Deep learning has been widely applied to seismic signal classification, predominantly through supervised learning, typically relying on large labeled datasets. However, since the process of labeling large volumes of seismic data by domain experts is time-consuming and prone to human error, labeled seismic datasets are scarce. To address the problem of limited labeled data availability, a novel approach for seismic event classification is proposed employing self-supervised learning techniques. Initially, a generative-based self-supervised learning model, specifically an auto-encoder, is designed to extract informative features from the Short Time Fourier Transform of seismic recordings. These features are classified into four categories: earthquakes, micro-earthquakes, rockfalls, and anthropogenic noise. Classification is performed using (a) unsupervised K-means clustering on unlabeled data and (b) semi-supervised approaches, where only 5 to 33.3% of the data are labeled. The proposed semi-supervised method achieves high performance on a publicly available Résif dataset with recall of 0.90 for earthquakes, 0.65 for micro-earthquakes, 0.91 for rockfalls, and 0.84 for noise signals when trained with 20% of the labeled data. Additionally, we introduce a novel method to improve data labeling efficiency by using Self-Organizing Maps to cluster features from large datasets into multiple nodes. Our results demonstrate that the experts can more effectively and confidently label a small number of nodes instead of labeling all the events in the large dataset, thereby reducing the experts’ workload to just 4.6% of the original effort and our study reveals that this approach provides an excellent trade-off between expert labeling effort and classification accuracy, making it a highly effective solution for seismic event labeling. To evaluate the generalization capability of our proposed self-supervised learning model, we tested it on two unseen seismic datasets: the globally distributed Stanford Earthquake Dataset and the regionally focused Pacific Northwest Curated Seismic Dataset. On Stanford Earthquake Dataset, the pre-trained model effectively extracted discriminative earthquake and noise features, achieving high clustering accuracies. The Pacific Northwest Curated Seismic Dataset further challenges generalization with heterogeneous and previously unseen event types such as explosions, and thunder. Despite this diversity, the pre-trained model still preserved meaningful feature separability and captured inter-class relationships among acoustically similar events. Overall, these findings highlight the model’s ability to generalize effectively across both global and regional seismic datasets, underscoring its potential for wide deployment in seismological monitoring and event characterization without extensive retraining.

In the ongoing energy transition, characterized by increased reliance on distributed renewable sources and smart grid technologies, the need for advanced and trustworthy artificial intelligence (AI) in energy management systems is crucial. Non-intrusive load monitoring (NILM), a method for inferring individual appliance energy consumption from aggregate smart meter data, has gained prominence for enhancing energy efficiency. However, advanced deep neural network models used in NILM, while effective, raise transparency and trust concerns due to their complexity. This paper introduces a novel explainability-informed NILM training framework, specifically designed for low-frequency NILM. Our approach aligns with principles for trustworthy AI, focusing on human agency and oversight, technical robustness, and transparency, incorporating explainability directly into the training phase of a NILM model. We propose a novel iterative, explainability-informed NILM training algorithm that uses attribution priors to guide model optimization, including implementation and evaluation of the framework across multiple state-of-the-art NILM architectures, namely, convolutional, recurrent, and dilated causal layers. We introduce a novel Robustness-Trust metric to measure joint improvement in predictive and explainability performance, utilizing explainability metrics of faithfulness, robustness and effective complexity while analyzing model predictive performance against NILM-specific regression and classification metrics. Results broadly show that robust models achieve better explainability, while explainability-enhanced models can lead to improved model robustness. Together, our results demonstrate significant improvements in robustness and transparency of NILM systems across various appliances, model architectures, measurement scales, types of buildings, and energy usage patterns. This work paves the way for more transparent and trustworthy deployments in AI-driven energy systems.

The deployment of deep neural networks (DNNs) on resource-constrained edge devices necessitates efficient, low-complexity algorithms. Knowledge distillation (KD) addresses this through a student-teacher paradigm, transferring knowledge from complex teacher models to simpler student models. Current KD methods often optimize student performance without adequately addressing the reliability and interpretability of transferred knowledge, thus presenting challenges in maintaining both robustness and decision transparency. This paper introduces an Interpretability and Reliability-driven Knowledge Distillation (IR-KD) framework that enhances teacher model interpretability through perception-aligned gradients while leveraging hidden information from weak labels to optimize knowledge transfer. Our approach ensures compressed models remain computationally efficient while improving interpretability, which is essential for trustworthy edge AI deployment. We demonstrate improved predictive performance and model interpretability in non-intrusive load monitoring (NILM) applications as a case study. Quantitative explainability metrics confirm that perception-aligned gradients provide more faithful explanations, validating our approach’s effectiveness in developing reliable and transparent edge AI systems.

