Junjian Li
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
Junjian Li has a total of 43 co-authors. Here are the top co-authors by number of shared publications:
- Honglong Chen
China University of Petroleum (East China)
12 shared publications - Yudong Gao
China University of Petroleum (East China), Qingdao, China
8 shared publications - Weifeng Liu
College of Control Science and Engineering, China University of Petroleum (East China), Qingdao, China
6 shared publications - Peng Sun
Hunan University
5 shared publications - Feng Xia
School of Computing Technologies, RMIT University
3 shared publications - Zhe Li
College of Control Science and Engineering, China University of Petroleum (East China), Qingdao, China
3 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6708-6717
Dual Inversion for Text-to-Image Diffusion Models: From Both Prompt and Noise Perspectives
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 6708-6717
Open AccessDual Inversion for Text-to-Image Diffusion Models: From Both Prompt and Noise Perspectives
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 22  (6) , 7746-7758
A Triple Stealthy Backdoor: Hidden in Spatial, Frequency, and Feature Domains
Backdoor attacks pose significant security risks to deep neural networks (DNNs). These attacks involve models that make intentionally incorrect (and potentially targeted) predictions on poisoned inputs containing carefully crafted triggers, while operating normally with clean inputs. Prior studies have investigated the invisibility of backdoor triggers to improve attack stealthiness. However, they primarily concentrate on achieving invisibility solely in the spatial domain, ignoring the generation of invisible triggers in the frequency and feature domains. This constraint makes the poisoned images vulnerable to detection by recent defense mechanisms. To tackle this problem, we introduce a Triple stealthy BAckdoor attack approach, termed TriBA, which simultaneously ensures the invisibility of triggers in all the spatial, frequency, and feature domains, to achieve desirable attack performance, while ensuring strong stealthiness. Specifically, we initially utilize Wavelet Transform to embed the high-frequency information from the trigger image into the clean image to ensure effective attack performance. Then, to achieve strong stealthiness across both spatial and frequency domains, we integrate Fourier Transform and Cosine Transform to blend the poisoned image and clean image in the frequency domain. Furthermore, TriBA adopts an attack strategy to make the backdoor features similar to clean features in the feature space, which guarantees trigger invisibility in the feature domain while maintaining attack effectiveness. We theoretically prove the effectiveness of this strategy. Finally, TriBA has been comprehensively evaluated on four datasets against popular image classifiers, demonstrating a marked improvement over existing state-of-the-art backdoor attacks in terms of both attack success rate and stealthiness.
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 22  (6) , 7746-7758
A Triple Stealthy Backdoor: Hidden in Spatial, Frequency, and Feature Domains
Backdoor attacks pose significant security risks to deep neural networks (DNNs). These attacks involve models that make intentionally incorrect (and potentially targeted) predictions on poisoned inputs containing carefully crafted triggers, while operating normally with clean inputs. Prior studies have investigated the invisibility of backdoor triggers to improve attack stealthiness. However, they primarily concentrate on achieving invisibility solely in the spatial domain, ignoring the generation of invisible triggers in the frequency and feature domains. This constraint makes the poisoned images vulnerable to detection by recent defense mechanisms. To tackle this problem, we introduce a Triple stealthy BAckdoor attack approach, termed TriBA, which simultaneously ensures the invisibility of triggers in all the spatial, frequency, and feature domains, to achieve desirable attack performance, while ensuring strong stealthiness. Specifically, we initially utilize Wavelet Transform to embed the high-frequency information from the trigger image into the clean image to ensure effective attack performance. Then, to achieve strong stealthiness across both spatial and frequency domains, we integrate Fourier Transform and Cosine Transform to blend the poisoned image and clean image in the frequency domain. Furthermore, TriBA adopts an attack strategy to make the backdoor features similar to clean features in the feature space, which guarantees trigger invisibility in the feature domain while maintaining attack effectiveness. We theoretically prove the effectiveness of this strategy. Finally, TriBA has been comprehensively evaluated on four datasets against popular image classifiers, demonstrating a marked improvement over existing state-of-the-art backdoor attacks in terms of both attack success rate and stealthiness.
