Article
Acoustics
Fei Gao, Bing Li, Lei Chen, Zhongyu Shang, Xiang Wei, Chen He
Summary: An optimized softmax classifier based on the traditional softmax classifier is proposed, and a convolution neural network (CNN) framework is built to accurately classify signals with similar curves. Through a comparative experiment, the classifier's performance is evaluated in terms of loss curve decline rate, classification accuracy, and feature visualization, showing high classification accuracy and strong robustness.
Article
Computer Science, Artificial Intelligence
Karnati Mohan, Ayan Seal, Ondrej Krejcar, Anis Yazidi
Summary: Automatic facial expression recognition is a challenging task in computer vision with applications in human-computer interaction and behavioral psychology. Previous works have focused on handcrafted features, but accurately extracting all relevant features is difficult due to emotional state variations. This study proposes FER-net, a convolutional neural network, and demonstrates its superiority over twenty-one state-of-the-art methods on multiple benchmark datasets by achieving high accuracy rates for distinguishing seven facial expressions.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Mathematics
Yuhua Ding, Zhenmin Tang, Fei Wang
Summary: This paper presents a single-sample face recognition method based on a shared generative adversarial network. By generating and expanding the gallery dataset, and utilizing a deep convolutional neural network for feature extraction and classifier training, it can effectively recognize single-sample faces.
Article
Engineering, Aerospace
Ali K. Abed, Rami Qahwaji, Ahmed Abed
Summary: In recent years, there has been a growing interest in near-real-time solar data processing for space weather applications. This study utilizes deep learning to establish an automated system for short-term solar flare forecasts, achieving promising results in predicting solar activity within the next 24 hours.
ADVANCES IN SPACE RESEARCH
(2021)
Article
Computer Science, Artificial Intelligence
Arati Kushwaha, Ashish Khare, Om Prakash
Summary: This paper proposes a simple and computationally efficient deep convolutional neural network (DCNN) architecture for human activity recognition based on multiscale processing. By increasing the width and depth of the network, the proposed architecture improves the utilization of computational resources and is flexible for both small and large datasets.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Engineering, Multidisciplinary
Junxiang Wang, Changshu Zhan, Di Yu, Qiancheng Zhao, Zhijie Xie
Summary: This paper proposes a method based on a stacked sparse autoencoder combined with a softmax classifier for fault diagnosis of rolling bearings. The method extracts frequency-domain features of vibration signals using a stacked sparse autoencoder and utilizes an improved K-fold cross-validation to obtain pre-train set, train set, and test set. The performance of the model is evaluated based on accuracy, macro-precision, macro-recall, and macro-F1 score. The proposed model is validated with high accuracy using data from Case Western Reserve University and XJTU-SY.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Brijnesh Jain
Summary: This article explores the combination of linear models and dynamic time warping for time series classification. The theoretical results, including the Representation Theorem and the Matrix Complexity Lemma, highlight the characteristics of warped-linear models. Empirical findings show that using dtw-score to replace the inner product of linear models can significantly enhance predictive performance.
KNOWLEDGE AND INFORMATION SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Rihem Mahmoud, Selma Belgacem, Mohamed Nazih Omri
Summary: Behavior recognition in video sequences is an interesting task in computer vision. This paper proposes a novel deep architecture for isolated actions and large scale continuous hand gesture recognition. By extracting gesture characteristics and analyzing each sequence, the proposed approach achieves high accuracy and precision in recognition.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Information Systems
V. Betcy Thanga Shoba, I. Shatheesh Sam
Summary: Age estimation is a challenging process influenced by various factors. To improve the accuracy of aging facial recognition and reduce computational time, the ADRBSRC method is proposed. It preprocesses images using adaptive bilateral filtering, extracts important features using ADRFL, and classifies age groups using BBSRC.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Environmental Sciences
Ruida Ye, Yuan Ren, Xiangyang Zhu, Yujing Wang, Mingyue Liu, Lifen Wang
Summary: Non-cooperative space object pose estimation is a key technique for spatial on-orbit servicing. This paper proposes an end-to-end learning algorithm based on dual-channel transformer for non-cooperative space object pose estimation. By directly detecting satellite pose information using RGB images, the algorithm significantly improves recognition efficiency.