Deep learning has found extensive applications in classifying seismic signals in recent years. However, as a black box algorithm, deep learning is still rarely exploited in real-world applications, such as landslide monitoring. This is particularly a concern for geoscientists who prefer to classify seismic signals based on their physical properties, through feature engineering. To build trust in deep learning model outputs, we propose a CNN multi-classifier architecture to classify seismic signals into four classes (earthquake, micro-quake, rockfall and noise), and explain its outputs based on Layer-wise Relevance Propagation. We demonstrate that the provided explanations can lead to a more interpretable model by relating network outputs to geophysical phenomena and showing that distinguishing features extracted by the network are aligned with those identified by geoscientists as pertinent to classes of interest.

The labelling of large seismic datasets is a challenging problem. Currently the methods most favoured by geoscientists are based on well known geophysical properties with STA/LTA ratio pickers remaining highly trusted to generate results which can be quickly attributed due to their ability to pick relatively high Signal to Noise Ratio (SNR) events with high speed and accuracy. We aim to improve on the ability of deep learning methods by the unsupervised clustering of events which can help to visually identify results as belonging to a certain cluster with high confidence without the need for event by event processing. From our previous work we use a Siamese model trained with known labels from an open source dataset we show performance as a classifier and then expand on the method by showing clustering of events, where an expert can have high confidence that certain events are correctly identified, or require further evaluation.

The frequency and intensity of devastating landslides have been increasing worldwide. Timely prediction of slope failure can save lives and protect property. Slope movement is a result of several meteorological and hydrogeophysical variables, such as temperature and moisture content, but this complex relationship is still not well understood. To predict and characterise a slope failure, multiple measurands are usually collected. Since these numerous variables in the predictor set may cause significant increase in complexity, it becomes necessary to use methods that determine the relative importance of measurands that contribute directly to slope failure. To this end, we investigate three methods of visualisation of the feature space and dimensionality reduction, namely Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE) and Linear Discriminant Analysis (LDA), to analyse a range of surface and subsurface measurements from multiple sensors focusing on five stages of slope movement and then make failure predictions using XGBoost regression by setting as predictors two most important components from the extracted features. The results clearly show that LDA better clusters the data points and distinguishes the five different stages of slope movement, including two failures during the period of study encompassing eight years.

Induced microseismic monitoring has gained increased interest recently, to support various subsurface activities, including geothermal exploration and oil and gas production. To accurately detect and locate origins of microseismisity, deep learning-based methods have become popular due to their high accuracy when trained on large well-labelled datasets. However, though a huge amount of publicly available seismic measurements is available, laballed data to train models is very scarce, since labelling is time consuming and requires very specialist knowledge. Building on our prior work on active learning for time-series data, we propose an active learning method that cleverly picks only a small number of samples to query and stops when the proposed stopping criterion is met. We demonstrate that the proposed approach can save up to 83% of labelling effort even when transferred to a well with different sensing equipment from those used to build the training set.

Advanced machine learning algorithms require large datasets, along with good-quality labels to reach state-of-the-art performance. Although measurements themselves can often be easily available, the labelling process is usually a bottleneck. To address this, active learning approaches exploit the fact that different samples provide varying levels of information to the algorithm. However, these approaches often rely on several unrealistic assumptions — an oracle is assumed to provide error-free labels, all at the same cost and effort. We propose novel active learning-based methods for classification of time series measurements, typically obtained from sensors continuously measuring highly fluctuating environmental conditions including electricity consumption, and demonstrate their effectiveness for home energy management applications, where data labelling is a challenge. A new acquisition function is proposed, which accounts for both model and labelling uncertainty and class balancing. A stopping criterion is designed to stop the active learning process after an optimal point is achieved, to reduce labelling effort. We assess the effect of labelling errors on classification performance and propose two ways of mitigating their effects: (i) a re-labelling mechanism based on similarity of provided labels; (ii) a revised loss function based on confidence levels provided by experts. We validate our contributions for energy disaggregation task in a real-world scenario with three application domain experts. Our results show that the proposed methodology significantly improves performance of algorithms transferred to unseen domains with reduced number of labelled samples — from 61% reduction for dishwasher to 93% reduction for kettle.

Smart meters allow the grid to interface with individual buildings and extract detailed consumption information using nonintrusive load monitoring (NILM) algorithms applied to the acquired data. Deep neural networks, which represent the state of the art for NILM, are affected by scalability issues since they require high computational and memory resources, and by reduced performance when training and target domains mismatched. This article proposes a knowledge distillation approach for NILM, in particular for multilabel appliance classification, to reduce model complexity and improve generalization on unseen data domains. The approach uses weak supervision to reduce labeling effort, which is useful in practical scenarios. Experiments, conducted on U.K.-DALE and REFIT datasets, demonstrated that a low-complexity network can be obtained for deployment on edge devices while maintaining high performance on unseen data domains. The proposed approach outperformed benchmark methods in unseen target domains achieving a$F_{1}$-score 0.14 higher than a benchmark model 78 times more complex.

