNN) for semi-active RIS-aided MISO systems can obtain approximate performances to traditional hand-crafted convex optimization-based algorithms like semi-definite relaxation (SDR) with much lower computational complexity.