Search

Combining synthetic aperture radar and optical imagery can effectively improve crop type identification. However, optical remote sensing imagery is limited by cloud contamination. In this study, fake optical images were generated using seven image-to-image translation methods and their performance in improving crop classification accuracies was evaluated. Although the sensitivity analysis of the classification models showed lower similarity between real and fake images in the near-infrared band compared to the green and red bands, a significant improvement was confirmed after adding fake images created by generative adversarial networks (GANs). In this study, we generated fake optical images using GAN–based SAR–optical image transformations and evaluated whether adding these fake optical images contributes to improving the accuracy of crop identification from SAR images. This was especially true for the signed attribute vector image-to-image transformation (SAVI2I) method, which was the most effective, achieving an overall accuracy (OA) of 81.3 % with an allocation disagreement (AD) of 11.8 % and a quantity disagreement (QD) of 6.9 %. In contrast, the OA, AD, and QD were respectively 75.9 %, 18.2 %, and 5.9 % when only vertical-horizontal polarization, vertical–vertical polarization, or the polarization ratio were applied. As a result, it was demonstrated that utilizing fake images generated through GAN–based SAR–optical image transformations is effective in improving the accuracy of crop identification from SAR images. [ABSTRACT FROM AUTHOR]

Chinese landscape paintings, realistic landscape photographs, and oil paintings each possess unique artistic characteristics and painting features. Image-to-image translation between these three domains is an extremely challenging task. Existing image-to-image translation networks suffer from deficiencies in preserving content or conveying style, posing difficulties in achieving this task. To address this issue, we propose a novel two-stage multi-domain generative adversarial network approach (TMGAN). We add edge maps as additional guidance input and implement content control to better retain content information. In addition, we design the IOST (In/Out module for Style Transfer) module to better assist the style transfer task. By employing a clever design, we decompose the image translation task into two stages: content extraction and style injection. In the content extraction stage, TMGAN extracts high-resolution edge images from content images. In the style injection stage, TMGAN takes the high-resolution edge image as input and injects the specified style for generation. Notably, we accomplish this two-stage task using only a single multi-domain generator network. Extensive qualitative and quantitative experiments conducted against the baseline model validate the exceptional performance of TMGAN. Furthermore, to facilitate further research, we release MLHQ, a high-quality multi-domain landscape dataset. [ABSTRACT FROM AUTHOR]

Balancing security with image quality is a critical challenge in image encryption, particularly for applications like medical imaging that require high visual fidelity. Traditional encryption methods often fail to preserve image integrity and are vulnerable to advanced attacks. This paper introduces CryptoGAN, a novel GAN-based model designed for image encryption. CryptoGAN employs an architecture to effectively encrypt a dataset of 2000 butterfly images with a resolution of 256x256 pixels, integrating Generative Adversarial Networks (GANs) with symmetric key encryption. Using a U-Net Generator and a PatchGAN Discriminator, CryptoGAN optimizes for key metrics including Structural Similarity Index (SSIM), entropy, and correlation measures. CryptoGAN's performance is comprehensively compared against state-of-theart models such as Cycle GAN-based Image Steganography, EncryptGAN, and DeepEDN. Our evaluation, based on metrics like SSIM, entropy, and PSNR, demonstrates CryptoGAN's superior ability to enhance encryption robustness while maintaining high image quality. Extensive experimental results confirm that CryptoGAN effectively balances security and visual fidelity, making it a promising solution for secure image transmission and storage. This study is supported by a literature survey and detailed analysis of the model architecture, underscoring CryptoGAN's significant contributions to the field of image encryption. [ABSTRACT FROM AUTHOR]