- 30 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27  (15209210) , 5909-5921
Black-Box Adversarial Defense Based on Image Decomposition and Reconstruction
Adversarial attacks have challenged the security of deep neural networks (DNNs) recently. The most prominent adversarial attack methods include backdoor attacks, adversarial examples, etc. These attack methods inject triggers or perturbations into images, leading to extremely dangerous security vulnerability in deep learning domain. The various forms of adversarial attacks can contaminate DNNs with their distinct characteristics. The complexity of adversarial attack poses a great challenge to designing a general defense strategy. In this paper, we propose a novel defense method against most of adversarial attacks through Image Decomposition and Reconstruction (IDR). Our method can be applied to poisoned images without the need for internal information about the model or any prior knowledge of the clean/poisoned images. We apply a linear transformation on the poisoned image to destroy the perturbations or triggers and deploy a pre-trained diffusion model to reconstruct the original information. In particular, we propose a novel reverse process that utilizes the consistency of range-null space decomposition to guide the generation of purified images. The decomposition of the range-null space can guarantee the retrieval of image information, which enhances the robustness of our method and contributes to the reliable purification of poisoned images. We assess the effectiveness of our proposed IDR against various prevalent backdoor attacks, adversarial examples and Image-Scaling attack methods. The experimental results highlight the outstanding defensive capabilities of our proposed IDR, demonstrating an exceptionally high defense success rate.
- 30 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27  (15209210) , 5909-5921
Black-Box Adversarial Defense Based on Image Decomposition and Reconstruction
Adversarial attacks have challenged the security of deep neural networks (DNNs) recently. The most prominent adversarial attack methods include backdoor attacks, adversarial examples, etc. These attack methods inject triggers or perturbations into images, leading to extremely dangerous security vulnerability in deep learning domain. The various forms of adversarial attacks can contaminate DNNs with their distinct characteristics. The complexity of adversarial attack poses a great challenge to designing a general defense strategy. In this paper, we propose a novel defense method against most of adversarial attacks through Image Decomposition and Reconstruction (IDR). Our method can be applied to poisoned images without the need for internal information about the model or any prior knowledge of the clean/poisoned images. We apply a linear transformation on the poisoned image to destroy the perturbations or triggers and deploy a pre-trained diffusion model to reconstruct the original information. In particular, we propose a novel reverse process that utilizes the consistency of range-null space decomposition to guide the generation of purified images. The decomposition of the range-null space can guarantee the retrieval of image information, which enhances the robustness of our method and contributes to the reliable purification of poisoned images. We assess the effectiveness of our proposed IDR against various prevalent backdoor attacks, adversarial examples and Image-Scaling attack methods. The experimental results highlight the outstanding defensive capabilities of our proposed IDR, demonstrating an exceptionally high defense success rate.
- 22 October 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 138
: Backdoor attack with bokeh effects via latent separation suppression- 22 October 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 138
: Backdoor attack with bokeh effects via latent separation suppression- 2 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9  (1) , 861-873
Generative Adversarial Network Based Image-Scaling Attack and Defense Modeling
Deep learning has achieved remarkable performance in computer vision applications, and gradually becomes one of the mainstream technologies. Image scaling, as an indispensable data pre-processing procedure for most of the computer vision applications, is implemented to resize the unmatched data to fit the input sizes of deep learning models. However, such kind of data pre-processing is vulnerable to be exploited to launch an attack called image-scaling attack, which can generate attack images presenting entirely different appearances after scaling. In this paper, we develop the attack model based on generative adversarial network (GAN) to fuse the source and target images, called Attack-GAN, which guarantees the generated attack images are imperceptible as well as being scaled to the target images. Experiments show that the superiority of Attack-GAN over the optimization method is that it can generate attack images with better camouflage and achieve higher attack success rates as well as accelerate the generation of attack images. Furthermore, we propose the Defense-GAN to learn and approximate the distribution of unperturbed images from attack images, which aims to eliminate and compensate the adversarial pixels. Based on different capabilities, customized defense strategies for different defenders are developed to resist the image-scaling attack. Experimental results manifest the proposed defense strategies are effective against the image-scaling attack, i.e., the generator of Defense-GAN can recover the attack images and retain the original ‘semantics’.