Article
Computer Science, Software Engineering
Arati Kushwaha, Prashant Srivastava, Ashish Khare
Summary: Human activity recognition (HAR) is in high demand for automated monitoring applications, however, developing an efficient algorithm faces various challenges. This study proposes a simple and computationally efficient deep CNN architecture using multi-layer information fusion to improve HAR performance, and extensive experiments on publicly available datasets demonstrate its superiority over other existing methods.
CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE
(2023)
Article
Computer Science, Information Systems
Lin Yan, Mingyong Zeng, Shuai Ren, Zhangkai Luo
Summary: Encrypted traffic identification using a deep residual convolution network, which fuses a Softmax classifier with its angular variant for better discrimination, improves representation learning and achieves excellent results on a public dataset.
IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS
(2021)
Article
Plant Sciences
Ping Li, Rongzhi Jing, Xiaoli Shi
Summary: A method for recognizing apple diseases based on modified convolutional neural networks (MCNN) is proposed in this study. By introducing Inception network, employing global average pooling and using modified Softmax classifier, the recognition performance is improved. The feasibility of the algorithm is demonstrated through experiments on apple disease image datasets.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Computer Science, Information Systems
J. S. Bindhu, K. Pramod
Summary: This paper introduces a pixel and texture based classification method using cellular automata algorithm, which improves the classification accuracy by incorporating the classification techniques. The method is applied to identify land cover and land use, and the results show that the texture-based classification achieves higher accuracy than the pixel-based classification.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Yuan Zhang, Yonggang Zhang, Lele Peng, Lianghua Quan, Shubin Zheng, Zhonghai Lu, Hui Chen
Summary: The softmax function is widely used in deep neural networks and its hardware performance is crucial for the training and inference of DNN accelerators. Traditional softmax functions are complex and existing hardware architectures are resource-consuming or have low precision. In this study, we propose a base-2 softmax function that shows feasibility and good accuracy in neural network training. We also design a low-complexity and high-precision architecture using the symmetric-mapping lookup table method, achieving the best performance and efficiency compared to the latest works.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2022)
Article
Computer Science, Artificial Intelligence
Yan Zhang, Yifei Li, Youyong Kong, Jiasong Wu, Jian Yang, Huazhong Shu, Gouenou Coatrieux
Summary: This paper proposes a graph self-construction and fusion network (GSCFN) for semi-supervised brain tissue segmentation in MRI by fusing multiple types of image features. Experimental results demonstrate the superiority of the proposed method compared to approaches based on a single feature type.
Article
Physiology
Henry Areiza-Laverde, Cindy Dopierala, Lotfi Senhadji, Francois Boucher, Pierre Y. Gumery, Alfredo Hernandez
Summary: The study evaluates the feasibility of acquiring ECG and ACC data from an implant in the gastric fundus and finds a high correlation with gold standard data. Results show the potential for chronic cardiovascular monitoring from an implantable cardiac device at the gastric location, with the main challenge being the optimization of the signal-to-noise ratio, especially for handling specific noise sources.
FRONTIERS IN PHYSIOLOGY
(2021)
Article
Computer Science, Artificial Intelligence
Jiasong Wu, Xiang Qiu, Jing Zhang, Fuzhi Wu, Youyong Kong, Guanyu Yang, Lotfi Senhadji, Huazhong Shu
Summary: The proposed Generative Fractional Scattering Networks (GFRSNs) use more expressive fractional wavelet scattering networks as the encoder and can generate better images on testing datasets compared to the original Generative Scattering Networks (GSNs).
FRONTIERS IN NEUROROBOTICS
(2021)
Article
Computer Science, Hardware & Architecture
Ismail Hadj Ahmed, Abdelghani Djebbari, Amar Kachenoura, Lotfi Senhadji
Summary: Heart rate variability signal is a valuable tool for cardiovascular system diagnostics, used to detect arrhythmia and abnormalities in the autonomic nervous system. Quadratic time-frequency analysis of the signal can help detect various cardiovascular pathologies. A client-server telemedical platform was developed for real-time remote monitoring of cardiovascular function in arrhythmia patients, allowing immediate interaction in case of alarms.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Zhijian Sun, Zhuhong Shao, Yuanyuan Shang, Bicao Li, Jiasong Wu, Hui Bi
Summary: This paper investigates a multi-stage convolutional neural network with predefined filters to capture nonlinear structures within data and more representational image features. The network consists of cascaded blocks of random Fourier mapping, two-dimensional principal component analysis, and activation operation. Experimental results show that the proposed network has significantly advantageous both in terms of accuracy and computational time compared to the existed algorithms.
