PDBSNet: Pixel-Shuffle Downsampling Blind-Spot Reconstruction Network for Hyperspectral Anomaly Detection

Wang, Degang; Zhuang, Lina; Gao, Lianru; Sun, Xingming; Huang, Min

doi:10.1109/tgrs.2023.3276175

Cited by 39 publications

(2 citation statements)

References 62 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Convolutional Neural Networks (CNNs) are recognised for their efficiency in processing local features through their convolutional layers, while Vision Transformers (ViTs) excel at capturing global dependencies in an image through self-attention mechanisms (Maurí cio et al, 2023). The PixelShuffle operation, also known as sub-pixel convolution, is a technique mainly used for upscaling images in super-resolution tasks (Wang et al, 2023b). The combination of Convolutional Vision Transformer structures and PixelShuffle operations in CycleGANs is a powerful tool for advanced image processing tasks.…”

Section: Introductionmentioning

confidence: 99%

NC2C-TransCycleGAN: Non-Contrast to Contrast-Enhanced CT Image Synthesis Using Transformer CycleGAN

Hou,

Liu,

Zhang

et al. 2024

View full text Add to dashboard Cite

Background: Lung cancer poses a great threat to human life and health. Although the density differences between lesions and normal tissues shown on enhanced CT images is very helpful for doctors to characterize and detect lesions, contrast agents and radiation may cause harm to the health of patients with lung cancer. By learning the mapping relationship between plain CT image and enhanced CT image through deep learning methods, high quality synthetic CECT image results can be generated based on plain scan CT image. It has great potential to help save treatment time and cost of lung cancer patients, reduce radiation dose and expand the medical image dataset in the field of deep learning. Methods: In this study, plain and enhanced CT images of 71 lung cancer patients were retrospectively collected. The data from 58 lung cancer patients were randomly assigned to the training set, and the other 13 cases formed the test set. The Convolution Vison Transformer structure and PixelShuffle operation were combined with CycleGAN, respectively, to help generate clearer images. After random erasing, image scaling and flipping to enhance the training data, paired plain and enhanced CT slices of each patient are input into the network as input and labeled, respectively, for model training.Finally, the peak signal-to-noise ratio, structural similarity and mean square error are used to evaluate the image quality and similarity. Results: The performance of our proposed method is compared with CycleGAN and Pix2Pix on the test set, respectively. The results show that the SSIM value of the enhanced CT images generated by the proposed method improve by 2.00% and 1.39%, the PSNR values improve by 2.05% and 1.71%, and the MSE decreases by 12.50% and 8.53%, respectively, compared with Pix2Pix and CycleGAN. Conclusions: The experimental results show that the improved algorithm based on CylceGAN proposed in this paper can synthesize high-quality lung cancer synthetic enhanced CT images, which is helpful to expand the lung cancer image data set in the deep learning research. More importantly, this method can help lung cancer patients save medical treatment time and cost.

show abstract

Section: Introductionmentioning

confidence: 99%

NC2C-TransCycleGAN: Non-Contrast to Contrast-Enhanced CT Image Synthesis Using Transformer CycleGAN

Hou,

Liu,

Zhang

et al. 2024

View full text Add to dashboard Cite

show abstract

“…For the unsupervised HAD methods, the dominant behavior is to achieve anomaly detection by reconstructing backgrounds without anomalies. So, the latest CNN-based method provides a new solution for hyperspectral anomaly detection which designs a blind spot strategy to train a background reconstruction network to detect anomalies [31]. The AE-based anomaly detectors consider that the main background distribution in HSI can be more easily reconstructed by a well-designed AE network than that of anomalies.…”

mentioning

confidence: 99%

Tucker Decomposition-Based Network Compression for Anomaly Detection With Large-Scale Hyperspectral Images

Wang,

Zhao

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Deep learning methodologies have demonstrated considerable effectiveness in hyperspectral anomaly detection (HAD). However, the practicality of deep learning-based HAD in real-world applications is impeded by challenges arising from limited labeled data, large-scale hyperspectral images and constrained computational resources. In light of these challenges, this paper introduces a convolutional neural network-based HAD model through the incorporation of Tucker decomposition, named as TD-CNND. Drawing inspiration from transfer learning, the proposed model initially constructs pixel sample pairs from known labeled hyperspectral images in the source domain, feeding them into the designed CNN to train the network learning spectral feature differences to obtain a convolutional neural network containing knowledge from the source domain. Subsequently, to prevent the need for network retraining caused by structural changes and to reduce model parameters for improving detecting timeliness, a general network compression scheme based on Tucker decomposition is applied to the convolutional neural network, where the convolutional layers of the above CNN undergo Tucker tensor decomposition to compress the network and alleviate parameter redundancy. Finally, spectral features realignment is used to recover the detection accuracy loss caused by Tucker tensor decomposition. In addition, a dual-windows structure is implemented during the detection phase, incorporating spatial information to the aforementioned spectrallevel learning model, facilitating spectral-spatial collaborative hyperspectral anomaly detection. Experimental evaluations using three real hyperspectral datasets and artificially expanded datasets demonstrate that, in comparison with state-of-the-art methods, the proposed TD-CNND method exhibits effectiveness and superiority in terms of both time cost and detection accuracy, where the notable advantages in terms of time cost become more pronounced with an increasing number of pixels.

show abstract

Deep Learning for Anomaly Detection in Time-Series Data: An Analysis of Techniques, Review of Applications, and Guidelines for Future Research

Usmani,

Abdul Aziz,

Jaafar

et al. 2024

IEEE Access

View full text Add to dashboard Cite

PDBSNet: Pixel-Shuffle Downsampling Blind-Spot Reconstruction Network for Hyperspectral Anomaly Detection

Cited by 39 publications

References 62 publications

NC2C-TransCycleGAN: Non-Contrast to Contrast-Enhanced CT Image Synthesis Using Transformer CycleGAN

NC2C-TransCycleGAN: Non-Contrast to Contrast-Enhanced CT Image Synthesis Using Transformer CycleGAN

Tucker Decomposition-Based Network Compression for Anomaly Detection With Large-Scale Hyperspectral Images

Deep Learning for Anomaly Detection in Time-Series Data: An Analysis of Techniques, Review of Applications, and Guidelines for Future Research

Contact Info

Product

Resources

About