Structural planes are one of the key factors controlling the stability of rock masses. A comprehensive understanding of the spatial distribution characteristics of structural planes is essential for accurately identifying key blocks, analyzing rock mass stability, and addressing various rock engineering challenges. This study compares the effectiveness of four stochastic structural plane generation methods—the Monte Carlo method, the Copula-based method, generative adversarial networks (GAN), and denoised diffusion models (DDPM)—in generating stochastic structural planes and capturing potential correlations between structural plane parameters. The Monte Carlo method employs the mean and variance of three parameters of the measured factual structural planes to generate data that follow the same distributions. The other three methods take the entire set of measured factual structural planes as the overall input to generate structural planes that exhibit the same probability distributions. Five sets of structural planes on four rock slopes in Norway are examined as an example. The validation and analysis were performed using histogram comparison, data feature comparison, scatter plot comparison, and linear regression analysis. The results show that the Monte Carlo method fails to capture the potential correlation between the dip direction and dip angle despite the best fit to the measured factual structural planes. The Copula-based method performs better with smaller datasets, and GAN and DDPM are better at capturing the correlation of measured factual structural planes in the case of large datasets. Therefore, in the case of a limited number of measured structural planes, it is advisable to employ the Copula-based method. In scenarios where the dataset is extensive, the deep generative model is recommended due to its ability to capture complex data structures. The results of this study can be utilized as a valuable point of reference for the accurate generation of stochastic structural planes within rock masses. [ABSTRACT FROM AUTHOR]

Identifying the root causes of part-to-part variation is a central problem in most six-sigma programs, especially of modern manufacturing processes. This is challenging as the sources and patterns of the variation are often unknown or previously unidentified. A small literature aims to address this problem by discovering unknown, previously unidentified variation sources, in a manner that helps understand their nature, from only a sample of measurement data. However, the common solution of this literature is unideal for this objective in terms of both methodology and metrology aspects. This paper proposes a convolutional generative modeling framework for optical scanning data to address this limitation. The proposed approach can correctly discover the true variation sources and visualize their individual patterns in two manufacturing examples, without any prior knowledge of the variation. The approach also outperforms state-of-the-art methods in these examples. [ABSTRACT FROM AUTHOR]

With the advent of generative adversarial networks (GAN), an astonishing advancement has been made in the generation of art painting in recent years. However, existing methods still suffer problems such as color confusion or blurred details. In addition, most of those works centered around the generation of western art painting, while less attention was paid to Chinese traditional arts. Moreover, the lack of traditional Chinese painting datasets is also one of the reasons for the delayed development. To solve the above problems, our research focuses on the synthesis of multi-style traditional Chinese paintings. Firstly, we collect and sort out more than 1000 traditional Chinese paintings, including line drawings, meticulous paintings, ink paintings. Secondly, we propose a Chinese art generative adversarial network (abbreviated as CA-GAN) to decouple the latent vector based on attention mechanism. CA-GAN maps an image to content space and attribute space and fuses them to generate high-quality traditional Chinese art paintings. Meanwhile, a content discriminator is presented to check the consistency of mapping process based on cross-cycle consistency constraint. To make the generated images more artistic, MS-SSIM loss and Charbonnier loss functions are adopted to improve the performance of our model. Experiments have been conducted to verify the effectiveness and the generalization ability of our model. Compared with other state-of-the-art methods, the Chinese art paintings generated by CA-GAN are more vivid and realistic, and the resolutions of them are increased to 280 × 280 . [ABSTRACT FROM AUTHOR]

Blur is one of the common types of damage in the image. Image de-blurring is one of the biggest and most common challenges in the field of image processing. In this article, we present a suitable method for the caused blind blur removal by the motion in the image, in which, in addition to the original image recovery, we recover the kernel of the blur, which is the unknown. The blur kernel is a function that describes the amount and type of blur caused by imaging a point source of light. In our article, the end-to-end learning method for motion blur removal is presented. The presented learning in this paper is the basis of the content loss and the conditional generative adversarial networks. This method gets advanced efficiency in terms of the appearance of the visual and the similarity of the structure. This approach is faster than five times over similar methods. Also, we introduce a new method for artificial motion blur image generation from sharp images that provides the possibility of the increase of the real data. Although the focus of this article is on the removal of the caused blur by the motion on the natural images, our presented approach is capable of removing blur on text images as well, and it partially covers the blur that varies with the location. According to the comparison of existing approaches and the conducted tests, our presented approach has the greatest image output quality value in terms of peak signal-to-noise ratio parameters. [ABSTRACT FROM AUTHOR]

