Article
Computer Science, Artificial Intelligence
Zixing Zhang, Ding Liu, Jing Han, Kun Qian, Bjorn W. Schuller
Summary: This article introduces an unsupervised learning framework to learn vector representations of audio sequences for acoustic event classification. Experimental results show that this method significantly outperforms other state-of-the-art hand-crafted sequence features in AEC.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Jary Pomponi, Simone Scardapane, Aurelio Uncini
Summary: Bayesian Neural Networks (BNNs) optimize an entire distribution over their weights, providing advantages in interpretability, multi-task learning, and calibration. This paper proposes an optimized version of training BNNs by replacing the Kullback-Leibler divergence in the ELBO term with a Maximum Mean Discrepancy (MMD) estimator, showing empirical advantages over the state-of-the-art methods.
Article
Computer Science, Information Systems
Xinyue Dong, Tingjin Luo, Ruidong Fan, Wenzhang Zhuge, Chenping Hou
Summary: Label distribution learning (LDL) is a new learning paradigm that tackles label ambiguity. We propose an active label distribution learning via kernel maximum mean discrepancy (ALDL-kMMD) method to address the high annotation cost in LDL. ALDL-kMMD captures structural information and reduces the queried unlabeled instances through a nonlinear model and marginal probability distribution matching. The effectiveness of our method is validated through experiments on real-world datasets.
FRONTIERS OF COMPUTER SCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Valentina Sanguineti, Pietro Morerio, Alessio Del Bue, Vittorio Murino
Summary: This paper investigates the generation of acoustic images using off-the-shelf cameras and a single microphone, and explores their application in audio-visual scene understanding. Three different model architectures are proposed and evaluated through various downstream tasks to assess the quality of the generated acoustic images.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Information Systems
Iti Chaturvedi, Tim Noel, Ranjan Satapathy
Summary: This paper presents a novel algorithm for classifying sentiments in speech under environmental noise. By leveraging the vector space of emotional concepts and a new metric based on eigenvalues, it achieves better classification results in sentiment analysis of YouTube videos. Additionally, the model is also capable of identifying bird species from audio recordings in urban areas.
Article
Chemistry, Physical
Fuping Guo, Wei Li, Peng Jiang, Falin Chen, Yinghonglin Liu
Summary: This paper investigates the application of acoustic emission technique for damage detection and classification in carbon fiber-reinforced composites. By combining deep learning approach using InceptionTime model with acoustic emission data, high accuracy damage classification can be achieved, showing potential in handling data imbalances.
Article
Mathematics
Qihang Huang, Yulin He, Zhexue Huang
Summary: This paper proposes a maximum mean discrepancy-based self-supervised learning algorithm that iteratively refines a classifier using highly confident unlabeled samples. Experimental results demonstrate that the proposed algorithm provides better testing accuracy and kappa values compared to other self-training and co-training algorithms.
Article
Computer Science, Information Systems
Hei-Chia Wang, Sheng-Wei Syu, Papis Wongchaisuwat
Summary: This paper discusses the issue of information overload caused by an abundance of digital music and proposes a music autotagging system to address the shortage of tags. Research shows that a multitask learning classifier combining audio and lyric information outperforms a single-task learning method using just audio data.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yiming Huang, Hang Lei, Xiaoyu Li, Guowu Yang
Summary: The study introduces a quantum maximum mean discrepancy (qMMD) metric and designs a quantum generative adversarial model qMMD-GAN based on it, implemented under a hybrid quantum-classical approach. By demonstrating its application in generating unknown quantum states on quantum devices, the competitive performance of qMMD-GAN is shown through numerical experiments. This research not only provides a new direction for improving classical data processing tasks but also contributes to the field of physics research.
Article
Computer Science, Information Systems
Na Wang, Yunxia Liu, Liang Ma, Yang Yang, Hongjun Wang
Summary: Automatic modulation classification is important in various military and civilian applications, and deep learning methods have achieved remarkable success. However, few methods can effectively generalize across changes in channel conditions and signal parameters. In this paper, a method is proposed to simultaneously achieve good classification accuracy on annotated source data and unlabeled signals with varying symbol rates and sampling frequencies. Experimental results show that it outperforms state-of-the-art methods in terms of accuracy on both source and target datasets.
