Article
Environmental Sciences
Alvaro Accion, Francisco Argueello, Dora B. Heras
Summary: A new data augmentation scheme using data imputation and matrix completion methods for segment-based classification has been proposed in this study, successfully increasing classification performance. Data imputation methods applied to multispectral imagery are shown to be a valid means of performing data augmentation, improving model generalization for high spectral resolution imagery.
Article
Engineering, Biomedical
Mayank Mishra, Umesh C. Pati
Summary: Autism Spectrum Disorder (ASD) has a complex and heterogeneous nature, leading to late detection. Magnetic resonance imaging (MRI) has been crucial in detecting various brain disorders, including ASD, due to its noninvasive nature. While Functional MRI (fMRI) has been widely used for ASD detection, this study focuses on utilizing Structural MRI (sMRI) with deep learning approaches. The proposed ensemble model of Deep Convolution Neural Network (DCNN) with Adam and Nadam optimizers achieves high accuracy in ASD detection, outperforming state-of-the-art approaches.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Multidisciplinary Sciences
Oluwasegun Oladipo, Elijah Olusayo Omidiora, Victor Chukwudi Osamor
Summary: This study developed a novel age estimation system targeted at black faces, which outperformed the standard ANN system in terms of correct classification rate, by combining genetic algorithm and back propagation-trained artificial neural network with the local binary pattern feature extraction technique.
SCIENTIFIC REPORTS
(2022)
Article
Computer Science, Artificial Intelligence
Isack Farady, Chih-Yang Lin, Ming-Ching Chang
Summary: This study improves data augmentation by introducing a new offline pre-augmentation network (PreAugNet), which acts as a class boundary classifier to effectively screen the quality of augmented samples and improve image augmentation. PreAugNet can generate augmented samples and update decision boundaries via an independent support vector machine (SVM) classifier. The experiments show that these new augmentation samples can improve classification without changing the target network architecture.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Computer Science, Artificial Intelligence
Ruiting Hao, Chengwei Weng, Xinyu Liu, Xiaorong Yang
Summary: In this paper, an iterative approach based on data augmentation method is proposed for estimating censored QRNN model. The proposed method outperforms existing ones in terms of quantile loss and prediction interval width, and can be easily adapted to different censoring types.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Chemistry, Analytical
Jaehyeop Choi, Chaehyeon Lee, Donggyu Lee, Heechul Jung
Summary: Modern data augmentation methods like Mixup, CutMix, and the proposed SalfMix have shown improved performance in image recognition tasks, especially when combined as in HybridMix. These methods outperformed traditional single image-based approaches and achieved state-of-the-art results in various classification and object detection datasets.
Article
Computer Science, Artificial Intelligence
Ruiting Hao, Huanfeng Zheng, Xiaorong Yang
Summary: In this paper, an iterative estimation method based on the data augmentation algorithm is proposed for the censored Composite Quantile Regression Neural Network (CQRNN) model. Simulation studies and real data application show that the proposed method outperforms existing censored methods and produces results close to the uncensoring case.
APPLIED SOFT COMPUTING
(2022)
Article
Remote Sensing
Rowida Alharbi, Haikel Alhichri, Ridha Ouni, Yakoub Bazi, Maazen Alsabaan
Summary: This study presents an improved data augmentation technique called Quality-based Sample Selection (QSS) for remote sensing (RS) scene classification. It generates a large number of samples using geometric transformations and selects the best ones based on a quality criterion. QSS has been experimentally proven to improve the accuracy of RS scene classification.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Jiarun Yu, Yafeng Wang
Summary: A novel method using deep learning is proposed for direction-of-arrivals (DoAs) estimation in millimeter (mmWave) massive MIMO systems, without prior information about multipath number. The DoAs estimation is decomposed into three sub-problems, solved by corresponding convolutional neural networks (CNNs). The proposed method achieves accurate multipath number estimation using CNN-I, DoA estimation of the line-of-sight (LOS) path using CNN-II, and DoAs estimation of non-line-of-sight (NLOS) paths using CNN-III. The method learns the non-linear relationship between the sample covariance matrix of the received signal and the angles, resulting in superior performance in low signal-to-noise-ratio (SNR) regimes and lower root mean square error (RMSE) compared to other methods.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Saim Ervural, Murat Ceylan
Summary: This study proposes a classification model for neonatal diseases using data augmentation and artificial intelligence methodology, achieving high classification accuracy by combining images through convolutional neural networks.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Environmental Sciences
Rong Yang, Robert Wang, Yunkai Deng, Xiaoxue Jia, Heng Zhang
Summary: The random cropping data augmentation method is widely used to expand the dataset scale and enhance model robustness, but directly introducing it into training of CNN-based SAR image ship detector may introduce gradient noise and hurt detection performance. A training method based on feature map mask is proposed to effectively eliminate gradient noise, significantly improve detection performance, and without increasing inference cost.