In the Internet era, recommender systems become more and more significant in the daily life. The combination of generative adversarial networks (GAN) and recommended algorithm provides new opportunities for the development of this field. In previous recommendation systems based on GAN, the gradient feedback provided by the discriminator is binary, which does not comprehensively assist the generator. This inadequacy leads to issues such as unstable generator performance and slow model iteration speed, thereby reducing the overall effectiveness of recommendations. Multi-dimensional gradient feedback generative adversarial networks (MGFGAN) is proposed to address above problems. According to the type of generated multidimensional user rating vector, the model incorporates AutoEncoder in the discriminator to provide more diversified feedback for the generator, aiming to make the generated data more closely match the user’s preferences of the model. However, it brings the problem of increasing computational complexity to the model. Therefore, MGFGAN introduces a negative sampling module in the generator, which makes the generator take into account both items of interest and disinterest to the user, thus accelerating the generator to quickly generate data consistent with the real user distribution and improving the efficiency of the model. Finally, the MGFGAN is carried out experimental simulation on the public datasets FilmTrust and Ciaos. Experimental results show that the recommendation performance of MGFGAN outperforms other recommendation models based on GAN and achieves improvements in efficiency and stability.

Characteristic viscosity is a key indicator of the quality of polyester melts, whose accurate prediction can help to identify potential quality problems of polyester melts in advance, adjust the process parameters in time and reduce enterprise losses. Considering the data incompleteness, data time series and high dimensional redundancy of the polyester melt production process, a method is proposed to predict the characteristic viscosity of polyester melt under incomplete data. A missing data generative adversarial nets (MDGAN) with a convolutional neural network discriminator and an attention long short-term memory neural network generator is designed to address the data incompleteness problem caused by the extreme production environment of polyester melts, and the missing data is filled by the adversarial generation mechanism. The extreme gradient boosting-bidirectional gated recurrent unit (XGBoost-BiGRU) is designed to predict the viscosity of polyester melts based on high dimensional redundancy and temporal characteristics prediction. The actual data test results of a polyester fiber manufacturer in Zhejiang show that the filling accuracy of the MDGAN algorithm at different missing rate data sets is better than that of data filling algorithms such as KNN,RF,MICE,and GAIN. The XGBoost-BiGRU characteristic viscosity prediction method has significant advantages over STL-GPR, CAGRU, BiGRU. In combination of MDGAN characteristic viscosity prediction, the method proposed can effectively solve the problem of predicting the characteristic viscosity of polyester melts under incomplete data.

Abstract The number of patents increases quickly, while more and more low-quality patents are emerging. It’s important to identify high-quality patents from massive data quickly and accurately for organizational R&D decision-making and patent layout. However, due to low percentage of high-quality patents, it is challenging to identify them efficiently. In order to solve above problem, we reconstruct the existing index system for identifying high-quality patents by adding 4 features from technological strength of patentees. Furthermore, we propose an improved model by integrating resampling technique and ensemble learning algorithm. First, generative adversarial networks (GAN) are used to expand minority samples. Second, Extreme Gradient Boosting algorithm (XGBoost) with Bayesian optimization (BO) is used to identify high-quality patents. For clarity, this model is called a GAN-BO-XGBoost model. To test the effectiveness of above model, we use patent data in field of lithography technology. Tenfold cross-validation is carried out to evaluate the performance between our proposed model and other models. The results show that GAN-BO-XGBoost model performs better and it’s more stable than other models.