- 2 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9  (1) , 861-873
Generative Adversarial Network Based Image-Scaling Attack and Defense Modeling
Deep learning has achieved remarkable performance in computer vision applications, and gradually becomes one of the mainstream technologies. Image scaling, as an indispensable data pre-processing procedure for most of the computer vision applications, is implemented to resize the unmatched data to fit the input sizes of deep learning models. However, such kind of data pre-processing is vulnerable to be exploited to launch an attack called image-scaling attack, which can generate attack images presenting entirely different appearances after scaling. In this paper, we develop the attack model based on generative adversarial network (GAN) to fuse the source and target images, called Attack-GAN, which guarantees the generated attack images are imperceptible as well as being scaled to the target images. Experiments show that the superiority of Attack-GAN over the optimization method is that it can generate attack images with better camouflage and achieve higher attack success rates as well as accelerate the generation of attack images. Furthermore, we propose the Defense-GAN to learn and approximate the distribution of unperturbed images from attack images, which aims to eliminate and compensate the adversarial pixels. Based on different capabilities, customized defense strategies for different defenders are developed to resist the image-scaling attack. Experimental results manifest the proposed defense strategies are effective against the image-scaling attack, i.e., the generator of Defense-GAN can recover the attack images and retain the original ‘semantics’.
- Open Access
- 19 June 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Autonomous and Adaptive Systems
IBAQ: Frequency-Domain Backdoor Attack Threatening Autonomous Driving via Quadratic Phase
The rapid evolution of backdoor attacks has emerged as a significant threat to the security of autonomous driving models. An attacker injects a backdoor into the model by adding triggers to the samples, which can be activated to manipulate the model’s inference. Backdoor attacks can lead to severe consequences, such as misidentifying traffic signs during autonomous driving, posing a risk of causing traffic accidents. Recently, there has been a gradual evolution of frequency-domain backdoor attacks. However, since the change of both amplitude and its corresponding phase will significantly affect image appearance, most of the existing frequency-domain backdoor attacks change only the amplitude, which results in a suboptimal efficacy of the attack. In this work, we propose an attack called IBAQ, to solve this problem by blurring semantic information of the trigger image through the quadratic phase. Initially, we convert the trigger and benign sample to YCrCb space. Then, we perform the fast Fourier transform on the Y channel, blending the trigger image’s amplitude and quadratic phase linearly with the benign sample’s amplitude and phase. IBAQ achieves covert injection of trigger information within amplitude and phase, enhancing the attack effect. We validate the effectiveness and stealthiness of IBAQ through comprehensive experiments.
- 19 June 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Autonomous and Adaptive Systems
Open AccessIBAQ: Frequency-Domain Backdoor Attack Threatening Autonomous Driving via Quadratic Phase
The rapid evolution of backdoor attacks has emerged as a significant threat to the security of autonomous driving models. An attacker injects a backdoor into the model by adding triggers to the samples, which can be activated to manipulate the model’s inference. Backdoor attacks can lead to severe consequences, such as misidentifying traffic signs during autonomous driving, posing a risk of causing traffic accidents. Recently, there has been a gradual evolution of frequency-domain backdoor attacks. However, since the change of both amplitude and its corresponding phase will significantly affect image appearance, most of the existing frequency-domain backdoor attacks change only the amplitude, which results in a suboptimal efficacy of the attack. In this work, we propose an attack called IBAQ, to solve this problem by blurring semantic information of the trigger image through the quadratic phase. Initially, we convert the trigger and benign sample to YCrCb space. Then, we perform the fast Fourier transform on the Y channel, blending the trigger image’s amplitude and quadratic phase linearly with the benign sample’s amplitude and phase. IBAQ achieves covert injection of trigger information within amplitude and phase, enhancing the attack effect. We validate the effectiveness and stealthiness of IBAQ through comprehensive experiments.
