Yuanyuan Zhang
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
Research Subjects
Yuanyuan Zhang has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
- Jinbo Xiong
Fujian University of Technology
10 shared publications - Renwan Bi
School of Cyber Science and Engineering, Minjiang University, Fuzhou, China
7 shared publications - Youliang Tian
Guizhou University
3 shared publications - Chun Lung Philip Chen
South China University of Technology
2 shared publications - Li Lin
Fujian Normal University
2 shared publications - Qi Li
Nanjing University of Posts and Telecommunications
2 shared publications
Latest Publications
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 3543-3558
Privacy-Preserving Multi-Modal Object Fusion for Connected Autonomous Vehicles: Resilience Against Malicious Third-Party Attacks
Connected autonomous vehicles (CAVs) utilize multi-modal sensors, such as LiDAR and high-definition cameras, to collect diverse types of sensing data. Fusing object detection information from these two modalities facilitates more accurate environmental perception. In this context, lightweight secret sharing techniques are employed to protect information privacy, enabling further calculation while effectively alleviating the computational resource constraints of CAVs. Meanwhile, such techniques require an additional third-party to generate some necessary random numbers. Addressing the challenges of privacy disclosure of multi-modal object information and the reliability of random numbers, we propose a malicious third-party-resistant privacy-preserving multi-modal object fusion model, termed MPOF. First, we develop a series of secure computation protocols that do not rely on time-consuming cryptographic primitives, including secure multiplication, secure sharing conversion, and secure comparison. Leveraging the idea of sacrificial verification, we can effectively detect malicious behavior by the third-party during the random number generation process. Second, we construct a secure object bounding-box matching module based on arithmetic secret sharing (ASS), enabling similarity calculation and matching of bounding-boxes between point cloud and image modalities. Additionally, we design a secure object score fusion module that achieves fusion and updating through secure implementations of convolution, ReLU, and Maxout operations. Detailed theoretical analysis and experimental results demonstrate that, compared to secure computation protocols using homomorphic encryption for random number generation, the proposed protocols reduce computational overhead by five orders of magnitude. Furthermore, the MPOF model constructed by integrating these protocols is secure, accurate, and efficient.
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 3543-3558
Privacy-Preserving Multi-Modal Object Fusion for Connected Autonomous Vehicles: Resilience Against Malicious Third-Party Attacks
Connected autonomous vehicles (CAVs) utilize multi-modal sensors, such as LiDAR and high-definition cameras, to collect diverse types of sensing data. Fusing object detection information from these two modalities facilitates more accurate environmental perception. In this context, lightweight secret sharing techniques are employed to protect information privacy, enabling further calculation while effectively alleviating the computational resource constraints of CAVs. Meanwhile, such techniques require an additional third-party to generate some necessary random numbers. Addressing the challenges of privacy disclosure of multi-modal object information and the reliability of random numbers, we propose a malicious third-party-resistant privacy-preserving multi-modal object fusion model, termed MPOF. First, we develop a series of secure computation protocols that do not rely on time-consuming cryptographic primitives, including secure multiplication, secure sharing conversion, and secure comparison. Leveraging the idea of sacrificial verification, we can effectively detect malicious behavior by the third-party during the random number generation process. Second, we construct a secure object bounding-box matching module based on arithmetic secret sharing (ASS), enabling similarity calculation and matching of bounding-boxes between point cloud and image modalities. Additionally, we design a secure object score fusion module that achieves fusion and updating through secure implementations of convolution, ReLU, and Maxout operations. Detailed theoretical analysis and experimental results demonstrate that, compared to secure computation protocols using homomorphic encryption for random number generation, the proposed protocols reduce computational overhead by five orders of magnitude. Furthermore, the MPOF model constructed by integrating these protocols is secure, accurate, and efficient.