Android is the most targeted mobile operating system for malware attacks. Most modern anti-malware solutions largely incorporate deep learning or machine learning techniques to detect malwares. In this paper, we conduct a comprehensive analysis on 10 deep learning and 5 machine learning classifiers in their abilities to identify Android malware applications. We used 1-gram dataset, 2-gram dataset and image dataset generated from the system call co-occurrence matrix for our experiments. Among the machine learning classifiers, XGBoost with 2-gram dataset showed the highest F1-score of 0.98. Also, the deep learning classifiers such as extreme learning machine with the system call images demonstrated the best F1-score of 0.952. We experimented using Gabor filters to investigate classifier performance on textures extracted from system call images. We observed an F1-score of 0.953 using the extreme learning machine with the Gabor images. We generated the Gabor image dataset by combining the images generated by passing system call images through 25 different Gabor configurations. In addition, to enhance the performance of the baseline classifiers, we considered the combination of autoencoders with machine learning classifiers. We observed that the amalgam of autoencoder with Random Forest displayed the best F1-score of 0.98. To evaluate the effectiveness of the aforesaid classifiers with diverse features on adversarial examples, we simulated a black-box based attack using a Generative Adversarial Network. The True Positive Rate of XGBoost on the 1-gram dataset, Random Forest on the 2-gram dataset and the Extreme Learning Machine on the system call image dataset significantly dropped to 0 from 0.98, 0.001 from 0.99 and 0 from 0.984 after the attack. Our experiments exposed a crucial vulnerability in classifiers used in modern anti-malware systems. A similar event in a real-world system could potentially render grave catastrophes. To defend against such probable attacks, we should continue further research and develop adequate security mechanisms. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Frontiers of Computer Science & Technology is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The number of patents increases quickly, while more and more low-quality patents are emerging. It's important to identify high-quality patents from massive data quickly and accurately for organizational R&D decision-making and patent layout. However, due to low percentage of high-quality patents, it is challenging to identify them efficiently. In order to solve above problem, we reconstruct the existing index system for identifying high-quality patents by adding 4 features from technological strength of patentees. Furthermore, we propose an improved model by integrating resampling technique and ensemble learning algorithm. First, generative adversarial networks (GAN) are used to expand minority samples. Second, Extreme Gradient Boosting algorithm (XGBoost) with Bayesian optimization (BO) is used to identify high-quality patents. For clarity, this model is called a GAN-BO-XGBoost model. To test the effectiveness of above model, we use patent data in field of lithography technology. Tenfold cross-validation is carried out to evaluate the performance between our proposed model and other models. The results show that GAN-BO-XGBoost model performs better and it's more stable than other models. [ABSTRACT FROM AUTHOR]

The convergence of Internet of Things (IoT) and edge computing allows for the collection and sharing of data from human wearable devices. With the huge amount of data transferred among those devices, privacy lies at the forefront of the concerns that must be addressed while preserving the usefulness of the shared data. This research proposes a microAggregation-generative based privacy-preserving model for human activity recognition by analyzing IoT data. Although generative deep neural networks have been widely used for data perturbation and privacy-preserving models, data leakage and disclosing private information of the training samples through linkage attacks remain as major threats when employing traditional anonymization approaches. In addition, noisy records pose a threat to both privacy and quality of the resulting anonymized data. To address these challenges, we propose a novel approach to perturb IoT data using Generative Adversarial Networks (GAN) and microAggregation while preserving both data privacy and utility. Our approach reduces the size of the original dataset by employing an entropy-preserving measure to discard outlier records when data is microaggregated. The performance of the proposed approach was measured using several criteria such as Classification Accuracy, Precision, Recall, and F-score to compare before and after anonymization. As a result, the proposed GAN-MicroAggregation privacy-preserving technique showed a remarkable performance in terms of preserving accuracy after anonymization. Moreover, the privacy of the anonymized data was measured, showing the benefits of the proposed approach when sharing IoT datasets with minimal data inference attack surface. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Shanghai Jiao Tong University (1006-2467) is the property of Journal of Shanghai Jiao Tong University Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Face images distributed widely on social networks are vulnerable to inferring sensitive information by unauthorized automatic identification systems, which poses a threat to user privacy. To protect face privacy, several methods have been proposed to generate highly transferable adversarial faces to remove identity information. However, the results generated by existing methods still suffer from obvious perturbations that make visual perception poor, which is not friendly for sharing on social networks. This paper proposes an adversarial face generation scheme via beautification, i.e., Adv-beauty. Adv-beauty utilizes a face matcher and a beautification discriminator to collaboratively supervise the training process of the generator, prompting the generator to produce a beauty-like perturbation on the original face to confront the face matcher. In other words, the pixel changes produced by the beauty mask the undesirable visual effects produced by the perturbations. In addition, this paper sets an adversarial threshold for identity loss to prevent face distortion due to excessive deviation of identity features. Sufficient experiments show that Adv-beauty maintains good visual results and is effectively against unknown face recognition classifiers and commercial APIs.