Knowledge distillation (KD) is a machine learning technique widely used in recent years for the task of domain adaptation and complexity reduction. It relies on a Student-Teacher mechanism to transfer the knowledge of a large and complex Teacher network into a smaller Student model. Given the inherent complexity of large Deep Neural Network (DNN) models, and the need for deployment on edge devices with limited resources, complexity reduction techniques have become a hot topic in the Non-intrusive Load Monitoring (NILM) community. Recent literature in NILM has devoted increased effort to domain adaptation and architecture reduction via KD. However, the mechanism behind the transfer of knowledge from the Teacher to the Student is not clearly understood. In this work, we aim to address the aforementioned issue by placing the KD NILM approach in a framework of explainable AI (XAI). We identify the main inconsistency in the transfer of explainable knowledge, and exploit this information to propose a method for improvement of KD through explainability guided learning. We evaluate our approach on a variety of appliances and domain adaptation scenarios and demonstrate that solving inconsistencies in the transfer of explainable knowledge can lead to improvement in predictive performance.

Automatic seismic signal classification methods are extensively investigated to reduce or replace manual interpretation, with great potential in previous research. Discriminative seismic wave propagation physical characteristics, such as velocities and accelerations, are rarely considered for classification. A multitask learning scheme is proposed that utilises the seismic wave equation and three-dimensional (3D) P-wave velocityVpmodel for signal representation learning. The classifier uses the obtained latent feature maps on a convolutional neural network architecture for classification of rockfall, slide quake, earthquake, and natural/anthropogenic noise events, recorded at an ongoing landslide. Our experimental results show that our approach outperforms state-of-the-art methods.

Passive seismics help us understand subsurface processes, e.g. landslides, mining, geothermal systems etc. and help predict and mitigate their effects. Continuous monitoring results in long seismic records that may contain various sources, which need to be classified. Manual detection and labeling of recorded seismic events is not only time consuming but can also be inconsistent when done manually, even in the case where it is done by the same expert. Therefore, an automated approach for classification of continuous microseismic recordings based on a Convolutional Neural Network (CNN) is proposed, with a multiclassifier architecture that classifies earthquakes, rockfalls and low signal to noise ratio quakes. Furthermore, we propose three CNN architectures that take as input time series data, Short Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT) maps. The suitability of these three networks is rigorously assessed over five months of continuous seismometer recordings from the active Super-Sauze landslide in France. We observe that all three architectures have excellent and very similar performance. Furthermore, we evaluate transferability to a geographically distinct seismically active site in Larissa, Greece. We demonstrate that the proposed network is able to detect all 86 catalogued earthquake events, having only been trained on the Super-Sauze dataset and shows good agreement with manually detected events. This is promising as it could replace painstaking manual labelling of events in large recordings.

As the need for using energy-efficient machinery escalates, energy consumption estimation plays an important role in decision support and planning in the agri-sector. Within the present research study, energy consumption in dairy farms was examined. A deep learning-based load disaggregation approach was used to develop data-driven models to quantify individual energy consumption of milk production-related devices of dairy farms, from a single aggregate measurement. According to the experiments conducted on three dairy farms in Germany, load disaggregation from a single aggregate meter is a viable, cheaper alternative to submetering multiple pieces of equipment to accurately quantify electricity consumption at scale in dairy farms in order to provide the decision support needed to inform measures for tackling climate change.

Microseismic monitoring has been increasingly used in the past two decades to illuminate (sub)surface processes, such as landslides, due to its ability to record small seismic waves generated by soil movement and/or brittle behavior of rock. Understanding the evolution of landslide processes is of paramount importance in predicting or even avoiding an imminent failure. Microseismic monitoring recordings are often continuous, noisy, and consist of signals emitted by various sources. Manually detecting and distinguishing the signals emitted by an unstable slope is challenging. Research on automated end-to-end denoising, detection, and classification of microseismic events, as an early warning system, is still in its infancy. To this effect, our work is focused on jointly evaluating and developing suitable approaches for signal denoising, accurate event detection, nonsite-specific feature construction, feature selection, and event classification. We propose an automated end-to-end system that can process big data sets of continuous seismic recordings fast and demonstrate applicability and robustness to a wide range of events (distant and local earthquakes, slidequakes, anthropogenic noise, etc.). Algorithmic contributions lie in novel signal processing and analysis methods with fewer tunable parameters than the state of the art, evaluated on two field data sets and benchmarked against the state of the art.

Classification of seismic events detected from seismic recordings has been gaining popularity for improved interpretation of subsurface processes, e.g., volcanic systems, earthquake activity, induced seismicity and slope stability, in particular landslides. However, due to the variability of signal representation for different classes in the temporal and spectral space, a large feature space is used to discriminate classes. The consequence is additional complexity on the classifier and overfitting. So far, there has been little attempt to address dimensionality reduction via feature selection. In this paper, we propose an iterative, alternating graph feature and classifier learning method for micro-seismic signals via graph Laplacian regularization and normalized graph Laplacian regularization. Using recorded micro-seismic events from an active landslide, we demonstrate improved classification accuracy with a relatively small feature space compared to the state-of-the-art.