- 17 February 2025
- Published by Springer Nature
- p. 280-289
FPIM: Fair and Privacy-Preserving Incentive Mechanism in Mobile Crowdsensing
- 17 February 2025
- Published by Springer Nature
- p. 280-289
FPIM: Fair and Privacy-Preserving Incentive Mechanism in Mobile Crowdsensing
- 18 January 2025
- Published by Elsevier in Vehicular Communications
- Vol. 52
Privacy-enhanced multi-region data aggregation for Internet of Vehicles
- 18 January 2025
- Published by Elsevier in Vehicular Communications
- Vol. 52
Privacy-enhanced multi-region data aggregation for Internet of Vehicles
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (4) , 3785-3799
Achieving Efficient Privacy-Preserving Mixed Data Quality Assessment in Mobile Crowdsensing
In mobile crowdsensing (MCS) applications, the single type data is inadequate to reflect the complexities of the real world and meet precise task requirements. Currently, there are few works that focus on mixed data in the context of MCS, and there is no work considering the credit issues of sensing platform. The privacy, fairness, and reliability of assessing the quality of mixed data remain unguaranteed. Therefore, we design a high-efficiency and privacy-preserving mixed data quality assessment scheme which adopts a dual-server architecture, designs secure k-prototype clustering for quality assessment, and conducts anomaly detection to eliminate anomalous data. Furthermore, we design a fair and reliable allocation mechanism to fairly allocate reward to users based on fixed and floating reward mechanisms for incentivizing rational users to submit high-quality mixed data. To prevent payment defaults by the sensing platform, we design verifiable credential to restrict them, ensuring payment fairness and transactional reliability. Finally, through theoretical analysis and experimental evaluation, we demonstrate the effectiveness and security of the proposed scheme. The results indicate that in terms of efficiency, the time overhead of mixed data quality assessment has been significantly reduced by three orders of magnitude compared to existing schemes.
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (4) , 3785-3799
Achieving Efficient Privacy-Preserving Mixed Data Quality Assessment in Mobile Crowdsensing
In mobile crowdsensing (MCS) applications, the single type data is inadequate to reflect the complexities of the real world and meet precise task requirements. Currently, there are few works that focus on mixed data in the context of MCS, and there is no work considering the credit issues of sensing platform. The privacy, fairness, and reliability of assessing the quality of mixed data remain unguaranteed. Therefore, we design a high-efficiency and privacy-preserving mixed data quality assessment scheme which adopts a dual-server architecture, designs secure k-prototype clustering for quality assessment, and conducts anomaly detection to eliminate anomalous data. Furthermore, we design a fair and reliable allocation mechanism to fairly allocate reward to users based on fixed and floating reward mechanisms for incentivizing rational users to submit high-quality mixed data. To prevent payment defaults by the sensing platform, we design verifiable credential to restrict them, ensuring payment fairness and transactional reliability. Finally, through theoretical analysis and experimental evaluation, we demonstrate the effectiveness and security of the proposed scheme. The results indicate that in terms of efficiency, the time overhead of mixed data quality assessment has been significantly reduced by three orders of magnitude compared to existing schemes.
- 26 August 2024
- Published by Elsevier in Journal of Information Security and Applications
- Vol. 85
Achieving lightweight, efficient, privacy-preserving user recruitment in mobile crowdsensing
- 26 August 2024
- Published by Elsevier in Journal of Information Security and Applications
- Vol. 85
Achieving lightweight, efficient, privacy-preserving user recruitment in mobile crowdsensing
- 22 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (3) , 41-47
Privacy-Preserving Outsourcing Learning for Connected Autonomous Vehicles: Challenges, Solutions, and Perspectives
Although data sharing and fusion between connected autonomous vehicles (CAVs) can effectively enhance environment awareness and improve driving safety, it has to face severe challenges of privacy disclosure. Outsourcing encrypted data to edge servers for data analysis and model learning can alleviate this issue without imposing additional computing load on CAVs. In this article, we propose a privacy-preserving outsourcing learning (PPOL) framework based on lightweight additive secret sharing (ASS). Firstly, we propose a privacy-preserving outsourcing object detection method over secretly shared image and point cloud data. Secondly, we construct a privacy-preserving outsourcing depth estimation model over fused stereo image shares, aiming to provide a feasible solution for outsourcing data fusion and learning. Finally, we point out open issues under the PPOL framework, and give research perspectives in aspects of privacy-preserving data fusion for multi-frame, multi-modal and multi-view data, as well as privacy-preserving model optimization.