We designed a generative adversarial network and an attention network to solve the brainwave emotion-classification problem. Using spatial attention and channel attention superposition to normalize and enhance the raw EEG data, we effectively solved the defects in the EEG data with weak features and easily disturbed them. First, a cognitive map of the brain in the emotional state was constructed by extracting graphical features from EEG signals. Simultaneously, generative adversarial networks are used to add noise to the cognitive map to generate similar data. The volume of the brain cognitive map has been expanded. The problem of insufficient EEG signal data was solved, and the accuracy and robustness were improved. Finally, we compared the processing abilities of different neural networks using EEG and adversarial signals. Compared with other deep learning models and parameter optimization methods, the proposed model achieved a detection accuracy of 94.87% on the SEED dataset.

The richness of textures and semantic information from RGB images can be supplemented in computer vision by the robustness of thermal images to light variations and weather artifacts. While many models rely on inputs from one sensor modality, image translation among modalities can be a solution. The existing works use large models that only work in one translation direction. This cause problems in limited computation applications, as well as a lack of flexibility to work interchangeably for different modalities. Three channel cameras extract visually rich features, but processing them on embedded platforms becomes a bottleneck. Furthermore, edge computing systems impose the burden of compressing data to be sent elsewhere. To address these issues, we propose a novel architecture with a single lightweight encoder capable of working in dual mode, encoding inputs from both grayscale an thermal images into very compact latent vectors. The encoding is then used for cross-modal image translation, grayscale image colorization and thermal image reconstruction, thus allowing 1) different downstream tasks on different modalities, 2) visually rich features from grayscale images and 3) data compression. Four different generators are employed and the training occurs in adversarial fashion with two discriminators. The loss function proposed contains not only adversarial terms but also reconstruction error terms. They induce consistency and contrast preservation across translation and reconstruction. The results backed by evaluation over multiple metrics demonstrate that the model performs the tasks with competitive quality of translation/reconstruction of images with different lighting conditions. Finally, we perform ablation studies to demonstrate the effectiveness of loss terms combined.

Collecting large-scene synthetic aperture radar (SAR) images with targets of interest (TOI) has been a challenging task. To embed TOI slices into measured large scenes can be a good solution. Current methods for SAR TOI slice generation are mainly based on a single data source. Poor background variability of generated images leads to difficulty in naturally embedding TOI slices into large scenes. This article presents a SAR target background conversion network (BCNet), which combines TOI slices with large-scene slices under the same operating condition. The background of TOI slices is converted to large-scene background while preserving the target scattering characteristics to enhance the target background diversity and variability. Background conversion is a special image style transfer, and BCNet uses CycleGAN as the baseline model. The problem that the baseline model may result in target missing is analyzed by Bayesian theory, and then, a new loss function Bysloss is designed to preserve the characteristics of target shadow and scattering center. A new image fusion module has been developed to generate training data for robust background conversion. In addition, the generated high-quality background conversion images are used for two-way recognition performance verification, large scene, and generated TOI slices fusion verification, respectively. The experimental results have shown that the generated data can be successfully used for SAR automatic target recognition in few-shot conditions, and also have strong potential in generating large-scene SAR images with TOI, SAR deception jamming, and augmenting the target detection dataset.