Article
Computer Science, Artificial Intelligence
Michele Scarpiniti, Francesco Colasante, Simone Di Tanna, Marco Ciancia, Yong-Cheol Lee, Aurelio Uncini
Summary: This paper proposes a Deep Belief Network-based approach for audio signal classification to enhance work activity identification and remote surveillance of construction projects. By testing and validating on ten classes of construction equipment and tools, the effectiveness of the approach is demonstrated with a high accuracy of up to 98% on the test set.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Automation & Control Systems
Yibin Li, Yan Song, Lei Jia, Shengyao Gao, Qiqiang Li, Meikang Qiu
Summary: The article proposes an intelligent fault diagnosis method based on an improved domain adaptation method, utilizing training feature extractors and ensemble learning for industrial equipment health monitoring, effectively addressing domain mismatch issues.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Computer Science, Artificial Intelligence
Asif Iqbal Middya, Baibhav Nag, Sarbani Roy
Summary: This research work explores model-level fusion to find the optimal multimodal model for emotion recognition using audio and video modalities. The proposed models achieve high predictive accuracies on benchmark datasets and are shown to be effective compared to existing emotion recognition models.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Chemistry, Analytical
Minh-Khue Ha, Thien-Luan Phan, Duc Hoang Ha Nguyen, Nguyen Hoang Quan, Ngoc-Quan Ha-Phan, Congo Tak Shing Ching, Nguyen Van Hieu
Summary: This paper discusses an innovative approach in integrating radar technology and machine learning for effective surveillance systems. The approach involves signal acquisition, signal processing, and feature-based classification, using techniques such as short-time Fourier transform (STFT), mel spectrogram, mel frequency cepstral coefficients (MFCCs), and a simplified 2D convolutional neural network architecture. Experimental results show that the 2D CNN model trained on MFCC features outperforms other methods in object classification.
Article
Computer Science, Information Systems
Alireza Amirshahi, Anthony Thomas, Amir Aminifar, Tajana Rosing, David Atienza
Summary: Recent years have witnessed an increase in using deep learning models to monitor epilepsy patients based on electroencephalographic (EEG) signals. However, these approaches often struggle to generalize well beyond the data they were trained on and manual labeling of EEG signals is a time-consuming process. In this study, the Maximum-Mean-Discrepancy Decoder (M2D2) is introduced for automatic localization and labeling of seizures in EEG recordings, showing significantly better generalization performance compared to other state-of-the-art deep learning approaches when evaluated on data from a different clinical setting.
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS
(2023)
Article
Engineering, Multidisciplinary
Qiu-Hu Zhang, Yi-Qing Ni
Summary: This article proposes a new statistical decision philosophy for structural damage detection. It achieves damage detection by deriving the posterior probability of damage presence from Bayesian testing. Two principles are defined to achieve an optimal trade-off between false positive and false negative risks.
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2023)
Article
Chemistry, Multidisciplinary
Da-Zhi Dang, Chun-Cheung Lai, Yi-Qing Ni, Qi Zhao, Boyang Su, Qi-Fan Zhou
Summary: In this paper, an efficient and robust image classification model is proposed for railway status identification using ultrasonic guided waves (UGWs). Experimental studies using a hybrid sensing system consisting of a PZT actuator and FBG sensors are conducted. Comparative studies evaluate the performance of the UGW signals obtained by FBG sensors and AE sensors. The proposed image classifier achieves high accuracy and has the potential to be applied to mass railway monitoring systems in the future.
APPLIED SCIENCES-BASEL
(2023)
Article
Engineering, Civil
Bei-Yang Zhang, Yi-Qing Ni
Summary: A data-driven optimal sensor placement strategy is proposed in this study, which aims at accurately reconstructing mode shapes for vibration-based structural damage detection by temporarily deploying a few vibration sensors on a target bridge. This strategy is also applicable for the upgrade of a long-term structural health monitoring system by using historical data collected from the current system.
ENGINEERING STRUCTURES
(2023)
Article
Engineering, Multidisciplinary
Bei-Yang Zhang, Yi-Qing Ni
Summary: This paper proposes a novel adaptive modelling framework for sparse polynomial chaos expansion. It automatically determines the truncation degree and training sample set, and alleviates the curse of dimensionality issue in polynomial chaos expansion. The framework incorporates an adaptive basis selection strategy, a sequential sampling strategy and a sparse representation method to improve the precision and convergence rate of the model.
APPLIED MATHEMATICAL MODELLING
(2023)
Article
Engineering, Civil
Zheng-Wei Chen, Guang-Zhi Zeng, Yi-Qing Ni, Tang-Hong Liu, Ji-Qiang Niu, Hua-Dong Yao
Summary: This study investigates the reduction of aerodynamic drag in high-speed trains by proposing an air blowing configuration on the head and tail cars. The analysis of aerodynamic drag and slipstream velocity under different blowing velocities reveals that blowing speeds of Ub = 0.05Ut, 0.10Ut, 0.15Ut, and 0.20Ut cause reductions in total drag coefficient (Cd) by 5.81%, 10.78%, 13.70%, and 15.43% compared to the without-blowing case. The blowing measure creates an air gap between the flow and train surface, resulting in a reduction of viscous and pressure drag. Considering blowing cost, efficiency, and flow structure evolution, a blowing speed of Ub = 0.10Ut is recommended. This speed leads to reduction ratios of 9.18%, 12.77%, 10.90%, and 10.78% for the head, middle, tail car, and total train's aerodynamic drag, respectively.
