Multi-modal image fusion (MMIF) enhances the information content of the fused image by combining the unique as well as common features obtained from different modality sensor images, improving visualization, object detection, and many more tasks. In this work, we introduce an interpretable network for the MMIF task, named FNet, based on an $\ell _{0}$ℓ0-regularized multi-modal convolutional sparse coding (MCSC) model. Specifically, for solving the $\ell _{0}$ℓ0-regularized CSC problem, we design a learnable $\ell _{0}$ℓ0-regularized sparse coding (LZSC) block in a principled manner through deep unfolding. Given different modality source images, FNet first separates the unique and common features from them using the LZSC block and then these features are combined to generate the final fused image. Additionally, we propose an $\ell _{0}$ℓ0-regularized MCSC model for the inverse fusion process. Based on this model, we introduce an interpretable inverse fusion network named IFNet, which is utilized during FNet's training. Extensive experiments show that FNet achieves high-quality fusion results across eight different MMIF datasets. Furthermore, we show that FNet enhances downstream object detection and semantic segmentation in visible-thermal image pairs. We have also visualized the intermediate results of FNet, which demonstrates the good interpretability of our network.

Bearing fault diagnosis is crucial for ensuring the reliability and safety of industrial systems, particularly in preventing operational failures and maintaining product quality. Traditional signal processing methods and deep learning algorithms, while useful, often overlook the complex structural relationships within sensor data, limiting their diagnostic effectiveness. To address this, we present a novel Digital Twin-Driven Fault Diagnosis Framework that integrates graph-based learning techniques with advanced signal analysis. Our approach employs XGBoost and GraphSAGE embeddings to capture both spatial and temporal correlations within the current signals. The raw sensor data is then processed using a sliding window technique and time-frequency domain features are extracted then transformed into a graph structure that represents the intricate relationship in the signal. GraphSAGE is then applied to these graph structures, generating embeddings that enhance fault detection accuracy. Additionally, XGBoost is utilized for classification, improving the overall robustness of the system. The proposed method, deployed on edge devices, delivers real-time diagnostics, providing a scalable and efficient solution for industrial applications. Experimental results using real-world datasets demonstrate that our method significantly outperforms state-of-the-art algorithms, improving detection accuracy and Area under the curve (AUC) scores up to 97% and 99%, respectively.

Seismic Fault Detection is a crucial aspect of oil exploration. While traditional deep learning methods struggle to handle complex seismic data patterns, training a deep learning model solely on synthetic seismic data may not yield satisfactory results. This research paper, involves utilizing a pre-trained vision-based transformer to extract relevant features from seismic data. By leveraging the knowledge learned from a different but related task, the model can capture general fault patterns in field data. In this framework, a portion of the pre-trained model architecture is employed and trained using a Structural Similarity-based loss function to learn fault-related features. This allows the model to adapt to the faulted structures of a geological dataset and improve fault detection performance in field data applications. In comparison to the state-of-the-art method, the proposed method yields improved results on real field dataset.

Conventional hard-label and soft-label labeling strategies for Speech Emotion Recognition (SER) fail to capture the diversities in annotator expertise in perceiving complex human emotions. This study introduces novel soft-label approaches that integrate emotion-specific annotator abilities and optimize database utilization by incorporating non-consensus and majority-voted non-target class utterances. We evaluate our methods primarily on the IEMOCAP database, with additional verification using the MSP-Podcast database, employing features from wav2vec 2.0 and WavLM models. Compared to the conventional hard-label approach, our method improves Unweighted Accuracy (UWA) by 6.80% and 2.73% with WavLM features for the IEMOCAP and MSP-Podcast databases, respectively. The results also outperform other state-of-the-art soft-label approaches in SER literature. Our study shows the importance of accounting for the annotator’s expertise and the inherent subjectivity in emotional perception for improving the SER performance.