In the novel power system of urban grid, the multiple resources increase and the data collection becomes more difficult, which lead to a higher random missing data rate. It is difficult to meet the demand for refined analysis and decision making. For the frequent missing data problem in the distribution network, a new missing data filling method for power systems based on fluctuation cross-correlation analysis (FCCA) and generative adversarial network (GAN) is proposed in this paper. Firstly, a multi-dimensional feature extraction method for strongly correlated grid data is proposed by fusing FCCA. Secondly, based on kernel principal component analysis (KPCA), the multi-dimensional feature dataset is dimensionally reduced. Finally, an improved GAN structure is designed, which integrates multi-dimensional features of power grid equipment data to reconstruct low dimensional vectors. The missing data is accurately filled in, and the integrity and availability of the new power system measurement data is improved. The algorithm is validated using real grid data, and the proposed method is also tested in a city grid. The results show that the proposed method has higher filling accuracy than the traditional data filling methods. Therefore, it is conformed that in the case of continuous and significant data environment, integrating strong correlation features for data filling has significant advantages in improving the integrity and availability of measurement data.

Achieving neural synchronization, one must be able to evaluate the degree of cooperation across Artificial Neural Networks (ANNs) on various sides, regardless of each network's particular weights. However, traditional approaches suffer from delays in evaluating collaboration, thereby jeopardizing the concealment of neural coordination. Furthermore, there is a paucity of study on employing a trustworthy Pseudo-Random Number Generator (PRNG) to produce a common input and reciprocate training a group of ANNs. This paper introduces the use of a Generative Adversarial Network (GAN) to successfully handle these issues and synchronize a collection of neural networks for session key switch over. This approach enables efficient and effective assessment of the final synchronization state among multiple ANNs. Reciprocal learning is employed to achieve synchronization between two neural networks and distribute the neural key through a single channel. When the ANNs have previously generated identical outputs, coordination is assessed based on this criterion. The proposed method offers several advantages, including: (1) the generation of ANN input sequences using a PRNG based on a GAN. Additionally, a neural feed-forward structure is utilized, incorporating inputs from a non-random "counter" to represent the statefulness of the PRNG. (2) Moreover, a complex ANNs ring or B-tree-guided group is leveraged to facilitate reciprocal neuronal alignment, leading to the creation of the session key via the public network, (3) The suggested methodology takes into account simple, geometry, and majority attacks, (4) The proposed strategy enables two communication partners to detect full synchronization more rapidly compared to previous approaches. The effectiveness of this recommended approach was thoroughly tested, and the results indicate its superiority over similar methods described in the existing literature. [ABSTRACT FROM AUTHOR]

This paper proposes CCME- GAN (circulatory correction multiscale evaluation-generative adversarial networks) based on the cycle consistency loss, aiming at the problems of texture clutter and poor quality of generated images when the generative adversarial network of image conversion class is dealing with the task of unsupervised style transfer. Firstly, in the design of the network architecture, a multiscale evaluation network architecture based on the three-layer semantic information of images is proposed to enhance the transfer effect from the source domain to the target domain. Secondly, in the improvement of the loss function, a multiscale adversarial loss and a cyclic correction loss are proposed to guide the optimization iteration direction of the model with a stricter target, and generate pictures with better visual quality. Finally, in order to prevent the problem of pattern collapse, this paper adds an attention mechanism in the encoding stage of style features to extract important feature information, and then introduces the ACON activation function in each stage of the network to strengthen the nonlinear expression ability of the network and avoid neuron necrosis. The experimental results show that the FID value of this paper method is reduced by 21.80% and 34.33% compared with CycleGAN and ACL-GAN on the landscape painting style migration dataset. In addition, in order to verify the generalization ability of the model, the generalization experiments are compared on two public datasets, Vangogh2Photo, and Monet2Photo and the FID values are decreased by 7.58%, 18.14% and 4.65%, 6.99% respectively. [ABSTRACT FROM AUTHOR]