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 70  (4) , 7044-7055
Image-Scaling Attack on Image Signal Processing Pipelines in Deep Neural Networks-Based Outdoor Vision Applications
Adversarial attacks pose a potential threat to the deployment of deep neural networks (DNNs) on the consumer electronic (CE) devices for outdoor use (e.g., face recognition and fire monitoring). Generally, most of the DNN-based CE vision devices are equipped with image signal processing (ISP) pipelines to implement RAW-to-RGB transformations and embedded into data preprocessing modules for efficient image processing. However, none of the existing adversarial attacks takes into account the impacts of both ISP pipeline and data preprocessing. In this paper, we develop an image-scaling attack targeting on the ISP pipeline and preprocessing process, which are the indispensable components in the DNN-based CE vision applications. Specially, the crafted adversarial RAW can be transformed into the attack image that presents an entirely different appearance once being scaled to a specific-size image. Notably, the attack is driven by the gradient information from the ISP pipeline. Thus, we first consider the gradient-available ISP pipeline, i.e., the gradient information can be directly utilized in the generation process of adversarial RAW. Subsequently, the scope of the attack is then expanded to encompass the gradient-unavailable ISP pipelines, where the gradient information cannot be directly computed. To obtain the effective gradient information for attacking, we design a proxy model as the gradient oracles that can well learn the RAW-to-RGB transformations of the target gradient-unavailable ISP pipeline. Then the attack can be launched by utilizing the approximated gradient information from the proxy model. Extensive experiments show that the proposed adversarial attacks can craft adversarial RAW data against the target ISP pipelines in vision applications with high attack success rates. We investigate the potential vulnerabilities in the vision devices and emphasize the urgent requirement for robust security measures to ensure the dependable deployment of these devices in outdoor CE products.
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 70  (4) , 7044-7055
Image-Scaling Attack on Image Signal Processing Pipelines in Deep Neural Networks-Based Outdoor Vision Applications
Adversarial attacks pose a potential threat to the deployment of deep neural networks (DNNs) on the consumer electronic (CE) devices for outdoor use (e.g., face recognition and fire monitoring). Generally, most of the DNN-based CE vision devices are equipped with image signal processing (ISP) pipelines to implement RAW-to-RGB transformations and embedded into data preprocessing modules for efficient image processing. However, none of the existing adversarial attacks takes into account the impacts of both ISP pipeline and data preprocessing. In this paper, we develop an image-scaling attack targeting on the ISP pipeline and preprocessing process, which are the indispensable components in the DNN-based CE vision applications. Specially, the crafted adversarial RAW can be transformed into the attack image that presents an entirely different appearance once being scaled to a specific-size image. Notably, the attack is driven by the gradient information from the ISP pipeline. Thus, we first consider the gradient-available ISP pipeline, i.e., the gradient information can be directly utilized in the generation process of adversarial RAW. Subsequently, the scope of the attack is then expanded to encompass the gradient-unavailable ISP pipelines, where the gradient information cannot be directly computed. To obtain the effective gradient information for attacking, we design a proxy model as the gradient oracles that can well learn the RAW-to-RGB transformations of the target gradient-unavailable ISP pipeline. Then the attack can be launched by utilizing the approximated gradient information from the proxy model. Extensive experiments show that the proposed adversarial attacks can craft adversarial RAW data against the target ISP pipelines in vision applications with high attack success rates. We investigate the potential vulnerabilities in the vision devices and emphasize the urgent requirement for robust security measures to ensure the dependable deployment of these devices in outdoor CE products.