Review
Computer Science, Information Systems
Sandeep Kumar Gupta, Neeta Nain
Summary: This paper presents a study on prediction models for facial age and gender recognition, providing a review of conventional and deep learning methods, analyzing their pros and cons, and offering insights for future research. Additionally, the paper lists the databases used for benchmarking results and their properties in constrained and unconstrained environments.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Interdisciplinary Applications
Rishi Raj, Jimson Mathew, Santhosh Kumar Kannath, Jeny Rajan
Summary: This paper proposes a non-linear data augmentation technique called Crossover for Convolutional Neural Networks in Medical Image Classification. The technique creates new samples by applying two-point crossover on the existing training dataset. Experimental results show that the proposed technique performs better in terms of increased accuracy and reduced loss compared to traditional linear and label preserving techniques.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2022)
Article
Computer Science, Hardware & Architecture
Eric Ke Wang, Juntao Yu, Chien-Ming Chen, Saru Kumari, Joel J. P. C. Rodrigues
Summary: This paper addresses the issue of data insufficiency in dialog systems by proposing a data augmentation technique based on a GAN model. The augmented data is effectively generated to improve the robustness of parameter estimation of unknown data. Furthermore, a new N-gram language model is used to evaluate multiple recognition candidates, resulting in enhanced performance of the speech recognition system.
MOBILE NETWORKS & APPLICATIONS
(2022)
Article
Computer Science, Interdisciplinary Applications
Prateek Bansal, Rico Krueger, Daniel J. Graham
Summary: Spatial count data models are used to predict phenomena frequencies in geographically distinct entities. Variational Bayes method offers faster estimation and better performance in both simulation studies and empirical applications.
COMPUTATIONAL STATISTICS & DATA ANALYSIS
(2021)
Article
Engineering, Electrical & Electronic
Running Zhao, Xiaolin Ma, Xinhua Liu, Jian Liu
Summary: This study proposes a continuous human motion recognition method based on radar data, which optimizes model representation, reduces computational complexity, and achieves better performance than existing methods by effectively utilizing all time information through an end-to-end network structure.
IEEE SENSORS JOURNAL
(2021)
Article
Engineering, Electrical & Electronic
Running Zhao, Xiaolin Ma, Xinhua Liu, Fangmin Li
Summary: This paper proposes a novel method for continuous human motion recognition using micro-Doppler signatures, which can effectively handle non-target micro motion interference. By signal preprocessing and tailored network design, accurate recognition of continuous human motion is achieved even in the presence of mixed interference.
IEEE SENSORS JOURNAL
(2021)
Article
Chemistry, Analytical
Xinhua Liu, Wenqian Wei, Hailan Kuang, Xiaolin Ma
Summary: This paper proposes a 3D central difference convolutional network (CDCA-rPPGNet) with an attention mechanism to measure heart rate. The proposed method effectively combines spatial and temporal features and achieves accurate heart rate measurement.
Article
Chemistry, Analytical
Hailan Kuang, Fanbing Lv, Xiaolin Ma, Xinhua Liu
Summary: Studies have shown that ordinary color cameras can detect the subtle color changes of the skin caused by the heartbeat cycle, enabling non-contact monitoring of pulse. This paper proposes an efficient spatiotemporal attention network (ESA-rPPGNet) for recovering high-quality rPPG signals for heart rate variability analysis. The network utilizes 3D depth-wise separable convolution, a lightweight attention block called 3D shuffle attention (3D-SA), and ConvGRU to capture dependencies and improve long-term spatiotemporal feature learning. Experimental results demonstrate better performance and robustness compared to existing methods in remote HRV analysis.
Article
Chemistry, Multidisciplinary
Hailan Kuang, Haoran Chen, Xiaolin Ma, Xinhua Liu
Summary: This paper proposes a keyword detection and context filtering method based on the Self-Attention mechanism for document-level relation extraction (RE). It introduces a Self-Attention Memory (SAM) module in ConvLSTM to process the document context and capture keyword features. By searching for word embeddings with high cross-attention of entity pairs, critical local features are updated and recorded to enhance the performance of the final classification model.
APPLIED SCIENCES-BASEL
(2022)
Article
Chemistry, Multidisciplinary
Xiaolin Ma, Kaiqi Wu, Hailan Kuang, Xinhua Liu
Summary: This paper introduces a span-based joint extraction method based on dynamic context and multi-feature fusion (SPERT-DC) which improves the compatibility with longer text and enhances context information by dynamically selecting context regions and combining with entity labels, thus improving the performance of entity relation extraction.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Information Systems
Xinhua Liu, Jie Tian, Hailan Kuang, Xiaolin Ma
Summary: This paper proposes a multi-camera stereo calibration method based on circular calibration plate, focusing on the extraction of pattern features during the calibration process. Experimental results show that this method has better calibration effect and accuracy compared to traditional calibration methods, with an average reduction of over 0.006 pixels in the multi-camera reprojection error.
Article
Chemistry, Multidisciplinary
Hailan Kuang, Kewen Wu, Xiaolin Ma, Xinhua Liu
Summary: This article discusses a method for Chinese grammatical error correction (GEC) based on iterative training and sequence tagging (CGEC-IT). The method dynamically generates target tags for each round and utilizes conditional random fields to improve overall labeling results. Experimental results show that the proposed method outperforms previous work by up to 2% on the F0.5 score, demonstrating the effectiveness of iterative training in the Chinese GEC model.
APPLIED SCIENCES-BASEL
(2022)