Recent advances in image acquisition and analysis have resulted in disruptive innovation in physical rehabilitation systems facilitating cost-effective, portable, video-based gait assessment. While these inexpensive motion capture systems, suitable for home rehabilitation, do not generally provide accurate kinematics measurements on their own, image processing algorithms ensure gait analysis that is accurate enough for rehabilitation programs. This paper proposes high-accuracy classification of gait phases and muscle actions, using readings from low-cost motion capture systems. First, 12 gait parameters, drawn from the medical literature, are defined to characterize gait patterns. These proposed parameters are then used as input to our proposedmulti-channel time-series classificationand gait phase reconstruction methods. Proposed methods fully utilize temporal information of gait parameters, thus improving the final classification accuracy. The validation, conducted using 126 experiments, with 6 healthy volunteers and 9 stroke survivors with manually-labelled gait phases, achieves state-of-art classification accuracy of gait phase with lower computational complexity compared to previous solutions.1

Energy disaggregation of appliances using non-intrusive load monitoring (NILM) represents a set of signal and information processing methods used for appliance-level information extraction out of a meter's total or aggregate load. Large-scale deployments of smart meters worldwide and the availability of large amounts of data, motivates the shift from traditional source separation and Hidden Markov Model-based NILM towards data-driven NILM methods. Furthermore, we address the potential for scalable NILM roll-out by tackling disaggregation complexity as well as disaggregation on houses which have not been 'seen' before by the network, e.g., during training. In this paper, we focus on low rate NILM (with active power meter measurements sampled between 1-60 seconds) and present two different neural network architectures, one, based on convolutional neural network, and another based on gated recurrent unit, both of which classify the state and estimate the average power consumption of targeted appliances. Our proposed designs are driven by the need to have a well-trained generalised network which would be able to produce accurate results on a house that is not present in the training set, i.e., transferability. Performance results of the designed networks show excellent generalization ability and improvement compared to the state of the art.

Appliance fault in buildings resulting in abnormal energy consumption is known as an anomaly. Traditionally, anomaly detection is performed either at aggregate, i.e., meter-level, or at appliance level. Meter-level anomaly detection does not identify the anomaly-causing appliance, while appliance-level detection requires submetering each appliance in the building. Non-Intrusive Load Monitoring (NILM) has been proposed as an alternative to submetering to detect when appliances are running as well as estimate the appliance energy consumption. So far, applications have revolved around meaningful energy feedback. In this paper, we assess whether NILM can indeed be used for anomaly detection, as an alternative to submetering. We propose a supervised anomaly detection approach, AEM, and evaluate the effectiveness of NILM for anomaly detection. The proposed approach first learns an appliance's normal operation and then monitors its energy consumption for anomaly detection. We resort to real data, aggregate and subme-tered data from the two-year long REFIT dataset. We explain why anomaly detection performs worse with NILM data as compared to submetered data, highlighting the need for new, anomaly-aware NILM approaches.

With growing interest in data-driven classification, deep learning is now prevalent due to its ability to learn feature mapping functions solely from data. For very small training sets, however, deep learning, even with traditional regularization techniques, often overfits, resulting in sub-par classification performance. In this paper, we propose a novel binary classifier deep learning method, based on an iterative quadratic programming (QP) formulation with a graph Laplacian regularizer (GLR), combining the merits of model-based and data-driven approaches. Specifically, the proposed network employs a convolutional neural network (CNN) to learn deep features, which are used to define edge weights for a graph to pose a convex QP problem. Further, we design a novel loss function to penalize samples at the class boundary during semi-supervised learning. Results demonstrate that, given a small-size training dataset, our network outperforms several state-of-the-art classifiers, including CNN, model-based GLR and dynamic graph CNN classifiers.

In digital signal processing, a shift-invariant filter can be represented as a polynomial expansion of a shift operation, that is, the Z-transform representation. When extended to graph signal processing (GSP), this would mean that a shift-invariant graph filter can be represented as a polynomial of the shift matrix of the graph. Prior work shows that this holds under the shift-enabled condition that the characteristic and minimum polynomials of the shift matrix are identical. While the shift-enabled condition is often ignored in the literature, this letter shows that this condition is essential for the following reasons. First, we prove that this condition is not just sufficient but also necessary for any shift-invariant filter to be representable by the shift matrix. Moreover, we provide a counterexample showing that given a filter that commutes with a non-shift-enabled graph, it is generally impossible to convert the graph into a shift-enabled graph with a shift matrix still commuting with the original filter. The result provides a deeper understanding of shift-invariant filters when applied in GSP and shows that further investigation of shift-enabled graphs is needed to make them applicable to practical scenarios.