- 13 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71  (2) , 6381-6391
Frequency is Devious: Deep Stealthy Backdoor in Artificial Intelligence-Based Consumer Internet of Things
Artificial Intelligence (AI) based Consumer Internet of Things (CIoT) flourishes at a rapid speed due to its excellent data collection ability, which plays an important role in optimizing Deep Neural Networks (DNNs) of AI-based CIoT. However, since the CIoT devices can be accessed without the user’s permission, the DNNs of AI-based CIoT face various security threats, especially the backdoor attack caused by data manipulation. Existing backdoor attacks only concentrate on designing imperceptible triggers in the spatial domain, but not the frequency domain, making the triggers easily detectable or even removable by recent defense methods or humans. In this paper, we reveal a DEep STealthy backdoor in DNNs of AI-based CIoT named DEST, which is more invisible in the spatial domain and even imperceptible in the frequency domain. Specifically, we use the singular value decomposition method to fuse the feature components of the trigger into the clean image which is wavelet and fourier transformed at the high frequency components, and use the fourier transform to limit the size of the fused components. We further provide theoretical explanations and literature support for feasibility of the proposed method. Extensive experiments on four datasets and three popular classifiers show that the proposed backdoor attack outperforms other state-of-the-art backdoor methods.
- 13 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71  (2) , 6381-6391
Frequency is Devious: Deep Stealthy Backdoor in Artificial Intelligence-Based Consumer Internet of Things
Artificial Intelligence (AI) based Consumer Internet of Things (CIoT) flourishes at a rapid speed due to its excellent data collection ability, which plays an important role in optimizing Deep Neural Networks (DNNs) of AI-based CIoT. However, since the CIoT devices can be accessed without the user’s permission, the DNNs of AI-based CIoT face various security threats, especially the backdoor attack caused by data manipulation. Existing backdoor attacks only concentrate on designing imperceptible triggers in the spatial domain, but not the frequency domain, making the triggers easily detectable or even removable by recent defense methods or humans. In this paper, we reveal a DEep STealthy backdoor in DNNs of AI-based CIoT named DEST, which is more invisible in the spatial domain and even imperceptible in the frequency domain. Specifically, we use the singular value decomposition method to fuse the feature components of the trigger into the clean image which is wavelet and fourier transformed at the high frequency components, and use the fourier transform to limit the size of the fused components. We further provide theoretical explanations and literature support for feasibility of the proposed method. Extensive experiments on four datasets and three popular classifiers show that the proposed backdoor attack outperforms other state-of-the-art backdoor methods.
- Open Access
- 24 March 2024
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 38  (3) , 1851-1859
A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives
Backdoor attacks pose serious security threats to deep neural networks (DNNs). Backdoored models make arbitrarily (targeted) incorrect predictions on inputs containing well-designed triggers, while behaving normally on clean inputs. Prior researches have explored the invisibility of backdoor triggers to enhance attack stealthiness. However, most of them only focus on the invisibility in the spatial domain, neglecting the generation of invisible triggers in the frequency domain. This limitation renders the generated poisoned images easily detectable by recent defense methods. To address this issue, we propose a DUal stealthy BAckdoor attack method named DUBA, which simultaneously considers the invisibility of triggers in both the spatial and frequency domains, to achieve desirable attack performance, while ensuring strong stealthiness. Specifically, we first use Wavelet Transform to embed the high-frequency information of the trigger image into the clean image to ensure attack effectiveness. Then, to attain strong stealthiness, we incorporate Fourier Transform and Cosine Transform to mix the poisoned image and clean image in the frequency domain. Moreover, DUBA adopts a novel attack strategy, training the model with weak triggers and attacking with strong triggers to further enhance attack performance and stealthiness. DUBA is evaluated extensively on four datasets against popular image classifiers, showing significant superiority over state-of-the-art backdoor attacks in attack success rate and stealthiness.
