Wyniki wyszukiwania - Biblioteka Nauki

Ten serwis zostanie wyłączony 2025-02-11.

Nowa wersja platformy, zawierająca wyłącznie zasoby pełnotekstowe, jest już dostępna.
Przejdź na https://bibliotekanauki.pl

Ograniczanie wyników

Znaleziono wyników: 4

Liczba wyników na stronie

Wyniki wyszukiwania

Wyszukiwano:
w słowach kluczowych: generative adversarial network

Sortuj według:

Ogranicz wyniki do:

Underwater image enhancement via efficient generative adversarial network

100%

Qian X. , Ge P.

tom Vol. 51, nr 4

483--497

Underwater image enhancement has been receiving much attention due to its significance in facilitating various marine explorations. Inspired by the generative adversarial network (GAN) and residual network (ResNet) in many vision tasks, we propose a simplified designed ResNet model based on GAN called efficient GAN (EGAN) for underwater image enhancement. In particular, for the generator of EGAN we design a new pair of convolutional kernel size for the residual block in the ResNet. Secondly, we abandon batch normalization (BN) after every convolution layer for faster training and less artifacts. Finally, a smooth loss function is introduced for halo-effect alleviation. Extensive qualitative and quantitative experiments show that our methods accomplish considerable improvements compared to the state-of-the-art methods.

Additional Look into GAN-based Augmentation for Deep Learning COVID-19 Image Classification

88%

Fedoruk O. , Klimaszewski K. , Ogonowski A. , Kruk M.

Machine Graphics and Vision

2023

tom Vol. 32, No. 3/4

107--124

Data augmentation is a popular approach to overcome the insufficiency of training data for medical imaging. Classical augmentation is based on modification (rotations, shears, brightness changes, etc.) of the images from the original dataset. Another possible approach is the usage of Generative Adversarial Networks (GAN). This work is a continuation of the previous research where we trained StyleGAN2-ADA by Nvidia on the limited COVID-19 chest X-ray image dataset. In this paper, we study the dependence of the GAN-based augmentation performance on dataset size with a focus on small samples. Two datasets are considered, one with 1000 images per class (4000 images in total) and the second with 500 images per class (2000 images in total). We train StyleGAN2-ADA with both sets and then, after validating the quality of generated images, we use trained GANs as one of the augmentations approaches in multi-class classification problems. We compare the quality of the GAN-based augmentation approach to two different approaches (classical augmentation and no augmentation at all) by employing transfer learning-based classification of COVID-19 chest X-ray images. The results are quantified using different classification quality metrics and compared to the results from the previous article and literature. The GAN-based augmentation approach is found to be comparable with classical augmentation in the case of medium and large datasets but underperforms in the case of smaller datasets. The correlation between the size of the original dataset and the quality of classification is visible independently from the augmentation approach.

A recombination generative adversarial network for intrusion detection

75%

Luo H. , Wan L.

tom Vol. 34, no. 2

323--334

The imbalance and complexity of network traffic data are hot issues in the field of intrusion detection. To improve the detection rate of minority class attacks in network traffic, this paper presents a method for intrusion detection based on the recombination generative adversarial network (RGAN). In this study, dual-stage game learning is used to optimize the discriminator for efficient identification of attack samples. In the first stage, the proposed model trains a deep convolutional generative adversarial network (DCGAN) integrated with the self-attention (SA) mechanism, and simultaneously trains an independent convolutional neural network (CNN) classifier integrated with the gated recurrent unit (GRU). This stage allows the generator to generate minority class attack samples that closely resemble real samples, while the independent classifier possesses the basic classification ability. In the second stage, the generator and the independent classifier of the DCGAN together constitute the second layer of the model - the generative adversarial network. Through dual-stage game learning, the classifier’s discrimination ability for the minority samples is optimized, and it serves as the final output of the discriminator. In addition, the introduction of reconstruction loss helps prevent the detection rate of false positive samples. Experimental results on the CSE-IDS-2018 dataset demonstrate that our model performs well compared with various other intrusion detection techniques in terms of detection accuracy, recall, and F1-score for minority class attacks.

Epileptic seizure prediction using scalp electroencephalogram signals

51%

Usman S. M. , Khalid S. , Bashir Z.

tom Vol. 41, no. 1

211--220

Epilepsy is a brain disorder in which patients undergo frequent seizures. Around 30% of patients affected with epilepsy cannot be treated with medicines/surgical procedures. Abnormal activity, known as the preictal state starts few minutes before the seizure actually occurs. Therefore, it may be possible to deliver medication prior to the occurrence of a seizure if initiation of the preictal state can predicted before the seizure onset. We propose an epileptic seizure prediction method that predicts the preictal state before the seizure onset using electroencephalogram (EEG) monitoring of brain activity. It involves three steps including preprocessing of EEG signals, feature extraction classification of preictal and interictal states. In our proposed method, we have used (i) Empirical model decomposition to remove noise from the EEG signals and Generative Adversarial Networks to generate preictal samples to deal with the class imbalance problem; (ii) Automated features have been extracted with three layer Convolutional Neural Networks and (iii) Classification between preictal and interictal states is done with Long Short Term Memory units. In this study, we have used CHBMIT dataset of scalp EEG signals and have validated our proposed method on 22 subjects of dataset. Our proposed seizure prediction method is able to achieve 93% sensitivity and 92.5% specificity with average time of 32 min to predict the seizure's onset. Results obtained from our method have been compared with recent state-of-the-art epileptic seizure prediction methods. Our proposed method performs better in terms of sensitivity, specificity and average anticipation time.