Deep learning-based synthetic aperture radar (SAR) ship detection methods have achieved significant progress. However, to locate the SAR ship targets accurately, it demands large-scale SAR images for training and oriented bounding box to annotate the location of the ship targets, which requires a lot of time and manpower. Most of the existing SAR image generation methods fail to generate SAR images and their location labels simultaneously. To address this problem, a cascaded generative adversarial network (CGAN) is proposed in this paper. The generator of CGAN consists of two cascaded parts, which are used to reconstruct the low-resolution and high-resolution SAR images, respectively, aiming to gradually improve the quality of the generated SAR ship images. To restrict the location and pixel value of the generated SAR ship targets, CGAN utilizes position and pixel value constraints as input during training. In this way, numerous SAR ship images and its oriented location labels can be obtained by different position and pixel value constraints. To evaluate the accuracy of the generated location label, we annotate the generated SAR ship images manually and compute the Intersection over Union (IoU) between the generated and the manual location labels. The comparative experiments on classical oriented target detection methods demonstrate that the SAR ship images generated by CGAN are conducive to improving the detection accuracy for oriented SAR ship targets. [ABSTRACT FROM AUTHOR]

Copyright of Electric Power Engineering Technology is the property of Editorial Department of Electric Power Engineering Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Journal of Frontiers of Computer Science & Technology is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Purpose: Lung tumor tracking during stereotactic radiotherapy with the CyberKnife can misrecognize tumor location under conditions where similar patterns exist in the search area. This study aimed to develop a technique for bone signal suppression during kV‐x‐ray imaging. Methods: Paired CT images were created with or without bony structures using a 4D extended cardiac‐torso phantom (XCAT phantom) in 56 cases. Subsequently, 3020 2D x‐ray images were generated. Images with bone were input into cycle‐consistent adversarial network (CycleGAN) and the bone suppressed images on the XCAT phantom (BSIphantom) were created. They were then compared to images without bone using the structural similarity index measure (SSIM) and peak signal‐to‐noise ratio (PSNR). Next, 1000 non‐simulated treatment images from real cases were input into the training model, and bone‐suppressed images of the patient (BSIpatient) were created. Zero means normalized cross correlation (ZNCC) by template matching between each of the actual treatment images and BSIpatient were calculated. Results: BSIphantom values were compared to their paired images without bone of the XCAT phantom test data; SSIM and PSNR were 0.90 ± 0.06 and 24.54 ± 4.48, respectively. It was visually confirmed that only bone was selectively suppressed without significantly affecting tumor visualization. The ZNCC values of the actual treatment images and BSIpatient were 0.763 ± 0.136 and 0.773 ± 0.143, respectively. The BSIpatient showed improved recognition accuracy over the actual treatment images. Conclusions: The proposed bone suppression imaging technique based on CycleGAN improves image recognition, making it possible to achieve highly accurate motion tracking irradiation. [ABSTRACT FROM AUTHOR]

Aiming at the problem that there is abnormal data during the process of collecting data from centrifugal pumps. The causes of abnormal data in the process of collecting data from centrifugal pumps, the optimization of the generative adversarial network（GAN） and the method of abnormal data detection were studied. A method for anomaly detection of centrifugal pump timing data using generative adversarial networks had been proposed（this method could optimize generative adversarial networks to solve the problem of gradient vanishing）. Firstly, the basic model in the framework of the GAN was established by using the long short-term memory neural network, and the temporal correlation of the capture data distribution was enhanced, and the problem of gradient disappearance was solved by using the Wasserstein distance method. Then, a centrifugal pump abnormal data detection experiment bench was built to collect data during the operation of the centrifugal pump and the reasons for the abnormal data were analyzed. Finally, the generator and discriminator of the GAN were trained based on normal data, and the loss score was constructed as a threshold to detect abnormal data by using reconstruction loss and discrimination loss. The research results show that the performance of the GAN in data anomaly detection is better than other unsupervised learning anomaly data detection algorithms such as isolated forest, auto-encoder（AE）, K-Means. The GAN can detect abnormal data of centrifugal pump with an accuracy rate of 89.5%, this method can effectively detect abnormal timing data of centrifugal pumps, achieving the goal of optimizing the database and improving the accuracy of rotating machinery fault diagnosis. In conclusion, the study provides a novel approach to detecting abnormal data during the process of collecting centrifugal pump data. [ABSTRACT FROM AUTHOR]

