Article
Geochemistry & Geophysics
Kecheng Chen, Xiaorong Pu, Yazhou Ren, Hang Qiu, Fanqiang Lin, Saimin Zhang
Summary: This article proposes a novel denoising framework for TEM signals using deep convolutional neural networks, which transforms the denoising task into an image denoising task. The framework includes a new signal-to-image transformation method and a deep CNN-based denoiser, achieving better performance.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Joaquin Ruiz, Gaston Schlotthauer, Leandro Vignolo, Marcelo A. Colominas
Summary: In this work, a time-varying wave-shape extraction algorithm based on a modified version of the adaptive non-harmonic model is proposed. The algorithm can accurately recover the time-varying wave-shape of non-stationary signals in various tasks.
Article
Engineering, Biomedical
Egle Butkeviciute, Liepa Bikulciene, Tomas Blazauskas
Summary: This study investigates the application of continuous non-invasive bio-signal recordings in daily life activities using smart devices and cloud-based technologies. A new ECG feature extraction algorithm for movement-contaminated signals was proposed and validated through comparisons with other methods. This research is significant for real-time data processing and health monitoring.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2022)
Article
Instruments & Instrumentation
Hai Liu, Qing An, Tingting Liu, Zhenghua Huang, Qian Deng
Summary: In this article, a novel infrared image destriping model with unidirectional gradient and sparsity constraint is proposed. Experimental results demonstrate that the model can effectively suppress stripe noises and preserve image structure, making it promising for visual detection on biomedical images.
INFRARED PHYSICS & TECHNOLOGY
(2022)
Article
Engineering, Electrical & Electronic
Joaquin Ruiz, Marcelo A. Colominas
Summary: The adaptive non-harmonic (ANH) model is a powerful tool for compactly representing oscillating signals with time-varying amplitude and phase, as well as non-sinusoidal oscillatory morphology. This paper addresses the problem of estimating the number of harmonic components of the waveform function in non-stationary signals and applies it to denoising and waveform estimation tasks. Experimental results demonstrate that the model can adaptively estimate the waveform of non-stationary signals and performs competitively in the presence of noise.
Article
Engineering, Electrical & Electronic
Baorui Dai, Gaetan Frusque, Qi Li, Olga Fink
Summary: Acoustic monitoring has great potential in infrastructure condition diagnosis. We propose an acceleration-guided acoustic signal denoising framework (AG-ASDF) based on learnable wavelet transform to extract relevant features and improve classification accuracy. A comparative study shows that AG-ASDF outperforms other methods in detecting slab track conditions.
IEEE SENSORS JOURNAL
(2022)
Article
Energy & Fuels
Noman Shabbir, Lauri Kutt, Bilal Asad, Muhammad Jawad, Muhammad Naveed Iqbal, Kamran Daniel
Summary: This article discusses research on power quality issues in modern power systems, focusing on improving power factor to mitigate the adverse effects of inductive loads on the system. Through the analysis of real-time data from a frequency converter, a hybrid solution based on wavelet transform and Fourier transform is proposed for diagnosing the causes of motor failure in ventilation systems.
Article
Computer Science, Artificial Intelligence
Chunwei Tian, Menghua Zheng, Wangmeng Zuo, Bob Zhang, Yanning Zhang, David Zhang
Summary: This paper proposes a multi-stage image denoising CNN with wavelet transform, using dynamic convolution, wavelet transform and enhancement, and residual block to improve denoising performance. Experimental results show that the proposed method outperforms popular denoising methods.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Shuang Xu, Jiangshe Zhang, Jialin Wang, Kai Sun, Chunxia Zhang, Junmin Liu, Junying Hu
Summary: This paper proposes a more interpretable network for guided image denoising, by building an observation model and using deep prior regularized optimization problem to design the network architecture. Experimental results show that the network outperforms several state-of-the-art methods in terms of performance.
INFORMATION FUSION
(2022)
Article
Biochemical Research Methods
Sean Fitzgerald, Eric Marple, Anita Mahadevan-Jansen
Summary: We propose a methodology for evaluating probe-based Raman spectroscopy systems for biomedical analysis. The method uses a biological standard sample and data analysis approach to accurately measure and compare signal quality. Dairy milk is selected as the standard due to its similarity to tissue properties and homogeneous nature. The results show that our method produces an experimental signal-to-noise ratio (SNR) that aligns with the theoretical value and reveals significant differences in the capabilities of different spectrographs to detect biological Raman spectra.