With growing ageing population, acquiring joint measurements with sufficient accuracy for reliable gait assessment is essential. Additionally, the quality of gait analysis relies heavily on accurate feature selection and classification. Sensor-driven and one-camera optical motion capture systems are becoming increasingly popular in the scientific literature due to their portability and cost-efficacy. In this paper, we propose 12 gait parameters to characterise gait patterns and a novel gait-phase classifier, resulting in comparable classification performance with a state-of-the-art multi-sensor optical motion system. Furthermore, a novel multi-channel time series segmentation method is proposed that maximizes the temporal information of gait parameters improving the final classification success rate after gait event reconstruction. The validation, conducted over 126 experiments on 6 healthy volunteers and 9 stroke patients with handlabelled ground truth gait phases, demonstrates high gait classification accuracy.

Localization of a viewer's region of interest (ROI) on eye gaze signal trajectories acquired by eye trackers is a widely used approach in scene analysis, image compression, and quality of experience assessment. In this paper, we propose a novel clustering approach for ROI estimation from potentially noisy raw eye gaze data, based on signal processing on graphs. The clustering approach adapts graph signal processing (GSP)-based classification by first cleverly selecting a starting data sample, and then classifying the remaining samples. Furthermore, Graph Fourier Transform is used to adjust GSP parameters on-the-fly to maximise accuracy. Experimental results show competitive clustering accuracy of our proposed scheme compared to Density-based spatial clustering of applications with noise (DB-SCAN), Distance-Threshold Identification (I-DT), and Mean-Shift on publicly available Shape Dataset and the potential of estimating ROI accurately on true eye tracker data1.

Optical motion capture systems have been widely adopted for human motion analysis in stroke rehabilitation because of real-time processing and high-accuracy features. However, these systems require a large laboratory space and multiple cameras and thus can be expensive and not transportable. In this paper, we propose a portable, cheap, single-camera motion analysis system to implement upper limb movement analysis. The proposed system consists of video acquisition, camera calibration, marker tracking, autonomous joint angle calculation, visualization, validation and classification. The validation with a state-of-the-art optical motion analysis system using Bland-Altman plot, a typical clinical measure, indicates that the proposed system can accurately capture elbow movement, trunk-tilt, and shoulder movement for diagnosis. Furthermore, the volunteers are explicitly classified into healthy and stroke groups via a support vector machine trained on statistics of the trunk-tilt and shoulder movement. Experimental results show that the proposed system can accurately capture the upper limb movement patterns, automatically classify stroke survivors using ordinal scale classification of upper limb impairment, and offer a convenient and inexpensive solution for upper limb movement analysis.

In an age where there is a strong dependency on electrical appliances for domestic routines, this paper proposes an algorithm for identifying domestic activities from non-intrusive smart meter aggregate data. We distinguish two types of activities: Type I activities are those that can be recognized using only smart meter data and Type II activities are recognized by combining smart meter data with basic environmental sensing (temperature and humidity). For both types of activities, we start by disaggregating the total power usage down to individual electrical appliances. Then, we build an indicative activity model to reason four domestic activities using the Dempster-Shafer theory of evidence. To validate our algorithms, we use real energy and environmental data collected in an actual UK household over a period of three months, benchmarked on a time-stamped log of activities. The results show that it is possible to detect four tested domestic daily activities with high accuracy based on the aggregate energy usage.

Human gait analysis for stroke rehabilitation therapy using video processing tools has become popular in recent years. This paper proposes a single-camera system for capturing gait patterns using a Kalman-Structural-Similarity-based algorithm which tracks multiple markers simultaneously. This algorithm is initialized by obtaining the user-selected blocks in the first frame of each video, and the tracker is implemented by using Structural-Similarity image quality assessment algorithm to detect each marker frame by frame within a search area determined by a discrete Kalman filter. Experimental results show the trajectories of the markers fixed on the joints of a human body. The obtained numerical results are used to generate gait information (e.g., knee joint angle) that is later used for diagnostics. The proposed method aims to explore an alternative and portable way to implement human gait analysis with significantly less cost compared to a state-of-the-art 3D motion capture system.

Multiple description coding (MDC) has been proposed as a possible solution to real-time video delivery over relay-assisted wireless networks to exploit path diversity. In this paper, we study layered multiple description video over relay-assisted mobile networks, such as LTE-A, and develop a framework that routes packets to different relays based on source-channel-relay parameters. The proposed system comprises, besides layered MDC, (i) application-layer forward error correction via Random linear codes (RLC) that are very suitable in the relaying scenarios, due to their rateless nature, (ii) unequal error protection (UEP), using the recently proposed expanding window technique, that relies on probabilistic scheduling of layered packets. We generate layered multiple descriptions using data slicing and data partitioning features of H.264/AVC and conduct simulations by modelling a two-relay LTE-A setup with the finite state Markov chain model. Simulation results show benefits of relaying with optimized routing and UEP.