Non-invasive and continuous blood pressure (BP) monitoring is essential for the early prevention of many cardiovascular diseases. Estimating arterial blood pressure (ABP) from photoplethysmography (PPG) has emerged as a promising solution. However, existing deep learning approaches for PPG-to-ABP reconstruction (PAR) encounter certain information loss, impacting the precision of the reconstructed signal. To overcome this limitation, we introduce an invertible neural network for PPG to ABP reconstruction (INN-PAR), which employs a series of invertible blocks to jointly learn the mapping between PPG and its gradient with the ABP signal and its gradient. INN-PAR efficiently captures both forward and inverse mappings simultaneously, thereby preventing information loss. By integrating signal gradients into the learning process, INN-PAR enhances the network’s ability to capture essential high-frequency details, leading to more accurate signal reconstruction. Moreover, we propose a multi-scale convolution module (MSCM) within the invertible block, enabling the model to learn features across multiple scales effectively. We have experimented on two benchmark datasets, which show that INN-PAR significantly outperforms the state-of-the-art methods in both waveform reconstruction and BP measurement accuracy. Codes can be found at: https://github.com/soumitra1992/INNPAR-PPG2ABP.

Improving the quality of underwater images is essential for advancing marine research and technology. This work introduces a sparsity-driven interpretable neural network (SINET) for the underwater image enhancement (UIE) task. Unlike pure deep learning methods, our network architecture is based on a novel channel-specific convolutional sparse coding (CCSC) model, ensuring good interpretability of the underlying image enhancement process. The key feature of SINET is that it estimates the salient features from the three color channels using three sparse feature estimation blocks (SFEBs). The architecture of SFEB is designed by unrolling an iterative algorithm for solving the ℓ1regulaized convolutional sparse coding (CSC) problem. Our experiments show that SINET surpasses state-of-the-art PSNR value by 1.05 dB with 3873 times lower computational complexity. Code can be found at: https://github.com/gargi884/SINET-UIE/tree/main.

Early fault detection in induction motors is critical for industrial reliability, with rotor faults representing challenging diagnostic scenarios due to their gradual development and subtle manifestations. Conventional fault diagnosis techniques suffer from limited effectiveness in early-stage detection and poor performance under variable operating conditions. This paper presents a novel Digital Twin framework enabling real-time motor behavior simulation, generating graph representations that capture spatial-temporal relationships between rotor components. Graph Convolutional Networks (GCNs) learn from these graph-encoded representations, leveraging neighborhood connectivity for enhanced fault feature extraction. The methodology encompasses signal acquisition from a 32-bit processor-based digital twin simulator, coupled circuit modeling of rotor fault conditions, graph construction encoding physical relationships, and specialized GCN architecture with anomaly detection capabilities. Comprehensive validation using simulated and experimental datasets demonstrates robustness under variable load conditions. The proposed method achieves 93% accuracy using GCN combined with CatBoost, significantly outperforming conventional approaches. This integration enhances real-time fault diagnosis capabilities, enabling proactive predictive maintenance and improving operational reliability.

Arcing faults pose a major threat in low-voltage distribution systems, as they generate intense heat, sparks, and plasma. These phenomena can cause fires, equipment damage, and serious safety hazards. This paper presents a novel modulation-based modeling approach to characterize electromagnetic radiation from arc faults. Using measured data, the modulating signal is extracted by analyzing the power spectral density properties of the electromagnetic radiation. The extracted modulating signal is then decomposed using a multilevel discrete wavelet transform to isolate sideband arcing features. This enables robust statistical feature extraction for improved arc fault detection.

An automatic depression detection (ADD) system using spoken language offers the opportunity to develop practical, low-cost tools to detect symptoms early. However, limited data availability, privacy concerns, and transcription efforts pose significant challenges. Recent advancements in foundational models, capable of understanding and processing multimodal inputs, present opportunities for enhancing ADD systems. This study explores various speech foundation models to investigate their impact on ADD. We leverage Whisper and MMS for automatic transcription and integrate speech and text embeddings into a language model optimized with low-rank adaptation (LoRA). In addition, we examine the effects of fine-tuning strategies and prompt formats on model performance. We used English and Bengali datasets to demonstrate the potential of our method in ADD, even with moderate-quality transcriptions. The best speech and language foundation models outperform baseline models on both datasets.

Interpreting seismic data involves finding out subsurface geologic information. In seismic data interpretation, one of the crucial steps is to delineate seismic faults. Natural gas and oil reservoirs are more likely to be present where seismic faults exist. In this paper, we develop a graph neural network based approach for finding faults in seismic data using graph total variation. Our proposed methodology begins with the first step, the extraction of patches for all training points (pixels). In the graph domain, these patches appear as individual graphs. We use the graph total variation as graph attributes and seismic amplitudes as node attributes for the graphs. The next step is implementing graph neural networks (GNNs) for graph classification and fault delineation. The proposed methodology offers higher accuracy and improved time complexity when implemented on real data.