- 22 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 38 (3) , 41-47
Privacy-Preserving Outsourcing Learning for Connected Autonomous Vehicles: Challenges, Solutions, and Perspectives
Although data sharing and fusion between connected autonomous vehicles (CAVs) can effectively enhance environment awareness and improve driving safety, it has to face severe challenges of privacy disclosure. Outsourcing encrypted data to edge servers for data analysis and model learning can alleviate this issue without imposing additional computing load on CAVs. In this article, we propose a privacy-preserving outsourcing learning (PPOL) framework based on lightweight additive secret sharing (ASS). Firstly, we propose a privacy-preserving outsourcing object detection method over secretly shared image and point cloud data. Secondly, we construct a privacy-preserving outsourcing depth estimation model over fused stereo image shares, aiming to provide a feasible solution for outsourcing data fusion and learning. Finally, we point out open issues under the PPOL framework, and give research perspectives in aspects of privacy-preserving data fusion for multi-frame, multi-modal and multi-view data, as well as privacy-preserving model optimization.
- Open Access
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 342-349
PracticalHAR: A Practical Method of Human Activity Recognition Based on Embedded Sensors of Mobile Phone
Human Activity Recognition (HAR) is the process of identifying human activity states based on collected data, with applications in health monitoring, rehabilitation therapy, and smart homes. However, accurate recognition often necessitates the use of multiple sensors, which can cause discomfort and significantly reduce practicality and convenience. Currently, smartphones with embedded sensors have broadened the potential for HAR, enabling healthcare monitoring, life logging, fitness tracking, and more. This paper conducts experiments on the WISDM public dataset and a self-collected dataset, demonstrating that a one-dimensional convolutional neural network structure does not necessarily underperform a two-dimensional one, using only accelerometer data. Moreover, the experimental results on the self-collected dataset suggest that utilizing the barometer built into smartphones can enhance activity recognition rates. Specifically, we first built the corresponding convolutional neural network model and conducted experiments on the WISDM dataset, and the accuracy reached 95.42%. Next, we collected data from the smartphone’s embedded barometer and accelerometer to construct a dataset. Experimental results on our self-collected dataset show that the inclusion of barometer data can improve the accuracy of activity recognition, especially stair-related activities. The F1-score for identifying walking, ascending stairs, and descending stairs increased by 6.7%, 7.39%, and 2.99% respectively.
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 342-349
Open AccessPracticalHAR: A Practical Method of Human Activity Recognition Based on Embedded Sensors of Mobile Phone
Human Activity Recognition (HAR) is the process of identifying human activity states based on collected data, with applications in health monitoring, rehabilitation therapy, and smart homes. However, accurate recognition often necessitates the use of multiple sensors, which can cause discomfort and significantly reduce practicality and convenience. Currently, smartphones with embedded sensors have broadened the potential for HAR, enabling healthcare monitoring, life logging, fitness tracking, and more. This paper conducts experiments on the WISDM public dataset and a self-collected dataset, demonstrating that a one-dimensional convolutional neural network structure does not necessarily underperform a two-dimensional one, using only accelerometer data. Moreover, the experimental results on the self-collected dataset suggest that utilizing the barometer built into smartphones can enhance activity recognition rates. Specifically, we first built the corresponding convolutional neural network model and conducted experiments on the WISDM dataset, and the accuracy reached 95.42%. Next, we collected data from the smartphone’s embedded barometer and accelerometer to construct a dataset. Experimental results on our self-collected dataset show that the inclusion of barometer data can improve the accuracy of activity recognition, especially stair-related activities. The F1-score for identifying walking, ascending stairs, and descending stairs increased by 6.7%, 7.39%, and 2.99% respectively.
