Article
Chemistry, Multidisciplinary
Hiskias Dingeto, Juntae Kim
Summary: While Machine Learning has security flaws, this paper proposes a Universal Adversarial Training algorithm using an AC-GAN to generate adversarial examples. By enhancing the AC-GAN architecture and comparing its performance to other models, it is shown that generative models are better suited for boosting adversarial security through adversarial training.
APPLIED SCIENCES-BASEL
(2023)
Article
Chemistry, Applied
Tong Lei, Qingxia Li, Da-Wen Sun
Summary: This study used a nondestructive Terahertz time-domain imaging system for evaluating the energy and moisture distributions of sunflower seed kernels. A semi-supervised model was developed to learn the kernel information and obtain high-quality chemical distribution maps.
Article
Computer Science, Artificial Intelligence
Christian M. Dahl, Emil N. Sorensen
Summary: This study proposes a novel bootstrap procedure for time series data based on Generative Adversarial networks (GANs). The researchers demonstrate that GANs can learn the dynamics of common stationary time series processes and generate additional samples from the process. The study also compares the performance of GAN sampling and circular block bootstrapping using simulations.
Article
Computer Science, Artificial Intelligence
Ailin Li, Lei Zhao, Zhiwen Zuo, Zhizhong Wang, Haibo Chen, Dongming Lu, Wei Xing
Summary: This paper presents a text-to-image generation method based on conditional Generative Adversarial Networks (GANs), which includes three components: contextual text embedding module, deep Mutual Information estimation module for stacked image generation, and text-image semantic relevance module. Experimental results demonstrate that the method is able to generate text-matched and more diverse images without quality degradation compared to state-of-the-art approaches.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2021)
Article
Computer Science, Artificial Intelligence
Yongqi Tian, Xueyuan Gong, Jialin Tang, Binghua Su, Xiaoxiang Liu, Xinyuan Zhang
Summary: With the rapid development of artificial intelligence, image generation based on deep learning has made significant progress. However, conventional GANs based on convolution are limited in capturing deep-level details and prone to overfitting due to spatial and channel restrictions. In this study, we propose a model called GIU-GANs that leverages a GIU module and representative BN to enhance the quality of generated images. Experimental results demonstrate that the proposed model achieves state-of-the-art performance on CIFAR-10 and CelebA datasets.
Article
Computer Science, Interdisciplinary Applications
Yiqing Shen, Arcot Sowmya, Yulin Luo, Xiaoyao Liang, Dinggang Shen, Jing Ke
Summary: Currently, data-driven based machine learning is widely used in clinical pathology analysis, and the challenge lies in the privacy and security assurance of distributed data samples and the variation of stain styles. To tackle these challenges, this study proposes a novel conditional GAN with one generator and multiple discriminators, implemented within a FL paradigm. The proposed model achieves similar performance to centralized learning and outperforms state-of-the-art stain-style transfer methods in image classification.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2023)
Article
Computer Science, Information Systems
Fei Ye, Adrian G. Bors
Summary: The paper introduces a novel deep learning model that learns disentangled and interpretable data representations for both supervised and unsupervised learning. It utilizes an optimization function based on mutual information maximization criterion and defines a lower bound for the mutual information between latent variables in joint distributions.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Pengcheng Zhao, Yanxiang Chen, Lulu Zhao, Guang Wu, Xi Zhou
Summary: The paper proposes a semantic consistency audio-to-image generative adversarial network (SCAIGAN) that can generate visual images with corresponding semantics directly from audio spectrograms. By utilizing self-attention mechanism, projection mechanism, and self-modulation batch normalization, the model is capable of generating clear and diverse images while achieving better classification accuracy than other methods.
Article
Computer Science, Artificial Intelligence
Xingyuan Chen, Peng Jin, Ping Cai, Hongjun Wang, Xinyu Dai, Jiajun Chen
Summary: This study examines the issues with language models based on generative adversarial networks (GANs) in generating texts, and proposes new metric functions to measure the distributional discrepancy between real and generated texts. Unlike existing methods, this study also evaluates the performance during the adversarial learning process.