Existing speech emotion recognition (SER) research is developed assuming static human emotions. However, emotions are dynamic in nature. In this work, we use IEMOCAP databases and modify the conventional SER systems by capturing the intra-utterance temporal dynamics of human emotions. We explore signal-level and feature-level descriptors of emotion variability and compute the utterance-wise mean-squared-successive-difference (MSSD). The MSSDs are used to recognize the utterances as low or high inertia. Emotion inertia defines the trait of emotion propagating from one time to the next. The emotional states in the low inertia utterances exhibit frequent transitions. We show that such transitions prominently degrade the conventional SER system’s performance. In our work, we detect the low inertia utterances during evaluation and proposed methodologies to nullify the stray emotion transitions. The modifications improve the overall performance by 2.87%.

Subsurface lithology identification from well log signals is a crucial step in geological exploration, providing essential information about rock formation properties and fluid flow. Accurate identification of lithofacies aids in reservoir characterization and hydrocarbon exploration. This letter presents a novel approach for lithofacies identification from well logs using graph-based feature extraction and classification. The existing instance-based methods ignore the sequential information in well log signals, which can provide valuable insights about the local lithology. The proposed approach treats each instance in a temporal sequence as a node in a graph that captures the local geological information by aggregating temporally neighborhood nodes to create an embedded feature space. Two separate aggregating schemes are proposed, one using a spatial kernel approach and the other using an attention-based network layer, to find the nonlinear relationship between the feature vectors and give more weights to the nearest vectors in the feature space. The graph structure allows the network to incorporate spatial and relational information between different well log features into the classification process, leading to improved accuracy of predictions. The experiment is run on real-world data from the oil and gas exploration field at Krishna-Godavari (KG) Basin, India. The proposed method outperforms traditional feature-based classification and provides a unique way to enhance the representation of the well log signals for lithofacies classification tasks.

Geological fault detection is a crucial aspect of oil exploration. With the advancements in deep learning, the challenging task of accurate fault detection has gained popularity. While traditional deep learning methods struggle due to the small sample problem and the labor-intensive fault labeling process, training a deep learning model solely on synthetic seismic data may not yield satisfactory results due to the disparities between synthetic and real seismic data. To mitigate the impact of these differences, we propose employing an instance weighting (IW)-based transfer learning (TL). This approach involves utilizing a pretrained deep-learning model to extract fault-related features from seismic data. By leveraging the knowledge learned from a different but related task, the TL model can capture general fault patterns that can be applicable to real seismic data. In this framework, a portion of the pretrained model is employed to learn fault-related features, which can then be fine-tuned using a smaller amount of labeled real seismic data. This allows the model to adapt to the complexities of the actual geological situation and improve fault detection performance in field data applications. The proposed method has been tested on the Indian Krishna Godavari Basin dataset. The method yields satisfying results in spite of the high imbalance between the fault and nonfault classes.

This article presents a data-driven probabilistic framework for estimating missing values in multivariate and time-varying interdependent IoT data streams during reefer container monitoring. It models the periodic fluctuations in the temperature and humidity due to the refrigeration cycle using a log-normal distribution, followed by the estimation of missing values using the sparse vector autoregression (sVAR) model. The accuracy of sVAR is improved by considering the spatio-temporal correlation of sensor signals while computing the model parameters and a kernel-based weighting scheme. It is applied to a dataset collected during an experiment. The results show that it outperforms a few baseline methods while providing a comprehensive solution for both point-missing values as well as large gap situations considering co-occurring and non-co-occurring cases.

Multimodal speech emotion recognition (MSER) is an emerging and challenging field of research due to its more robust characteristics than unimodal. However, in multimodal approaches, the interactive relations for model building using different modalities of speech representations for emotion recognition have not been well investigated yet. To address this issue, we introduce a new approach to capturing the deep temporal features of audio and text. The audio features are learned with a convolution neural network (CNN) and a Bi-directional Gated Recurrent Unit (Bi-GRU) network. The textual features are represented by GloVe word embedding along with Bi-GRU. A cross-modal transformers block is designed for multimodal learning to capture better inter- and intra-interactions and temporal information between the audio and textual features. Further, a self-attention (SA) network is employed to select more important emotional information from the fused multimodal features. We evaluate the proposed method on the IEMOCAP dataset on four emotion classes (i.e., angry, neutral, sad, and happy). The proposed method performs significantly better than the most recent state-of-the-art MSER methods.