Generative Adversarial Networks (GAN) show excellent performance in various problems of computer vision, computer graphics, and machine learning, but require large amounts of data and huge computational resources. There is also the issue of unstable training. If the generator and discriminator diverge during the training process, the GAN is subsequently difficult to converge. In order to tackle these problems, various transfer learning methods have been introduced; however, mode collapse, which is a form of overfitting, often arises. Moreover, there were limitations in learning the distribution of the training data. In this paper, we provide a comprehensive review of the latest transfer learning methods as a solution to the problem, propose the most effective method of fixing some layers of the generator and discriminator, and discuss future prospects. The model to be used for the experiment is StyleGAN, and the performance evaluation uses Fréchet Inception Distance (FID), coverage, and density. Results of the experiment revealed that the proposed method did not overfit. The model was able to learn the distribution of the training data relatively well compared to the previously proposed methods. Moreover, it outperformed existing methods at the Stanford Cars, Stanford Dogs, Oxford Flower, Caltech-256, CUB-200–2011, and Insect-30 datasets. [ABSTRACT FROM AUTHOR]

Violent crowd behavior detection has gained significant attention in the computer vision system. Diverse crowd behavior detection approaches are introduced to detect violent behavior but enhancing the recognition rate poses a complex task due to different crowd diversity, mutual occlusion between crowds, and diversity of monitoring scene. Therefore, a crowd behavior recognition mechanism is introduced by Conditional Autoregressive-Tunicate Swarm Algorithm based Generative Adversarial Network (CA-TSA based GAN) to detect violent behavior. Accordingly, the developed CA-TSA is modeled by inheriting Conditional Autoregressive Value at Risk by Regression Quantiles with Tunicate Swarm Algorithm. Initially, the features, such as Tanimoto based Violence Flows descriptor, Local Ternary patterns, and Gray level co-occurrence matrix are extracted from the video frames. Then, the crowd behavior recognition is done by the GAN, which finds the abnormal and the normal crowd behaviors. Here, GAN is trained by the proposed CA-TSA. Moreover, the performance of the proposed method is analyzed using ASLAN challenge dataset. The developed model has the accuracy, sensitivity, and specificity values of 93.688%, 94.261%, and 94.051%, respectively. [ABSTRACT FROM AUTHOR]

The emergence of single-cell RNA sequencing (scRNA-seq) technology has revolutionized the identification of cell types and the study of cellular states at a single-cell level. Despite its significant potential, scRNA-seq data analysis is plagued by the issue of missing values. Many existing imputation methods rely on simplistic data distribution assumptions while ignoring the intrinsic gene expression distribution specific to cells. This work presents a novel deep-learning model, named scMultiGAN, for scRNA-seq imputation, which utilizes multiple collaborative generative adversarial networks (GAN). Unlike traditional GAN-based imputation methods that generate missing values based on random noises, scMultiGAN employs a two-stage training process and utilizes multiple GANs to achieve cell-specific imputation. Experimental results show the efficacy of scMultiGAN in imputation accuracy, cell clustering, differential gene expression analysis and trajectory analysis, significantly outperforming existing state-of-the-art techniques. Additionally, scMultiGAN is scalable to large scRNA-seq datasets and consistently performs well across sequencing platforms. The scMultiGAN code is freely available at https://github.com/Galaxy8172/scMultiGAN. [ABSTRACT FROM AUTHOR]

Under the background of large-scale PV (photovoltaic) integration, generating typical operation scenarios of power systems is of great significance for studying system planning operation and electricity markets. Since the uncertainty of PV output and system load is driven by weather factors to some extent, using PV output, system load, and weather data can allow constructing scenarios more accurately. In this study, we used a TimeGAN (time-series generative adversarial network) based on LSTM (long short-term memory) to generate PV output, system load, and weather data. After classifying the generated data using the k-means algorithm, we associated PV output scenarios and load scenarios using the FP-growth algorithm (an association rule mining algorithm), which effectively generated typical scenarios with weather correlations. In this case study, it can be seen that TimeGAN, unlike other GANs, could capture the temporal features of time-series data and performed better than the other examined GANs. The finally generated typical scenario sets also showed interpretable weather correlations. [ABSTRACT FROM AUTHOR]

Copyright of Automobile Technology is the property of Automobile Technology Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)