CONNECTION SCIENCE
(2022)
Article
Computer Science, Artificial Intelligence
Yang Zhang, Ivor W. Tsang, Yawei Luo, Changhui Hu, Xiaobo Lu, Xin Yu
Summary: This paper proposes a Recursive Copy and Paste Generative Adversarial Network (Re-CPGAN) to recover authentic high-resolution (HR) face images while compensating for non-uniform illumination. The method utilizes internal and recursive external Copy and Paste networks to progressively offset illumination and enhance facial details. Extensive experiments show that the proposed method outperforms state-of-the-art methods in achieving authentic HR face images with a 16x magnification factor under uniform illumination conditions.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Hardware & Architecture
Stella Ho, Youyang Qu, Bruce Gu, Longxiang Gao, Jianxin Li, Yong Xiang
Summary: In the era of big data, a novel approach combining differential privacy and generative adversarial nets is proposed to generate synthetic datasets with indistinguishable statistic features. By employing a min-max game with three players, a deep generative model, DP-GAN model, is devised for synthetic data generation while maintaining a balance between privacy protection and data utility. Extensive simulation results on a real-world dataset demonstrate the superiority of the proposed model in terms of privacy protection, data utility, and efficiency.
JOURNAL OF NETWORK AND COMPUTER APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yanxiang Gong, Zhiwei Xie, Guozhen Duan, Zheng Ma, Mei Xie
Summary: Mode collapse is a significant issue in generative adversarial networks (GANs), where nonuniform sampling can result in missing subdistributions. Even when the generated distribution differs from the real one, GAN can achieve the minimum. To address this, a global distribution fitting (GDF) method with a penalty term and a local distribution fitting (LDF) method are proposed. Experiments demonstrate the effectiveness and competitive performance of GDF and LDF.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yoon-Jae Yeo, Min-Cheol Sagong, Seung Park, Sung-Jea Ko, Yong-Goo Shin
Summary: This paper introduces a novel normalization method called self pixel-wise normalization (SPN) in image generation technique. SPN improves generative performance by pixel-adaptive affine transformation without the need for an external guidance map. Experimental results demonstrate that SPN significantly enhances the performance of image generation technique.
APPLIED INTELLIGENCE
(2023)
Article
Biochemistry & Molecular Biology
Xihao Chen, Jingya Yu, Shenghua Cheng, Xiebo Geng, Sibo Liu, Wei Han, Junbo Hu, Li Chen, Xiuli Liu, Shaoqun Zeng
Summary: This article proposes an unsupervised method to normalize cytopathology image styles through a two-stage style normalization framework, achieving superior results on six cervical cell datasets from different hospitals and scanners. The method greatly improves the recognition accuracy of lesion cells on unseen cytopathology images, meaningful for model generalization.
COMPUTATIONAL AND STRUCTURAL BIOTECHNOLOGY JOURNAL
(2021)
Article
Computer Science, Artificial Intelligence
Jelica Vasiljevic, Friedrich Feuerhake, Cedric Wemmert, Thomas Lampert
Summary: This paper proposes a unified framework, HistoStarGAN, that can perform multiple stain transfer, stain normalization, and stain invariant segmentation in one inference of the model. We demonstrate the generalization abilities of the proposed solution in diverse and unseen stainings, which is the first such demonstration in the field. Moreover, the pre-trained HistoStarGAN model can be used as a synthetic data generator to improve the training of deep learning-based algorithms using fully annotated synthetic image data.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Liming Xu, Xianhua Zeng, He Zhang, Weisheng Li, Jianbo Lei, Zhiwei Huang
Article
Computer Science, Artificial Intelligence
Liming Xu, Xianhua Zeng, Weisheng Li, Zhiwei Huang
Article
Computer Science, Information Systems
Liming Xu, Xianhua Zeng, Weisheng Li, Ling Bai
Summary: This paper proposes IDHashGAN, a novel deep hashing model with generative adversarial networks, for retrieving incomplete data. The model integrates feature restoration, feature learning, and hash coding into an end-to-end framework, leading to significant improvements in performance.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Computer Science, Information Systems
Liming Xu, Xianhua Zeng, Weisheng Li, Bochuan Zheng
Summary: Due to metallic implants, the CT images of certain patients often have metal artifacts, which is a severe problem. Existing methods for reducing metal artifacts are still inadequate due to issues like symptom variance, second artifact, and poor subjective evaluation. To address these issues, a novel method based on GANs is proposed, which incorporates interactive information and imaging CT to overcome limitations of single-modal CT images. The method also includes an enhancement network to avoid second artifact and enhance edge and suppress noise. Experimental results show significant improvement over other methods in terms of both objective metrics and subjective evaluation.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2023)
Article
Engineering, Biomedical
Liming Xu, He Zhang, Lanyu Song, Yanrong Lei
Summary: Magnetic Resonance Imaging (MRI) is widely used in clinical treatment and image-guide therapy, but its high cost and long processing time pose financial burden and low throughput. This study proposes a bidirectional prediction method using multi-generative multi-adversarial nets (Bi-MGAN) to predict desired modality from acquired modality. The proposed method achieves significant improvements in preserving pathological features and anatomical structures, compared to other state-of-the-art methods.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2022)
Article
Computer Science, Artificial Intelligence
Liming Xu, Xianhua Zeng, Weisheng Li, Yicai Xie
Summary: This paper proposes a novel bidirectional hash_code-to-image translation model using multi-generative multi-adversarial nets to achieve high retrieval precision, reduce storage cost, and satisfactory user acceptance. The supervised manifold metric is introduced to optimize the hash codes, and multi-generative and multi-adversarial networks are utilized for hash mapping and inverse hash generation. Theoretical analysis is conducted to avoid unstable training and mode collapse. Experimental results demonstrate that the proposed method outperforms state-of-the-art approaches and significantly reduces storage cost.