This paper presents a new approach to detect weak signatures of fault frequency components associated with inner race bearing faults in low voltage, high current Permanent Magnet Synchronous Motor (PMSM) under light load conditions. Inner race faults in bearing produce very weak signatures in the stator current spectrum, which is difficult to detect due to the high amplitude fundamental component of motor current. The proposed method combines a preconditioning Extended Kalman Filter (EKF), Low-Rank Approximation of Hankel Matrix, and the Matrix Pencil Method (MPM) to achieve robust fault detection by suppressing the high magnitude fundamental component. The EKF is utilized to pre-process the current signal, reducing noise and extracting fault-related information. A subsequent Low-Rank Approximation generates a Hankel matrix, capturing the fault-related damped sinusoidal components in the presence of background noise. The MPM is then employed to estimate the fault-related frequency components. Experimental evaluations are conducted on synthetic and real-world data to assess the performance of the proposed method. Results indicate improved accuracy in estimating the fault frequency components, even in the presence of noise and other disturbances. Fault detection capability is significantly enhanced, demonstrating the effectiveness of the proposed approach. This work contributes to a comprehensive fault detection framework for PMSM motors, enabling early identification of inner race faults. The proposed method shows promise for practical implementation in industrial applications, facilitating timely maintenance and reducing the risk of catastrophic failures.

Speech emotion recognition (SER) is an essential component of human-computer interaction systems, enabling machines to understand and respond to human emotions. To our knowledge, this is the first study that explicitly proposes emotion elicitation information to perform curriculum learning (CL) in SER tasks by incorporating both acted and spontaneous data. We utilize IEMOCAP and BAUM-1 databases that comprise both acted and spontaneous speech samples. We use wav2vec 2.0 transformer layer extracted representations as features and train them on ECAPA-TDNN classifier. We train separate SER models using acted and spontaneous training data and perform same-elicitation and cross-elicitation performances as curriculum generator tasks. Following that, our approach involves categorizing the data into easy (acted) and hard (spontaneous) training subsets. We first train the SER model with the easy subset after that, sequentially feed the hard subset using batch sampling and learning rate scaling. By employing CL, the model gradually adapts to the increased difficulty and complexity, leading to improved robustness against elicitation variations across the evaluation data. Our experimental results show that the proposed CL approach improves SER performance by 2.77% and 3.41% for the IEMOCAP and BAUM-1 databases, respectively, compared to systems trained without CL.

The integration of Information and Communications Technology (ICT) for mental Health (m-Health) detection and control in the smart framework (e-Framework) opens the opportunity for developing the intelligent Human-device Interaction (HDI) system. m-Health detection using multi-modality cues: Respiration Wave Pattern (RWP), Heart Wave Pattern (HWP), and visual RGB facial expression of a person convey more accurate information to build a robust and efficient system for an early stage of m-Health recognition. This study proposes a k-nearest neighbor (k-NN) based Graph Neural Network (GNN) with the input domain modality of RGB, RWP, and HWP to interpret the six discrete m-Health, such as Anger, Fear, Disgust, Joy, Sad, and Surprise. The average classification performance of our proposed k-NN-GNN architecture is 98. 94% for our in-house experimental dataset, validated with the subjective evaluation ground truth data of SAM score, 98.04%, 98.07%, and 98.06% for AMIGOS, AMHUSE, and MAHNOB, respectively. A User Interface Andriod Application (UIA) connected to the e-Framework provides Negative Mental Health (NMH) distraction suggestions to the end user for an early stage of NMH detection and control.

In recent years, there has been significant progress in utilizing neural networks and deep learning methods for enhancing the delineation of seismic faults. However, the scarcity of labeled data has posed a challenge in training such networks, leading to a reliance on synthetic samples. Consequently, the task of annotation has become a crucial component within machine learning frameworks. Data labeling not only consumes considerable time but also necessitates a high level of precision. To address the above limitations, an Artificial Intelligence-powered interactive annotation tool has been developed. This tool aims to minimize the immense human effort involved in labeling data by offering an efficient and accurate solution. By leveraging the power of artificial intelligence, the tool enables faster and more precise annotation. The effectiveness and reliability of the proposed tool are affirmed through the observed enhancements in segmentation quality and the average speedup achieved in the annotation process.