BIOMEDICAL OPTICS EXPRESS
(2023)
Article
Engineering, Electrical & Electronic
Sahar Sadrizadeh, Nematollah Zarmehi, Ehsan Asadi Kangarshahi, Hamidreza Abin, Farokh Marvasti
Summary: The paper presents a low complexity algorithm for reconstructing signals corrupted by noise, with superior reconstruction quality compared to other state-of-the-art methods. The algorithm is versatile, effective for different types of noise, and has been validated through experimental applications in various scenarios.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Telecommunications
Saeid Yazdanpanah, Saman Shojae Chaeikar, Alireza Jolfaei
Summary: The COVID-19 pandemic presents new challenges for the healthcare industry, particularly regarding the exposure of hospital staff to the virus. In response, a hospital in New York implemented an audio-based communication system to protect nurses during the 2014 Ebola epidemic, which later evolved into an IoT healthcare solution for remote patient communication. However, this technology has also attracted criminals who use it for covert communication. Current steganalysis practices are not efficient enough for speech content, but this research proposes a new feature, PEAS, which can effectively discriminate between stego speech samples and clean ones with high sensitivity.
DIGITAL COMMUNICATIONS AND NETWORKS
(2023)
Article
Engineering, Biomedical
Emma Farago, Adrian D. C. Chan
Summary: This paper examines three methods for synthesizing motion artifact data and finds that the RNN method is the most promising in terms of synthesizing diverse motion artifact data that resembles the properties of experimental data.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2021)
Article
Engineering, Mechanical
Nitin Nagesh Kulkarni, Nicholas A. Valente, Alessandro Sabato
Summary: This study proposes a robust time-inferred autoencoder (TIA) framework for denoising optical data while preserving the system's spatio-temporal characteristics. Compared to existing methods, TIA achieves better accuracy in extracting spatio-temporal characteristics and offers more flexibility and automation than total variation denoising.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Review
Acoustics
Janek Groehl, Melanie Schellenberg, Kris Dreher, Lena Maier-Hein
Summary: Photoacoustic imaging (PAI) is a promising emerging imaging modality that enables spatially resolved imaging of optical tissue properties. However, extracting relevant tissue parameters from raw data is difficult, leading to the application of deep learning methods in PAI. These methods have advantages like fast computation times and adaptability to various problems, facilitating clinical translation.
Article
Energy & Fuels
Apostolos Vavouris, Benjamin Garside, Lina Stankovic, Vladimir Stankovic
Summary: The paper proposes a scalable methodology for detecting household EV charging events and load consumption from smart meter data, emphasizing the generalisability of the approaches across similar houses and different geographical regions and EV charging profiles. The effectiveness of different performance and generalisation loss metrics is also discussed.
Article
Energy & Fuels
Dandan Li, Jiangfeng Li, Xin Zeng, Vladimir Stankovic, Lina Stankovic, Changjiang Xiao, Qingjiang Shi
Summary: This study explores the importance of reducing carbon emissions in buildings to overall greenhouse gas emissions reductions and proposes a method for estimating appliance power consumption based on transfer learning. By utilizing a one-to-many model and incorporating appliance transfer learning and cross-domain transfer learning, this approach effectively estimates the power consumption of all appliances.
Article
Chemistry, Analytical
Jiangfeng Li, Lina Stankovic, Vladimir Stankovic, Stella Pytharouli, Cheng Yang, Qingjiang Shi
Summary: This paper proposes a novel multi-channel event-detection scheme based on Neyman-Pearson lemma and Multi-channel Coherency Migration (MCM) on the stacked signal across multi-channels to improve the classification of seismic signals. It also uses graph-based feature weight optimization as feature selection to enhance signal classification. Experimental results show that this method can identify 614 more seismic events compared to traditional detection approaches, and feature selection provides more focused feature sets while improving the classification performance.
Review
Chemistry, Analytical
Dong Han, Beni Mulyana, Vladimir Stankovic, Samuel Cheng
Summary: This review provides an overview of recent advances in deep reinforcement learning algorithms for robotic manipulation tasks. It covers the fundamental ideas of reinforcement learning and the parts of a reinforcement learning system. Various deep reinforcement learning algorithms, including value-based methods, policy-based methods, and actor-critic approaches, are discussed. The review also examines the challenges and solutions when applying these algorithms to robotics tasks, and highlights unsolved research issues and future directions for the subject.
Article
Chemistry, Analytical
Rachel Stephen Mollel, Lina Stankovic, Vladimir Stankovic
Summary: With the global roll-out of smart metering, the potential of higher resolution energy readings is being tapped into for accurate billing, improved demand response, and better-tuned tariffs. This paper focuses on the trustworthiness of the NILM model and proposes a naturally interpretable decision tree approach, as well as explainability tools, to improve appliance classification performance and feature selection. Experimental results show significant improvements in the classification performance of appliances like the toaster, dishwasher, and washing machine.
Article
Environmental Sciences
Apostolos Parasyris, Lina Stankovic, Vladimir Stankovic
Summary: In recent years, deep learning has been used in seismic inversion to analyze the subsurface, such as improving velocity model building. This study focuses on deep-learning-based inversion for velocity model building, using a conditional generative adversarial network (PIX2PIX) with ResNet-9 as the generator. Mathematical methodology is also used to generate samples of multi-stratified heterogeneous velocity models for training the architecture. The results show that the proposed architecture achieves state-of-the-art performance in reconstructing velocity models using only one seismic shot, reducing cost and computational complexity. The solution is also shown to be applicable to linear multi-layer models, curved or folded structures, structures with salt bodies, and higher-resolution structures built from geological images through quantitative and qualitative evaluation.