In order to estimate the amount of energy that will be generated by a wind farm and provide efficient power distribution planning, it is necessary to deliver information of wind speed at all wind turbines. This paper proposes a scheme for compressing wind speed measurements exploiting both temporal and spatial correlation between the turbine readings via distributed source coding. The proposed scheme relies on a correlation model based on true measurements. A compression scheme proposed is of low encoding complexity and uses a particle-filtering based belief propagation decoder that adaptively estimates the nonstationary noise of the correlation model. Simulation results using realistic models show significant performance improvements compared to the scheme that does not dynamically refine correlation.

The authors propose two practical power- and bandwidth-efficient systems based on amplify-and-forward and decode-and-forward schemes to address the problem of information exchange via a relay. The key idea is to channel encode each source's message by using a high-performance non-binary turbo code based on partial unit memory codes to enhance the bit-error-rate performance, then reduce the energy consumption and increase spectrum efficiency by using network coding (NC) to combine individual nodes’ messages at the relay before forwarding to the destination. Two simple and low complexity physical layer NC schemes are proposed based on combinations of received source messages at the relay. The authors also present the theoretical limits and numerical analysis of the proposed schemes. Simulation results under additive white Gaussian noise confirm that the proposed schemes achieve significant bandwidth savings and fewer transmissions over the benchmark systems which do not resort to NC. Theoretical limits for capacity and signal-to-noise ratio behaviour for the proposed schemes are derived. This study also proposes a cooperative strategy that is useful when insufficient combined messages are received at a node to recover the desired source messages, thus enabling the system to retrieve all packets with significantly fewer retransmission request messages.

Distributed video coding (DVC) is rapidly gaining popularity as a low cost, robust video coding solution, that reduces video encoding complexity. DVC is built on distributed source coding (DSC) principles where correlation between sources to be compressed is exploited at the decoder side. In the case of DVC, a current frame available only at the encoder is estimated at the decoder with side information generated from other frames available at the decoder. One of the main challenges in DVC design is that correlation among the source and side information needs to be estimated online and as accurately as possible. Since correlation dynamically changes with the scene, in order to exploit the robustness of DSC code designs, we integrate particle filtering (PF) with standard belief propagation (BP) decoding for inference on one joint factor graph to estimate correlation among source and side information. Correlation estimation is performed online as it is carried out jointly with decoding of the graph-based DSC code. Moreover, we demonstrate our joint bit-plane decoding with adaptive correlation estimation schemes within state-of-the-art DVC systems, which are transform-domain based with a feedback channel for rate adaptation. Experimental results show that our proposed system gives a significant performance improvement compared to the benchmark state-of-the-art DISCOVER codec (including correlation estimation) and the case without dynamic PF tracking, due to improved knowledge of timely correlation statistics via the combination of joint bit-plane decoding and particle-based BP (PBP) tracking.

We propose an adaptive distributed compression solution using particle filtering that tracks correlation, as well as performing disparity estimation, at the decoder side. The proposed algorithm is tested on the stereo solar images captured by the twin satellites system of NASA's Solar TErrestrial RElations Observatory (STEREO) project. Our experimental results show improved compression performance w.r.t. to a benchmark compression scheme, accurate correlation estimation by our proposed particle-based belief propagation algorithm, and significant peak signal-to-noise ratio improvement over traditional separate bit-plane decoding without dynamic correlation and disparity estimation.

In this paper, we propose an adaptive Distributed Video Coding (DVC) scheme that dynamically estimates correlation statistics of the scene in a video sequence to enhance belief-propagation (BP) Slepian-Wolf (SW) decoding. In order to exploit the robustness of distributed source coding (DSC) designs, we integrate particle filtering with standard BP decoding in one factor graph to estimate online correlation among source and side information. Our proposed system boasts improved performance over classical DVC without correlation estimation, due to improved knowledge of correlation statistics via the combination of bit-plane coding and particle-based BP tracking in each frame, as shown by our results.

This paper describes a complete low-complexity imaging system based on a single MEMS scanning mirror and a single photodetector, together with customized image enhancement algorithms based on sparse signal representation. Due to very low complexity of our developped optical set-up for image acquisition, resulting images suffer visible artifacts. We propose an iterative denoising-deblurring algorithm for image enhancement, which offers significant improvement over wavelet denoising with soft-thresholding. Several image enhancement algorithms are compared using the blind image quality indices (BIQI) as well as visual experience.

Processing of solar image data has become increasingly important for accurate space weather prediction and expanding our understanding about the Sun and Universe. To enable proper analysis, image denoising and contrast enhancement are essential for removal of all artifacts introduced within the acquisition process. Hence, this paper focuses on these two tasks applied on solar images corrupted with pixel dependent Poisson and zero-mean additive Gaussian noise. The denoising frameworks are build upon on two state-of-the-art techniques, K-SVD and BM3D (for natural images) where contrast enhancement of noisy solar images is performed jointly with noise removal using sparse coding adaptive dictionary learning. Results are given for two conventional sets of solar images.