Seismic sparse layer inversion (SLI) plays a crucial role in improving the resolution of seismic data for accurate subsurface characterization. However, achieving high accuracy and efficient runtime in SLI is of paramount importance for reservoir characterization. In this paper, we address the SLI problem by constructing a dictionary using odd and even reflection coefficients. The conventional Linear Programming (LP) approach suffers from the equal penalization of all model parameters, resulting in ghost layers and high computational costs on Central Processing Units(CPUs). To overcome these limitations, we propose an enhanced formulation by incorporating Tikhonov regularization, dynamically penalizing the model parameters based on the target seismic trace. Moreover, we optimize the computational runtime by leveraging the power of a Graphical Processing Unit (GPU)-based LP solver. Our algorithm is implemented on an NVIDIA RTX 4000 GPU with 8GB dedicated memory and tested on the Indian WADU dataset. Results demonstrate the efficacy of our proposed method in accurately delineating thin seismic layers compared to state-of-the-art techniques. Additionally, we highlight the superior runtime performance of the GPU-based LP formulation compared to its CPU implementation, further enhancing the efficiency of the SLI process.

Many activities such as drilling and exploration in the oil and gas industries rely on identifying seismic faults. Using graph high-frequency components as inputs to a graph convolutional network, we propose a method for detecting faults in seismic data. In Graph Signal Processing (GSP), digital signal processing (DSP) concepts are mapped to define the processing techniques for signals on graphs. As a first step, we extract patches of the seismic data centered around the points of concern. Each patch is then represented in a graph domain, with the seismic amplitudes as the graph signals. We attenuate the low-frequency components of the signal with the aid of a graph high-pass filter. By applying the graph Fourier transform, we obtain the graph high-frequency components. These graph high-frequency components act as inputs to a graph convolutional network (GCN). By classifying the patches using GCN, we identify the faults in data.

Seismic data is an aggregation of traces that shows a spatial and temporal sampling of reflected wavefields—the identification of reservoir localization based on thin layer selection from the earth layers. The challenge to finding thin layers is that the information acquired gets distorted limiting reservoir localization due to subsurface complexity and random environmental noise. Dictionary elements from structure functions represent reflectivity patterns used in inversion seismic data. The seismic traces is a superposition of constituting reflectivity patterns. Traditionally noise separation is done by utilizing different properties in the various transform domains. The results lack adaptability and consideration of physical layer properties. In this paper, the dictionary is carefully chosen based on spatial source wavefield considers horizontal and vertical wavefield components having translational and rotational components. The horizontal source wavefield components coupled with wedges formed from reflectivity pair then convolved with seismic wavelet. The generated output is reflectivity inversion is based on the source wavefield dictionary. The proposed method is robust to noise as the dictionary used in inversion is built with the same spatial source wavefield; hence it could significantly improve the resolution of seismic data.

Capacity estimation of lithium-ion (Li-ion) rechargeable batteries with adequate accuracy based on a small amount of charge-discharge cycling data are challenging. This is because the cycle data may not account for the considerable cell-to-cell variability that occurs during the aging process. Collecting long-term cycle data from numerous cells, on the other hand, is a costly and time-consuming operation in real-world applications. This article presents a semi-parametric Adaptive transfer learning method based on Gaussian process regression (AT-GPR) for assessing cell-level capacity despite only having access to a small dataset. It could be used to adapt transfer learning by automatically evaluating the similarity between the source and target tasks. Experimental results indicate that the proposed AT-GPR capacity estimation model may produce reliable prediction results, although the training data only accounts for 20% of the total dataset.

Seismic fault detection has become an important research topic in geo-science. Semblance-based coherence algorithm is widely used to detect seismic faults, folds, and fractures. However, the algorithm is computationally expensive on Central Processing Unit (CPU) when the seismic datasets are too large. Also, existing commercial geoscience software solutions use serial or batch processing modes using CPU-based computation which leads to a long execution time. In this paper, we present a fast and parallel implementation of semblance algorithm using General Purpose Graphical Processing Unit (GPGPU) powered with Compute Unified Device Architecture (CUDA). This is accomplished with a parallel kernel map of the algorithm through multiple threads. We also adopted a strategy for efficient memory occupancy and CPU-GPU communication with minimal latency. The algorithm is implemented on NVIDIA RTX 4000 GPU model with 8GB dedicated GPU memory and tested with Netherland F3 and Indian Krishna-Godavari (KG) Basin datasets. Our CUDA implementation achieved considerable speedup over its conventional CPU implementation on both datasets. Also, the proposed algorithm achieves faster times speedup are reported on both datasets over the commercial software OpenDtect. For extensive study, optimal runtime of the algorithm with variation of the parallel threads is also reported here.