Article
Energy & Fuels
Tamara Todic, Vladimir Stankovic, Lina Stankovic
Summary: With the widespread deployment of smart meters, Non-Intrusive Load Monitoring (NILM) has emerged as a promising application for informing energy management within buildings. However, existing deep learning NILM models have limitations in terms of flexibility and scalability. In this study, an active learning framework is proposed to improve transferability and reduce the cost of labeling, achieving optimal accuracy-labelling effort trade-off. The results demonstrate the potential of this approach in improving the performance of NILM models and reducing computational resources needed.
Article
Chemistry, Analytical
Alexander Jamieson, Laura Murray, Vladimir Stankovic, Lina Stankovic, Arjan Buis
Summary: This study performs an unsupervised cluster analysis on the activities of individuals with lower limb amputation (ILLAs) and individuals without gait impairment in free-living conditions. The use of a thigh-based accelerometer allows for the clustering of physical activity data, with tSNE showing the best performance. The analysis identifies clusters in walking-based activities, specifically ground walking and stair ambulation.
Article
Geochemistry & Geophysics
Jiangfeng Li, Minxiang Ye, Lina Stankovic, Vladimir Stankovic, Stella Pytharouli
Summary: This study proposes a multitask learning scheme that utilizes physical characteristics of seismic wave propagation and a 3-D P-wave velocity model for signal representation learning. The results show that this approach outperforms state-of-the-art methods.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Geochemistry & Geophysics
Jiaxin Jiang, Vladimir Stankovic, Lina Stankovic, Emmanouil Parastatidis, Stella Pytharouli
Summary: Passive seismics are used to understand underground processes and predict their effects. Manual detection of seismic events is time-consuming and prone to inconsistency, so an automated approach based on convolutional neural networks (CNN) is proposed. Three different CNN architectures are evaluated using continuous seismometer recordings from a landslide area, showing excellent performance. The proposed network can also detect earthquake events in a different seismic area, demonstrating its potential to replace manual labeling.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Automation & Control Systems
Giulia Tanoni, Lina Stankovic, Vladimir Stankovic, Stefano Squartini, Emanuele Principi
Summary: This article proposes a knowledge distillation approach for NILM, aiming to reduce model complexity and improve generalization on unseen data domains. Experimental results show that the approach outperforms benchmark methods in multilabel appliance classification.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Proceedings Paper
Construction & Building Technology
Apostolos Vavouris, Lina Stankovic, Vladimir Stankovic, Jiufeng Shi
Summary: This paper explores how three-phase metering can benefit nonintrusive load monitoring (NILM), especially for appliances that are difficult to disaggregate and not widely reported. The emphasis is on the performance and loss introduced when using different levels of granularity of low-frequency data for load disaggregation.
PROCEEDINGS OF THE 2022 THE 9TH ACM INTERNATIONAL CONFERENCE ON SYSTEMS FOR ENERGY-EFFICIENT BUILDINGS, CITIES, AND TRANSPORTATION, BUILDSYS 2022
(2022)
Proceedings Paper
Construction & Building Technology
Rachel Stephen Mollel, Lina Stankovic, Vladimir Stankovic
Summary: This paper demonstrates the importance of using explainability tools to explain the outcomes of a decision tree multi-classification approach for NILM, as well as how model explainability informs feature selection and ultimately improves performance.
PROCEEDINGS OF THE 2022 THE 9TH ACM INTERNATIONAL CONFERENCE ON SYSTEMS FOR ENERGY-EFFICIENT BUILDINGS, CITIES, AND TRANSPORTATION, BUILDSYS 2022
(2022)
Article
Engineering, Electrical & Electronic
Liyan Chen, Samuel Cheng, Vladimir Stankovic, Lina Stankovic, Qingjiang Shi
Summary: This article discusses the shift-enabled property of random network graphs, finding that the considered unweighted connected random network graphs are shift-enabled with high probability when the number of edges is moderately high, while very dense and fully connected graphs are not. Weighted connected graphs are usually shift-enabled, unless the number of edges is very low.
IEEE TRANSACTIONS ON SIGNAL AND INFORMATION PROCESSING OVER NETWORKS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Tamara Todic, Lina Stankovic, Vladimir Stankovic, Jiufeng Shi
Summary: In this research, a deep learning-based load disaggregation approach was used to quantify individual energy consumption of milk production-related devices in dairy farms. The experiments conducted in three dairy farms in Germany showed that load disaggregation from a single aggregate meter is a viable and cost-effective alternative to accurately measure electricity consumption and provide decision support.
2022 IEEE INTERNATIONAL CONFERENCE ON SMART COMPUTING (SMARTCOMP 2022)
(2022)