The paper presents a scalable compressive sampling (CS) scheme for video acquisition. The proposed solution enables progressive reconstruction of video frames with novel measurement matrices. Simulation results show significant performance improvements over the traditional CS technique for the base layer and slightly better performance for the final enhancement layer.

Distributed compression or Distributed Source Coding (DSC) refers to separate compression and joint decompression of multiple correlated sources. Though theoretical foundations were set almost forty years ago, driven by applications such as wireless video surveillance and wireless multimedia communications, DSC has become and still is a very active research area with interest from both academia and industry. While the first decade of this century has seen massive progress in code designs and achievable bounds, recent work is focussed on applications and solving practical DSC limitations such as large codeword size, accurate correlation estimation, etc. The goal of this paper is to provide an overview of the exciting trends and novel applications such as security, remote sensing, wireless data gathering of correlated data, biomedical imaging and infotainment, that DSC makes possible.

Acquiring and processing astronomical images is becoming increasingly important for accurate space weather prediction and expanding our understanding about the Sun and the Universe. These images are often rich in content, large in size and dynamic range. Efficient, low-complexity compression solutions are essential to reduce onboard storage, processing, and communication resources. Distributed compression is a promising technique for onboard coding of solar images by exploiting correlation between successively acquired images. In this paper we propose an adaptive distributed compression solution using particle filtering that tracks correlation, as well as performing disparity estimation, at the decoder side. The proposed algorithm is tested on the stereo solar images captured by the twin satellites system of NASA's STEREO project. Our experimental results show the significant PSNR improvement over traditional separate bit-plane decoding without dynamic correlation and disparity estimation.

Application-layer forward error correcting (AL-FEC) codes, providing error protection across blocks of packetized data, are becoming increasingly important in emerging networking applications. In this paper, we investigate AL-FEC solutions based on rate- compatible low-density parity-check (RC LDPC) codes, as they may offer excellent performance with low system complexity. Recent studies on RC LDPC codes are focused on physical layer incremental redundancy hybrid automatic-repeat-request (IR-HARQ) systems and typical wireless channel models. The focus of our study is RC LDPC code design as an AL-FEC solution in packet-level IR-HARQ systems assuming erasure channels between the source and the destination(s). We propose novel design methods for RC LDPC codes over erasure channels motivated by the dependance of finite-length LDPC code performance on the size of the smallest stopping set of its code graph. Simulation experiments demonstrate that the proposed RC LDPC design significantly outperforms the state-of-the-art RC LDPC codes in the error-floor region, and in particular, in the domain of poor channel conditions.

Scale Invariant Feature Transform (SIFT) is a very powerful technique for image registration. While SIFT descriptors accurately extract invariant image characteristics around keypoints, the commonly used matching approach for registration is overly simplified, because it completely ignores the geometric information among descriptors. In this paper, we formulate keypoint matching as a global optimization problem and provide a suboptimum solution using belief propagation. Experimental results show significant improvement over previous approaches.

This paper addresses the problem of wireless delivery of a captured scene from two cameras, which do not communicate with each other, to a central point for joint decoding. We exploit correlation among two camera views using distributed source coding and use complete complementary (CC) data spreading to combat multiple access interference and noise even when the transmitters are de-synchronized. Our distributed source coding scheme is based on uniform scalar quantization in the DCT domain and non-asymmetric Slepian-Wolf coding via turbo codes. The non-asymmetric Slepian-Wolf scheme enables efficient trade-off between the transmission rates of the two cameras. Simulation results indicate that the proposed system outperforms significantly two independently JPEG-encoded streams at low transmission rates.

This paper describes an optical system for low-complexity optical image acquisition based on a single scanning MEMS mirror and a single photodetector. The overall aim of the research is to investigate techniques for image acquisition at electromagnetic wavelengths where the cost and/or technical maturity of detector arrays pose a limitation. In contrast to similar systems built using a digital micromirror device (DMD), the present configuration has advantages of lower cost and potential applicability across a wide spectrum, ranging from visible to Terahertz frequencies. In the present arrangement, light at visible wavelengths from the object passes through a telescope and falls onto a small, scanning MEMS micromirror. The entire image of the object is projected onto the mirror surface and reflected towards a single photodetector with a pinhole at its entrance. Similarly to conventional scanning, by finely changing the tilt-angle of the mirror, the detector sees different areas of the projected image, thereby building up an image pixel-by-pixel. Resolution is increased by allowing for an overlap between neighbouring scanned areas. Iterative bilinear interpolation and wavelet denoising are employed to enhance image quality.

