Chang Lu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Chang Lu has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Dongshen Han
School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
2 shared publications - Caiyan Qin
School of Robotics and Advanced Manufacture, Harbin Institute of Technology, Shenzhen 518055, China
1 shared publication - Chaoning Zhang
School of Computing, Kyung Hee University, Yongin 17113, Republic of Korea
1 shared publication - Chaowen Hou
Science and Technology on Electromechanical Dynamic Control Laboratory, Beijing Institute of Technology, Beijing 100081, China
1 shared publication - Donggeon Kim
School of Computing, Kyung Hee University, Yongin 17104, Republic of Korea
1 shared publication - Mingming Zhang
School of Robotics and Advanced Manufacture, Harbin Institute of Technology, Shenzhen 518055, China
1 shared publication
Latest Publications
- Open Access
Implicit Neural Representation with Dead-Free Linear Unit for Remote Sensing Images
As a crucial component of multimodal sensing in modern AI agents, remote sensing images have attracted significant attention, for which neural representation is a promising direction. Implicit Neural Representations (INRs) using Multi-Layer Perceptrons (MLPs) have the ability to model images by learning an implicit mapping from pixel coordinates to pixel intensities. This paper revisits the ReLU activation function, a widely adopted non-linearity known for its dead region on the negative axis, within the context of MLP-based INRs. We introduce the Dead-Free Linear Unit (DeLU), a novel activation function that leverages a linearly transformed absolute value to eliminate inactive regions. By combining dead-free non-linearity with adaptive linear scaling, DeLU enhances the expressiveness of INR architectures, particularly those employing periodic activations. Extensive experiments across multiple remote sensing datasets, including LandCover.ai, LoveDA, INRIA, UAVid, and ISPRS Potsdam, validate the efficacy of our proposed method.
Open AccessImplicit Neural Representation with Dead-Free Linear Unit for Remote Sensing Images
As a crucial component of multimodal sensing in modern AI agents, remote sensing images have attracted significant attention, for which neural representation is a promising direction. Implicit Neural Representations (INRs) using Multi-Layer Perceptrons (MLPs) have the ability to model images by learning an implicit mapping from pixel coordinates to pixel intensities. This paper revisits the ReLU activation function, a widely adopted non-linearity known for its dead region on the negative axis, within the context of MLP-based INRs. We introduce the Dead-Free Linear Unit (DeLU), a novel activation function that leverages a linearly transformed absolute value to eliminate inactive regions. By combining dead-free non-linearity with adaptive linear scaling, DeLU enhances the expressiveness of INR architectures, particularly those employing periodic activations. Extensive experiments across multiple remote sensing datasets, including LandCover.ai, LoveDA, INRIA, UAVid, and ISPRS Potsdam, validate the efficacy of our proposed method.
- Open Access
- 3 January 2025
- Published by MDPI AG in Remote Sensing
- Vol. 17 (1) , 146
Transferable Targeted Adversarial Attack on Synthetic Aperture Radar (SAR) Image Recognition
Deep learning models have been widely applied to synthetic aperture radar (SAR) target recognition, offering end-to-end feature extraction that significantly enhances recognition performance. However, recent studies show that optical image recognition models are widely vulnerable to adversarial examples, which fool the models by adding imperceptible perturbation to the input. Although the targeted adversarial attack (TAA) has been realized in the white box setup with full access to the SAR model’s knowledge, it is less practical in real-world scenarios where white box access to the target model is not allowed. To the best of our knowledge, our work is the first to explore transferable TAA on SAR models. Since contrastive learning (CL) is commonly applied to enhance a model’s generalization, we utilize it to improve the generalization of adversarial examples generated on a source model to unseen target models in the black box scenario. Thus, we propose the contrastive learning-based targeted adversarial attack, termed CL-TAA. Extensive experiments demonstrated that our proposed CL-TAA can significantly improve the transferability of adversarial examples to fool the SAR models in the black box scenario.
- 3 January 2025
- Published by MDPI AG in Remote Sensing
- Vol. 17 (1) , 146
Open AccessTransferable Targeted Adversarial Attack on Synthetic Aperture Radar (SAR) Image Recognition
Deep learning models have been widely applied to synthetic aperture radar (SAR) target recognition, offering end-to-end feature extraction that significantly enhances recognition performance. However, recent studies show that optical image recognition models are widely vulnerable to adversarial examples, which fool the models by adding imperceptible perturbation to the input. Although the targeted adversarial attack (TAA) has been realized in the white box setup with full access to the SAR model’s knowledge, it is less practical in real-world scenarios where white box access to the target model is not allowed. To the best of our knowledge, our work is the first to explore transferable TAA on SAR models. Since contrastive learning (CL) is commonly applied to enhance a model’s generalization, we utilize it to improve the generalization of adversarial examples generated on a source model to unseen target models in the black box scenario. Thus, we propose the contrastive learning-based targeted adversarial attack, termed CL-TAA. Extensive experiments demonstrated that our proposed CL-TAA can significantly improve the transferability of adversarial examples to fool the SAR models in the black box scenario.