- Open Access
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-5
AntiMal: an Approach of Malware Detection Employing Swin Transformer
In the present era, the menace of malicious software is growing continuously, posing a significant challenge in the realm of cybersecurity. Currently, traditional machine learning algorithms have been widely applied to the analysis of malicious software and are considered effective methods. However, these approaches often require extensive feature engineering, feature learning, and feature representation, which can be time-consuming and resource-intensive. In this paper, a method for the identification of malicious software, named "AntiMal," is proposed. This method utilizes the Swin Transformer as the backbone network and combines it with image features for static analysis. It leverages the dataset provided by Microsoft for the Kaggle Microsoft Malware Classification Challenge. Initially, it converts the binary data of malicious code into grayscale images, treating every 8 bits as a pixel. Subsequently, it employs PyTorch to build and train the Swin Transformer, classifying the images in the validation set. The results demonstrate that this approach excels in the problem of malicious software classification, achieving an impressive accuracy of up to 95%, showcasing remarkable precision and generalization capabilities. Furthermore, it exhibits significant advantages when dealing with large datasets, requiring fewer computational resources, thus offering exceptional computational efficiency.
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessAntiMal: an Approach of Malware Detection Employing Swin Transformer
In the present era, the menace of malicious software is growing continuously, posing a significant challenge in the realm of cybersecurity. Currently, traditional machine learning algorithms have been widely applied to the analysis of malicious software and are considered effective methods. However, these approaches often require extensive feature engineering, feature learning, and feature representation, which can be time-consuming and resource-intensive. In this paper, a method for the identification of malicious software, named "AntiMal," is proposed. This method utilizes the Swin Transformer as the backbone network and combines it with image features for static analysis. It leverages the dataset provided by Microsoft for the Kaggle Microsoft Malware Classification Challenge. Initially, it converts the binary data of malicious code into grayscale images, treating every 8 bits as a pixel. Subsequently, it employs PyTorch to build and train the Swin Transformer, classifying the images in the validation set. The results demonstrate that this approach excels in the problem of malicious software classification, achieving an impressive accuracy of up to 95%, showcasing remarkable precision and generalization capabilities. Furthermore, it exhibits significant advantages when dealing with large datasets, requiring fewer computational resources, thus offering exceptional computational efficiency.
- Open Access
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 82-87
Marine litter detection based on YOLOv7 algorithm and data encryption protection
Marine litter can cause significant damage to marine biodiversity, threatening the marine food chain and spreading harmful substances, posing a significant impact on the ocean ecosystem. Autonomous Underwater Vehicles (AUVs) can automatically remove marine litter using sensors and trained models. this paper evaluates the YOLOv7 series of models that utilize deep learning to detect targets in a real underwater environment. In order to improve the performance of the model, we introduced two attention mechanisms, and the experimental results showed a 2.5% increase in Mean Average Precision(mAP) values. We used a large publicly available dataset of re-annotated open-water debris to train convolutional neural networks for target detection, and we evaluated the trained models on a subset of the dataset, to provide insights into the ability of deep learning to detect marine litter. In addition, to prevent attacks by malicious actors during AUVs cloud platform access, we introduced data encryption protection to ensure that the model’s predicted results can be correctly received by AUVs.
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 82-87
Open AccessMarine litter detection based on YOLOv7 algorithm and data encryption protection
Marine litter can cause significant damage to marine biodiversity, threatening the marine food chain and spreading harmful substances, posing a significant impact on the ocean ecosystem. Autonomous Underwater Vehicles (AUVs) can automatically remove marine litter using sensors and trained models. this paper evaluates the YOLOv7 series of models that utilize deep learning to detect targets in a real underwater environment. In order to improve the performance of the model, we introduced two attention mechanisms, and the experimental results showed a 2.5% increase in Mean Average Precision(mAP) values. We used a large publicly available dataset of re-annotated open-water debris to train convolutional neural networks for target detection, and we evaluated the trained models on a subset of the dataset, to provide insights into the ability of deep learning to detect marine litter. In addition, to prevent attacks by malicious actors during AUVs cloud platform access, we introduced data encryption protection to ensure that the model’s predicted results can be correctly received by AUVs.