CNNs(Convolutional Neural Networks) have a good performance on most classification tasks,but they are vulnerable when meeting adversarial *** and design of highly aggressive adversarial examples can help enhance the s...
CNNs(Convolutional Neural Networks) have a good performance on most classification tasks,but they are vulnerable when meeting adversarial *** and design of highly aggressive adversarial examples can help enhance the security and robustness of *** transferability of adversarial examples is still low in black-box ***,an adversarial example method based on probability histogram equalization,namely HE-MI-FGSM(Histogram Equalization Momentum Iterative Fast Gradient Sign Method) is *** each iteration of the adversarial example generation process,the original input image is randomly histogram equalized,and then the gradient is calculated to generate adversarial perturbations to mitigate overfitting in the adversarial *** effectiveness of the method is verified on the ImageNet *** with the advanced method I-FGSM(Iterative Fast Gradient Sign Method) and MI-FGSM(Momentum I-FGSM),the attack success rate in the adversarial training network increased by 27.9% and 7.7% on average,respectively.
Adding subtle perturbations to an image can cause the classification model to misclassify, and such images are called adversarial examples. Adversarial examples threaten the safe use of deep neural networks, but when ...
Adding subtle perturbations to an image can cause the classification model to misclassify, and such images are called adversarial examples. Adversarial examples threaten the safe use of deep neural networks, but when combined with reversible data hiding(RDH) technology, they can protect images from being correctly identified by unauthorized models and recover the image lossless under authorized models. Based on this, the reversible adversarial example(RAE) is rising. However, existing RAE technology focuses on feasibility, attack success rate and image quality, but ignores transferability and time complexity. In this paper,we optimize the data hiding structure and combine data augmentation technology,which flips the input image in probability to avoid overfitting phenomenon on the dataset. On the premise of maintaining a high success rate of white-box attacks and the image's visual quality, the proposed method improves the transferability of reversible adversarial examples by approximately 16% and reduces the computational cost by approximately 43% compared to the state-of-the-art method. In addition, the appropriate flip probability can be selected for different application scenarios.
Tokenizer, serving as a translator to map the intricate visual data into a compact latent space, lies at the core of visual generative models. Based on the finding that existing tokenizers are tailored to image or vid...
This paper proposes and designs a microstrip antenna that can work in the B1 band for Beidou with a double-layer substrate, and realizes the circularly polarized radiation characteristics. The antenna consists of a gr...
This paper proposes and designs a microstrip antenna that can work in the B1 band for Beidou with a double-layer substrate, and realizes the circularly polarized radiation characteristics. The antenna consists of a ground plane, two substrates, and two metal patches. The first layer of the patch is a rectangular structure as well as the second layer of the patch is a T-shaped structure with two vertical branches which are to achieve circular polarization. This structure adopts a simple coaxial feeding method, with the probe passing through the second layer of the substrate to connect the ground plane and the T-shaped strip. The simulation optimization results using the HFSS 2021 show that the axial ratio (AR) beam width in both planes exceeds 160°, and the AR at the design frequency is 0.18 dB. The impedance bandwidth is 70 MHz, which can meet the requirements of Beidou's operation.
A mimic multi-layer polarization-independent FSS with accessible angular-stability is reported using a three-layer structure, where the top layer and the bottom layer are realized based the metal rings that are same, ...
A mimic multi-layer polarization-independent FSS with accessible angular-stability is reported using a three-layer structure, where the top layer and the bottom layer are realized based the metal rings that are same, and the middle is constructed by a slotted metal structure. The mentioned three layers are aimed to separate the used two dielectric substrates. For a well illustration, the constructed dimensions of the installed FSS are $\boldsymbol{5\times 5}\mathbf{mm}^{2}$ . As a case of the FSS cell design, here the dimension is limited as $\boldsymbol{0.15}\lambda\times \boldsymbol{0.15}\lambda$ . Since two different structure is used, the FSS has double resonace, making a wide operation. From the numerical evaluations, the FSS can reject the signal from C-and X-band.
A compact Sub-6GHz multiple-input-multiple-output (MIMO) antenna is presented in this paper. The recommended MIMO antenna is electrically small (38mm × 38 mm× 1.6 mm). For good isolation and miniaturized siz...
详细信息
This paper introduces a compact wideband 6 port multiple-input multiple-output (MIMO) antenna for the 5G handset. Each antenna elements uses coupling feed, and I-shaped grounding structure is added to the unit to obta...
详细信息
Reconfigurable intelligent Surface (RIS) is a revolutionary technology in modern wireless communication systems, since it can adjust the wireless communication environment with passive components. In this paper, we co...
Reconfigurable intelligent Surface (RIS) is a revolutionary technology in modern wireless communication systems, since it can adjust the wireless communication environment with passive components. In this paper, we consider the system environment in the presence of Base Station (BS) interference and implement the estimation of the target Direction-of-Arrival (DOA). To improve the estimation performance, we propose a DOA estimation algorithm based on RIS. The proposed algorithm consists of an error term for the lorentzian bound function and a regular term for the atomic norm (termed as LBF-AN), which are implemented to eliminate system noise and signal interference, respectively. The simulation results show that the DOA estimation performance of the proposed algorithm is better than existing algorithms.
In this paper, we propose a DenseNet-CBAM model that utilizes DenseNet121 as the backbone network for feature extraction. The features are then weighted using the Convolutional Block Attention Module (CBAM), which inc...
In this paper, we propose a DenseNet-CBAM model that utilizes DenseNet121 as the backbone network for feature extraction. The features are then weighted using the Convolutional Block Attention Module (CBAM), which incorporates both channel and spatial attention mechanisms. The features are further processed with ReLU activation and adaptive average pooling operations. Finally, a linear classifier is applied to output the final category predictions. Experiment results show that the proposed model performs the best in terms of the ability in detecting patients with high-grade squamous intraepithelial lesions (HSIL).
Tandem mass spectrometry (MS/MS) is a widely used technique for protein identification, post-translational modifications, immunotherapy, and other applications. As the amount of MS/MS spectra data increases, new compu...
详细信息
暂无评论