JOURNAL OF WIND ENGINEERING AND INDUSTRIAL AERODYNAMICS
(2023)
Article
Engineering, Multidisciplinary
Yun-Ke Luo, Li-Zhong Song, Chao Zhang, Yi-Qing Ni
Summary: This research evaluates the efficiency of different noise mitigation measures on an elevated railway through in-situ measurements. Experimental results show that the rubber floating slab track can mitigate bridge-borne noise by 0-4 dB SPL, while the track acoustic absorber and track-side noise barrier combination can boost the insertion losses of SPL by 2-7 dB(A). The combined control strategy demonstrates better performance within the efficient noise reduction regions. The findings of this study can provide guidance for the design of noise control strategies for elevated railways.
Article
Engineering, Multidisciplinary
Cai Yi, Le Ran, Jiayin Tang, Qiuyang Zhou, Lu Zhou
Summary: In this paper, a multi-period pulse detection indicator called harmonic spectrum correlation kurtosis (HSCK) is proposed to effectively extract multi-fault features from compound fault signals of rotating machinery. A novel adaptive matching extraction strategy is introduced, which includes variational mode decomposition, adaptive plane paving method, and enhanced cyclic frequency estimation. The simulation results demonstrate the accuracy and effectiveness of this strategy for compound fault diagnosis.
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2023)
Article
Construction & Building Technology
Wei Liu, Yi-Qing Ni, Kohju Ikago, Wai Kei Ao
Summary: Rate-independent linear damping (RILD) is an effective option for addressing excessive displacement challenges in low-frequency structures. However, the physical realization of RILD remains challenging. This study proposes two models to promote progress in the application of RILD and examines their effectiveness in seismic control of low-frequency structures.
JOURNAL OF BUILDING ENGINEERING
(2023)
Article
Mathematics
Su-Mei Wang, You-Wu Wang, Yi-Qing Ni, Yang Lu
Summary: This study aims to develop a track-side online monitoring system for malfunction detection in the suspension controllers of maglev trains during their in-service operation. The proposed system utilizes accelerometers and a data acquisition unit to detect malfunctions and establish a dynamic model.
Article
Engineering, Mechanical
Ziquan Yan, Xiangyun Deng, Yi-Qing Ni, Linlin Sun
Summary: This study proposes an explicit finite element method to investigate the elastic layer effects in wheel-rail rolling contact. The method is validated by comparing it with Kalker's boundary element method. The results show that a harder layer introduces larger contact pressure, surface shear stress, and subsurface stress.
Article
Construction & Building Technology
Yuxuan Liang, Xiaomin Dong, Wai Kei Ao, Yi-Qing Ni
Summary: A study of magnetorheological seat suspension controlled by a novel tuning control strategy is conducted to reduce vibrations and avoid end-stop impacts. Both simulation and experiment results prove that the proposed control strategy shows good performances on vibration attenuation and end-stop impact reduction.
STRUCTURAL CONTROL & HEALTH MONITORING
(2023)
Article
Engineering, Civil
E. Deng, Xin-Yuan Liu, Huan Yue, Wei-Chao Yang, De-Hui Ouyang, Yi-Qing Ni
Summary: This study examines the impact of dunes next to desert urban motorways on the driving safety of sedans and proposes a cost-effective simulation scheme.
JOURNAL OF WIND ENGINEERING AND INDUSTRIAL AERODYNAMICS
(2023)
Article
Engineering, Mechanical
Chen Wang, Chenxi Wang, Youhong Ji, Gaolei Li, Gui-Lin Wen, Yi-Qing Ni, Siu-Kai Lai
Summary: This paper presents a high-performance tri-hybrid vibration-driven generator with quin-stability and speed amplification characteristics. It integrates three different mechanisms to enhance power density and exploits the quin-stable nonlinear behavior to create a shallow potential well. Experimental results demonstrate the exceptional performance of this design.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Engineering, Electrical & Electronic
Gao-Feng Jiang, Su-Mei Wang, Yi-Qing Ni, Wen-Qiang Liu
Summary: This article proposes an unsupervised domain adaptation (DA) approach to diagnose the health conditions of maglev rail joints in complex operational conditions, which can successfully identify the conditions even when the operation mode changes.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Automation & Control Systems
Feng Wang, Jiaqi Xia, Xiaoyuan Zhu, Xing Xu, Yi-Qing Ni
Summary: This article presents a real-time predictive energy management strategy with mode transition frequency constraints to improve the energy efficiency of plug-in hybrid electric vehicles. It ensures low mode transition frequency and high calculation efficiency.
IEEE-ASME TRANSACTIONS ON MECHATRONICS
(2023)