Compressive sampling is a novel framework that exploits sparsity of a signal in a transform domain to perform sampling below the Nyquist rate. In this paper, we apply compressive sampling to reduce the sampling rate of images/video. The key idea is to exploit the intra- and inter-frame correlation to improve signal recovery algorithms. The image is split into non-overlapping blocks of fixed size, which are independently compressively sampled exploiting sparsity of natural scenes in the Discrete Cosine Transform (DCT) domain. At the decoder, each block is recovered using useful information extracted from the recovery of a neighboring block. In the case of video, a previous frame is used to help recovery of consecutive frames. The iterative algorithm for signal recovery with side information that extends the standard orthogonal matching pursuit (OMP) algorithm is employed. Simulation results are given for Magnetic Resonance Imaging (MRI) and video sequences to illustrate advantages of the proposed solution compared to the case when side information is not used.

The paper combines distributed source coding and compressive sampling for efficient spectrum estimation. Two or more cognitive radios sample the spectrum compressively and independently compress their observations using multiterminal source coding. A central hub collects compressed streams from these radios before performing joint multiterminal source decoding followed by iterative signal reconstruction. Simulation results are provided for two radios performing practical multiterminal source coding with uniform scalar quantization and systematic turbo codes for Slepian-Wolf coding and error protection.

We propose a new technique, compress-spread forward (CSF), for high-performance wireless streaming from two base stations in parallel. CSF uses multiterminal source coding for efficient source compression and complete complementary sequences for error-free multiple access and synchronization. Our practical design shows significant performance gains due to spatial diversity and distributed source coding.

Fountain codes were introduced as an efficient and universal forward error correction (FEC) solution for data multicast over lossy packet networks. They have recently been proposed for large scale multimedia content delivery in practical multimedia distribution systems. However, standard fountain codes, such as LT or Raptor codes, are not designed to meet unequal error protection (UEP) requirements typical in real-time scalable video multicast applications. In this paper, we propose recently introduced UEP expanding window fountain (EWF) codes as a flexible and efficient solution for real-time scalable video multicast. We demonstrate that the design flexibility and UEP performance make EWF codes ideally suited for this scenario, i.e., EWF codes offer a number of design parameters to be “tuned” at the server side to meet the different reception criteria of heterogeneous receivers. The performance analysis using both analytical results and simulation experiments of H.264 scalable video coding (SVC) multicast to heterogeneous receiver classes confirms the flexibility and efficiency of the proposed EWF-based FEC solution.

Data streaming from multiple base stations to a client is recognized as a robust technique for multimedia streaming. However the resulting transmission in parallel over wireless channels poses serious challenges, especially multiple access interference, multipath fading, noise effects and synchronization. Spread spectrum techniques seem the obvious choice to mitigate these effects, but at the cost of increased bandwidth requirements. This paper proposes a solution that exploits complete complementary spectrum spreading and data compression techniques jointly to resolve the communication challenges whilst ensuring efficient use of spectrum and acceptable bit error rate. Our proposed spreading scheme reduces the required transmission bandwidth by exploiting correlation among information present at multiple base stations. Results obtained show 1.75 Mchip/sec (or 25%) reduction in transmission rate, with only up to 6 dB loss in frequency-selective channel compared to a straightforward solution based solely on complete complementary spectrum spreading.

Latest advances in sensor and communication technologies and computational intelligence will allow vehicles to observe their surroundings, process the collected information and communicate it with other vehicles, e.g., vehicles can sense adverse road conditions, experience trafic congestion, or detect accidents, and then adjust vehicle control. We investigate the problem of effective intelligence gathering in the perspectives of both communications and data fusion. The target system should have high processing and transmission accuracy for the gathered information as well as low communication load requirement. Assuming a binary source, we determine the theoretical bound for the error probability of gathered information and present a way of communicating the observation from individual vehicles to the destination. Compared to the flooding method, our model reduces the communication load by approximately N/log2N times, with N being the total number of the vehicles, without decreasing the accuracy of the gathered information.

This paper addresses a data-hiding problem where a source image needs to be hidden into another host image. We consider both cases when the decoder has access to the host image (non-blind data-hiding), and when the decoder does not (blind data-hiding). Our proposed solution combines complete complementary (CC) spreading sequences and nested scalar quantization (NSQ). To enhance security, the source image is encrypted prior to embedding. Our simulation results, for both non-blind and blind data-hiding, with AWGN attacks show competitive results. Indeed, for blind data-hiding, our design outperforms the traditional NSQ system by 8 dB.

Digital Fountain (DF) codes have recently been suggested as an efficient forward error correction (FEC) solution for video multicast to heterogeneous receiver classes over lossy packet networks. However, to adapt DF codes to low-delay constraints and varying importance of scalable multimedia content, unequal error protection (UEP) DF schemes are needed. Thus, in this paper, Expanding Window Fountain (EWF) codes are proposed as a FEC solution for scalable video multicast. We demonstrate that the design flexibility and UEP performancemake EWF codes ideally suited for this scenario, i.e., EWF codes offer a number of design parameters to be “tuned” at the server side to meet the different reception conditions of heterogeneous receivers. Performance analysis of H.264 Scalable Video Coding (SVC) multicast to heterogeneous receiver classes confirms the flexibility and efficiency of the proposed EWF-based FEC solution.