PATTERN RECOGNITION
(2023)
Article
Ecology
Quan Tang, Liming Xu, Bochuan Zheng, Chunlin He
Summary: Bird strikes in low-altitude areas can lead to economic losses and endanger the lives of airline passengers. This paper proposes an effective bird identification algorithm using a vision transformer (ViT) and a Mel frequency cepstral coefficient (MFCC) flow framework. The algorithm preprocesses the sound signal and extracts sound features using MFCC flow, which are then normalized for identification using a visual feature network. The proposed Transound method achieves significant improvements compared to recent state-of-the-art approaches.
ECOLOGICAL INFORMATICS
(2023)
Review
Computer Science, Artificial Intelligence
Liming Xu, Quan Tang, Jiancheng Lv, Bochuan Zheng, Xianhua Zeng, Weisheng Li
Summary: Image captioning, or report generation in medical field, is crucial for human-computer interaction and computer-aided diagnosis. With the development of deep learning technologies, image captioning has gained increasing attention in AI-related fields. This study provides a comprehensive overview of different deep image captioning methods in natural and medical fields, including feature representation, visual encoding, and language generation. It also discusses datasets, evaluations, typical caption methods, advantages and disadvantages of existing methods, and concludes with challenges.
Article
Computer Science, Artificial Intelligence
Liming Xu, Xianhua Zeng, Bochuan Zheng, Weisheng Li
Summary: In this paper, a novel cross-modal hashing method called multi-manifold deep discriminative cross-modal hashing (MDDCH) is proposed for large-scale medical image retrieval. The method integrates multiple sub-manifolds defined on heterogeneous data to preserve correlation among instances and improves the discriminative performance of hash codes. Experimental results show that the proposed method achieves comparable performance to state-of-the-art hashing methods.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Engineering, Biomedical
Wenwen Wu, Yanqi Huang, Xiaomei Wu
Summary: In this study, a 2D deep learning classification network SRT was proposed to improve automatic ECG analysis. The model structure was enhanced with the CNN and Transformer-encoder modules, and a novel attention module and Dilated Stem structure were introduced to improve feature extraction. Comparative experiments showed that the proposed model outperformed several advanced methods.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Chiheb Jamazi, Ghaith Manita, Amit Chhabra, Houssem Manita, Ouajdi Korbaa
Summary: In this study, a new dynamic and intelligent clustering method for brain tumor segmentation is proposed by combining the improved Aquila Optimizer (AO) and the K-Means algorithm. The proposed MAO-Kmeans approach aims to automatically extract the correct number and location of cluster centers and the number of pixels in each cluster in abnormal MRI images, and the experimental results demonstrate its effectiveness in improving the performance of conventional K-means clustering.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Alberto Hernando, Maria Dolores Pelaez-Coca, Eduardo Gil
Summary: This study applied a new algorithm to decompose the photoplethysmogram (PPG) pulse and identified changes in PPG pulse morphology due to pressure. The results showed that there was an increase in amplitude, width, and area values of the PPG pulse, and a decrease in ratios when pressure increased, indicating vasoconstriction. Furthermore, some parameters were found to be related to the pulse-to-pulse interval.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Jens Moeller, Eveline Popanda, Nuri H. Aydin, Hubert Welp, Iris Tischoff, Carsten Brenner, Kirsten Schmieder, Martin R. Hofmann, Dorothea Miller
Summary: In this study, a method based on texture features is proposed, which can classify healthy gray and white matter against glioma degrees 4 samples with reasonable classification performance using a relatively low number of samples for training. The method achieves high classification performance without the need for large datasets and complex machine learning approaches.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Amrutha Bhaskaran, Manish Arora
Summary: The study evaluates a cyclic repetition frequency-based algorithm for fetal heart rate estimation. The algorithm improves accuracy and reliability for poor-quality signals and performs well for different gestation weeks and clinical settings.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Manan Patel, Harsh Bhatt, Manushi Munshi, Shivani Pandya, Swati Jain, Priyank Thakkar, Sangwon Yoon