MACHINE VISION AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Fuzhi Wu, Jiasong Wu, Youyong Kong, Chunfeng Yang, Guanyu Yang, Huazhong Shu, Guy Carrault, Lotfi Senhadji
Summary: This paper establishes the connection between CNNs and signal modulation, explains the forward and back-propagation processes of CNNs, and verifies that modulating the signal to the appropriate energy spectrum distribution can improve classification and segmentation accuracy.
Article
Computer Science, Artificial Intelligence
Youyong Kong, Jiaxing Li, Ke Zhang, Jiasong Wu
Summary: Data augmentation can enhance the generalization performance of neural networks, but it is challenging for graph data due to its irregular non-Euclidean structure. This paper introduces MSSA-Mixup, a novel graph data augmentation method that extends the training distribution through interpolating multi-scale graph representation with self-attention. MSSA-Mixup effectively improves the generalization ability of GNNs, as demonstrated by extensive experiments on benchmark datasets.
PATTERN RECOGNITION LETTERS
(2023)
Article
Acoustics
Jiasong Wu, Qingchun Li, Guanyu Yang, Lei Li, Lotfi Senhadji, Huazhong Shu
Summary: In traditional speech denoising tasks, clean audio signals are usually used as training targets, but they are difficult to obtain. To address this, an end-to-end self-supervised training scheme called Only-Noisy Training (ONT) is proposed. ONT generates training pairs solely from noisy audio signals and consists of two modules: training audio pairs generation and speech denoising. Experimental results show that ONT reduces dependence on clean targets and achieves comparable or better performance than other training strategies. Availability-ONT can be found at https://github.com/liqingchunnnn/Only-Noisy-Training.
SPEECH COMMUNICATION
(2023)
Proceedings Paper
Acoustics
Xinyu Yuan, Wenhan Wang, Youyong Kong, Jiasong Wu, Guanyu Yang, Huazhong Shu
Summary: Prediction of brain functional activity is crucial for neuroscience research. The proposed Temporal Cross-Graph Network (TCGN) can effectively capture multi-modal spatial dependence and temporal patterns for accurate prediction of brain functional activity.
2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2022)
Article
Computer Science, Artificial Intelligence
Yuting He, Rongjun Ge, Xiaoming Qi, Yang Chen, Jiasong Wu, Jean-Louis Coatrieux, Guanyu Yang, Shuo Li
Summary: This work proposes a novel framework for few-shot medical image segmentation based on the learning registration to learn segmentation paradigm. The framework addresses the limitations in authenticity, diversity, and robustness of existing frameworks, and achieves state-of-the-art performance in few-shot medical image segmentation tasks.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Junxiao Sun, Yan Zhang, Jian Zhu, Jiasong Wu, Youyong Kong
Summary: This paper presents a semi-supervised learning framework based on a novel multi-scale graph cut loss function, which significantly reduces the need for labeled data in practical medical image processing. Experiments show that the proposed method is more effective compared to the currently known state-of-the-art methods.
2021 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Yuting He, Rongjun Ge, Jiasong Wu, Jean-Louis Coatrieux, Huazhong Shu, Yang Chen, Guanyu Yang, Shuo Li
Summary: The study introduces a HiFe priori rule and corresponding network structure HiFeNet, enhancing thin structure semantic segmentation by extracting and fusing high-frequency components, demonstrating significant advantages in experiments.
2021 21ST IEEE INTERNATIONAL CONFERENCE ON DATA MINING (ICDM 2021)
(2021)
Article
Biology
Haiting Yan, Yue Wang, Jingrong Zhang, Xinru Cui, Jiasong Wu, Jie Zhou, Yuan Chen, Jia Lu, Ruiyang Guo, Maggie Ou, Hongxu Lai, Zhiming Yu
Summary: Cryo-scanning electron microscopy (cryo-SEM) is an advanced technique used for studying the structures of biological samples with high water content, providing authentic images without the need for sample pretreatment. The main steps include cryoprocessing and SEM observation, which can be applied to study various root systems.