- 24 March 2024
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 38  (3) , 1851-1859
Open AccessA Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives
Backdoor attacks pose serious security threats to deep neural networks (DNNs). Backdoored models make arbitrarily (targeted) incorrect predictions on inputs containing well-designed triggers, while behaving normally on clean inputs. Prior researches have explored the invisibility of backdoor triggers to enhance attack stealthiness. However, most of them only focus on the invisibility in the spatial domain, neglecting the generation of invisible triggers in the frequency domain. This limitation renders the generated poisoned images easily detectable by recent defense methods. To address this issue, we propose a DUal stealthy BAckdoor attack method named DUBA, which simultaneously considers the invisibility of triggers in both the spatial and frequency domains, to achieve desirable attack performance, while ensuring strong stealthiness. Specifically, we first use Wavelet Transform to embed the high-frequency information of the trigger image into the clean image to ensure attack effectiveness. Then, to attain strong stealthiness, we incorporate Fourier Transform and Cosine Transform to mix the poisoned image and clean image in the frequency domain. Moreover, DUBA adopts a novel attack strategy, training the model with weak triggers and attacking with strong triggers to further enhance attack performance and stealthiness. DUBA is evaluated extensively on four datasets against popular image classifiers, showing significant superiority over state-of-the-art backdoor attacks in attack success rate and stealthiness.
- Open Access
- 19 March 2024
- Published by Springer Nature in Neural Processing Letters
- Vol. 56  (2) , 1-21
Enhanced Coalescence Backdoor Attack Against DNN Based on Pixel Gradient
Deep learning has been widely used in many applications such as face recognition, autonomous driving, etc. However, deep learning models are vulnerable to various adversarial attacks, among which backdoor attack is emerging recently. Most of the existing backdoor attacks use the same trigger or the same trigger generation approach to generate the poisoned samples in the training and testing sets, which is also commonly adopted by many backdoor defense strategies. In this paper, we develop an enhanced backdoor attack (EBA) that aims to reveal the potential flaws of existing backdoor defense methods. We use a low-intensity trigger to embed the backdoor, while a high-intensity trigger to activate it. Furthermore, we propose an enhanced coalescence backdoor attack (ECBA) where multiple low-intensity incipient triggers are designed to train the backdoor model, and then, all incipient triggers are gathered on one sample and enhanced to launch the attack. Experiment results on three popular datasets show that our proposed attacks can achieve high attack success rates while maintaining the model classification accuracy of benign samples. Meanwhile, by hiding the incipient poisoned samples and preventing them from activating the backdoor, the proposed attack exhibits significant stealth and the ability to evade mainstream defense methods during the model training phase.
- 19 March 2024
- Published by Springer Nature in Neural Processing Letters
- Vol. 56  (2) , 1-21
Open AccessEnhanced Coalescence Backdoor Attack Against DNN Based on Pixel Gradient
Deep learning has been widely used in many applications such as face recognition, autonomous driving, etc. However, deep learning models are vulnerable to various adversarial attacks, among which backdoor attack is emerging recently. Most of the existing backdoor attacks use the same trigger or the same trigger generation approach to generate the poisoned samples in the training and testing sets, which is also commonly adopted by many backdoor defense strategies. In this paper, we develop an enhanced backdoor attack (EBA) that aims to reveal the potential flaws of existing backdoor defense methods. We use a low-intensity trigger to embed the backdoor, while a high-intensity trigger to activate it. Furthermore, we propose an enhanced coalescence backdoor attack (ECBA) where multiple low-intensity incipient triggers are designed to train the backdoor model, and then, all incipient triggers are gathered on one sample and enhanced to launch the attack. Experiment results on three popular datasets show that our proposed attacks can achieve high attack success rates while maintaining the model classification accuracy of benign samples. Meanwhile, by hiding the incipient poisoned samples and preventing them from activating the backdoor, the proposed attack exhibits significant stealth and the ability to evade mainstream defense methods during the model training phase.