Article
Computer Science, Artificial Intelligence
Xin Li, Fengrong Bi, Lipeng Zhang, Jiewei Lin, Xiaobo Bi, Xiao Yang
Summary: This paper proposes an accurate and efficient end-to-end fault detection model for rotating machinery based on small-scale training data. The model, called FCK-DESN, utilizes fixed convolution kernels for spatial feature extraction and pattern recognition, and incorporates time-frequency information for fault detection. Case studies demonstrate that the FCK-DESN approach achieves higher recognition rates, greater efficiency, and lower data size requirements compared to popular deep learning methods.
APPLIED SOFT COMPUTING
(2022)
Article
Computer Science, Information Systems
Yan Song, Yiming Yin, Panfeng Xu
Summary: In this paper, a novel model named ECA-CRNN is proposed for emotion recognition using EEG signals. The model integrates the ECA-Net module into a modified CNN and GRU, allowing for more comprehensive feature extraction and improved recognition performance. The results on the DEAP dataset show that the model achieves recognition accuracies of 95.70% and 95.33% for arousal and valence, respectively, surpassing most existing methods.
Article
Computer Science, Information Systems
Farzad Baradaran, Ali Farzan, Sebelan Danishvar, Sobhan Sheykhivand
Summary: In this research, a model based on Deep Convolutional Neural Networks (DCNNs) is presented, which can reliably classify three emotions: positive, negative, and neutral, from EEG signals using musical stimuli. The proposed model achieved an accuracy of 98% for the binary classification of positive and negative emotions, and 96% accuracy for the trinary classification of positive, neutral, and negative emotions.
Review
Biology
Mahboobeh Jafari, Afshin Shoeibi, Marjane Khodatars, Sara Bagherzadeh, Ahmad Shalbaf, David Lopez Garcia, Juan M. Gorriz, U. Rajendra Acharya
Summary: Emotions play a crucial role in daily life and are considered a significant factor in human interactions. EEG signals, with their high spatial resolution, have the potential to be a powerful tool for emotion recognition. However, there are challenges such as signal variability, individual differences, and feature selection. Deep learning techniques show promise in addressing these challenges.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Computer Science, Artificial Intelligence
Qi Liu, Hongguang Liu
Summary: The study uses deep learning and EEG signals to classify criminal psychological emotions, constructs a classification method based on neural networks, verifies the effectiveness of the algorithm, and shows practical effects.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Chemistry, Multidisciplinary
Nikesh Bajaj, Jesus Requena Carrion
Summary: This paper proposes an approach to interpret deep representations of EEG signals using spatio-spectral feature images (SSFIs), which encode the activation patterns of neurons in each layer of a DNN. The experimental results show that low-level CNN features focus on larger regions, while high-level features focus on smaller regions, and different patterns can be discerned in different frequency bands using SSFI.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Zhongli Bai, Zeyu Li, Zhiwei Li, Yu Song, Qiang Gao, Zemin Mao
Summary: This study investigates the use of brain-computer interface technology for emotion recognition. EEG signals collected from 15 hearing-impaired subjects under different emotions were analyzed using various feature extraction methods and attention mechanisms to improve accuracy. The cross-subject problem was addressed with domain adaptation methods, achieving high accuracies in both subject-dependent and cross-subject scenarios.
Article
Biology
Yihan Wu, Min Xia, Li Nie, Yangsong Zhang, Andong Fan
Summary: In recent years, there has been a growing interest in emotion recognition based on electroencephalography (EEG) in the brain-computer interaction (BCI) field. This study proposed a Multi-Scales Bi-hemispheric Asymmetric Model (MSBAM) based on a convolutional neural network (CNN) structure, taking into account both the activity differences in the left and right brain hemispheres and the nonstationarity of EEG signals. The model achieved over 99% accuracy for the two-class classification of low-level and high-level states in each of four emotional dimensions, demonstrating its potential for designing deep learning models for emotion recognition.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Chemistry, Analytical
Gin Chong Lee, Chu Kiong Loo
Summary: This work proposes a novel unsupervised self-organizing network, called SO-ConvESN, for human action recognition. The network utilizes Recurrent Plots and Recurrence Quantification Analysis techniques for explaining and tuning ESN hyperparameters, and is cascaded with a CNN for action recognition. Experimental results demonstrate competitive recognition accuracy.
Article
Geochemistry & Geophysics
Ting Lu, Mengkai Liu, Wei Fu, Xudong Kang
Summary: This study proposes a new band-grouping guided multi-attention module to enhance the performance of spectral-spatial feature learning. Spectral bands are adaptively divided into multiple nonoverlapping groups, reducing complexity. A multi-attention mechanism is embedded into CNNs to learn group-specific spectral-spatial features. A spectral-spatial classification network is built, integrating pixelwise and patchwise learning to boost performance.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Valentin Gabeff, Tomas Teijeiro, Marina Zapater, Leila Cammoun, Sylvain Rheims, Philippe Ryvlin, David Atienza
Summary: This study addressed the lack of interpretability of neural network models in medical decision support by developing a deep learning model from EEG signals for online detection of epileptic seizures and associating model behavior with expert medical knowledge. The focus was on aggregating classification results, identifying frequency patterns, and recognizing signal waveforms. Results showed that the kernel size in the first layer significantly impacts feature interpretability and model sensitivity.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2021)
Article
Neurosciences
Ming Gao, Jie Mao
Summary: The main clinical manifestations of stroke include motor, language, sensory, and mental disorders, with the possibility of resulting in sequelae such as numbness, facial paralysis, and central paralysis if not effectively treated. Effective rehabilitation training is crucial for stroke patients to reduce the disease and restore motor function, especially considering the prevalence of upper limb paralysis among those affected.
FRONTIERS IN NEUROSCIENCE
(2021)
Article
Acoustics
Mehmet Bilal Er, Harun Cig, Ibrahim Berkan Aydilek
Summary: This study discusses the recognition of human emotions while listening to music using EEG signals, proposing a new emotion recognition method based on deep learning. The results show that the method performs well in distinguishing different emotions.
Article
Computer Science, Hardware & Architecture
Suat Toraman
Summary: A novel method using one-dimensional capsule networks for preictal/interictal recognition in scalp electroencephalogram signals achieved the best classification accuracy. The results indicated that important information about preictal/interictal recognition can be found in the 30 minutes before the onset of seizures. The proposed method brings a new perspective to seizure prediction studies using capsule networks.
COMPUTERS & ELECTRICAL ENGINEERING
(2021)
Article
Engineering, Multidisciplinary
Kuiyu Chen, Lingzhi Zhu, Si Chen, Shuning Zhang, Huichang Zhao
Summary: The study introduces a novel radar intra-pulse modulation recognition method based on high-order spectrums, which uses automatic soft thresholding to improve learning effectiveness in the feature learning process. The method demonstrates excellent classification performance and robustness under low signal-to-noise ratios.