Summary: Electroencephalogram (EEG) signals have been effectively used to measure and analyze neurological data and brain-related ailments. Artificial Intelligence (AI) algorithms, specifically the proposed CNN-FEBAC framework, show promising results in studying the EEG signals of autistic patients and predicting their response to stimuli with 91% accuracy.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Wencheng Gu, Kexue Sun
Summary: This research proposes an improved version of YOLOv5 (AYOLOv5) based on the attention mechanism to address the issue of low recognition rate in cell detection. Experimental results demonstrate that AYOLOv5 can accurately identify cell targets and improve the quality and recognition performance of cell pictures.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Anita Gade, V. Vijaya Baskar, John Panneerselvam
Summary: Analysis of exhaled breath is an increasingly used diagnostic technique in medicine. This study introduces a new NICBGM-based model that utilizes various features and weight optimization for accurate data interpretation and result optimization.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Arsalan Asemi, Keivan Maghooli, Fereidoun Nowshiravan Rahatabad, Hamid Azadeh
Summary: Biometric authentication systems can perform identity verification with optimal accuracy in various environments and emotional changes, while the performance of signature verification systems can be affected when people are under stress. This study examines the performance of a signature verification system based on muscle synergy patterns as biometric characteristics for stressed individuals. EMG signals from hand and arm muscles were recorded and muscle synergies were extracted using Non-Negative Matrix Factorization. The extracted patterns were classified using Support Vector Machine for authentication of stressed individuals.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Tianjiao Guo, Jie Yang, Qi Yu
Summary: This paper proposes a CNN-based approach for segmenting four typical DR lesions simultaneously, achieving competitive performance. This approach is significant for DR lesion segmentation and has potential in other segmentation tasks.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
G. Akilandasowmya, G. Nirmaladevi, S. U. Suganthi, A. Aishwariya
Summary: This study proposes a technique for skin cancer detection and classification using deep hidden features and ensemble classifiers. By optimizing features to reduce data dimensionality and combining ensemble classifiers, the proposed method outperforms in skin cancer classification and improves prediction accuracy.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Tuuli Uudeberg, Juri Belikov, Laura Paeske, Hiie Hinrikus, Innar Liiv, Maie Bachmann
Summary: This article introduces a novel feature extraction method, the in-phase matrix profile (pMP), specifically adapted for electroencephalographic (EEG) signals, for detecting major depressive disorder (MDD). The results show that pMP outperforms Higuchi's fractal dimension (HFD) in detecting MDD, making it a promising method for future studies and potential clinical use for diagnosing MDD.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
P. Nancy, M. Parameswari, J. Sathya Priya
Summary: Stroke is the third leading cause of mortality worldwide, and early detection is crucial to avoid health risks. Existing research on disease detection using machine learning techniques has limitations, so a new stroke detection system is proposed. The experimental results show that the proposed method achieves a high accuracy rate in stroke detection.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Shimin Liu, Zhiwen Huang, Jianmin Zhu, Baolin Liu, Panyu Zhou
Summary: In this study, a continuous blood pressure (BP) monitoring method based on random forest feature selection (RFFS) and a gray wolf optimization-gradient boosting regression tree (GWO-GBRT) prediction model was developed. The method extracted features from electrocardiogram (ECG) and photoplethysmography (PPG) signals, and employed RFFS to select sensitive features highly correlated with BP. A hybrid prediction model of gray wolf optimization (GWO) technique and gradient boosting regression tree (GBRT) algorithm was established to learn the relationship between BP and sensitive features. Experimental results demonstrated the effectiveness and advancement of the proposed method.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Engineering, Biomedical
Weijun Gong, Yurong Qian, Weihang Zhou, Hongyong Leng
Summary: The recognition of dynamic facial expressions is challenging due to various factors, and obtaining discriminative expression features has been difficult. Traditional deep learning networks lack understanding of global and temporal expressions. This study proposes an enhanced spatial-temporal learning network to improve dynamic facial expression recognition.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)