Convolution Neural Network (CNN) is one of the powerful deep learning tools used in many computer vision tasks; however, it is still in a premature state while dealing with sensor data due to the unavailability of a large data set. Here in this paper, we present a CNN-based fall detection system. Deep convolution generative adversarial network (DCGAN) is used to synthesize more fall data before analysis. Sensor data along with synthesized data is prepossessed, and time-varying spectrograms are derived. We apply CNN on raw spectrogram images of different activities to categorize a fall from other activities. As a result, we successfully attain a classification accuracy of 97.2%, surpassing the conventional machine learning methods and previous works on the same data by a large margin.

Seismic fault analysis is one of the most critical aspects of the oil and natural gas industries. Many crucial decisions like borehole drilling and exploration are taken based on the presence of a seismic fault. Manually identifying faults is an old and time taking method. However, many new methods have been developed in the recent past that automate seismic fault detection. Convolutional neural network (CNN) is the most used method among them. In this paper, we propose an approach for training CNNs using seismic attributes and data augmentation. A mixture of synthetic and real seismic data is used to create the training and testing datasets. Additionally, we augment training data to increase diversity. We consider three seismic attributes: gradient structure tensor (GST) based coherence, semblance based coherence, and local discontinuity. Then we extract 2D patches, which act as input to CNN.

Spectral decomposition helps the geophysicists in enhancing the data interpretation as certain geological features may get highlighted at a particular frequency. The transformation of 1-D seismic trace into the corresponding frequency components gives a better analysis of the stratigraphy in the subsurface. We propose the multispectral coherence approach to delineate the stratigraphic features such as faults. First, the data are spectrally decomposed using continuous wavelet transform as well as a recently developed synchrosqueezing wavelet transform. Once all the seismic traces are spectrally decomposed with a band of frequencies, we propose a modified spectral balancing technique that enhances the resolution of seismic data. The spectrally balanced data are then subjected to time–frequency (T-F) analysis, which results in multispectrum data of corresponding frequency of certain bandwidth. Gradient structure tensor-based coherence is applied on spectrally balanced data as well as selected frequency bands of T-F data as the stratigraphic features may get highlighted in a certain frequency band. Finally, all the coherence images are statistically fused using a weighted mean to get finer and sharper fault lines with very little noise. This proposed method helps to better visualize all the possible subtle and minor faults present in the data. Experimental results on field seismic data show that subtle and minor faults are more apparent and discernible using the proposed method.

The goal of lithofacies classification is to predict a facies profile at the well site by using the values of rock parameters obtained or computed in well log analysis. Knowledge of lithofacies profiles is required to map the depositional environment of the subsurfaces, from which hydrocarbon zones may be discovered. Statistical techniques give the most feasible categorization of lithological facies across the borehole by maximizing a model, which can predict the chance of a group of rock samples belonging to a specific class. The majority of existing algorithms categorize each sample in the well log separately and classify the different classes. However, mainly, these techniques failed for the complex environments, where the actual data is very imbalanced. In this work, this issue is addressed by using the multistage support vector machine (SVM) classification algorithm with modified regularization parameters for each class. In our case, minority classes are getting worse accuracy than majority classes, which is improved by applying the higher penalty parameter to the minority classes. The method is applied to the Krishna-Godavari basin dataset where seven lithofacies are identified. The entire statistical methodology allows us to transmit the ambiguity from data collected at the well site to the estimated facies profiles. The method’s benefits are validated by the field data application results. This contribution provides a useful method for lithofacies classification and reservoir projection using SVM.

Electrical problems such as aging of electrical conductors and loose connections lead to arc fault in the electrical systems. Arcing temperature approaches up to 1000 °C which melts the conductors and leads to electrical fire accidents. Due to arcing, the current in the path gets reduced and thus conventional protection circuit may not be triggered. Thus, there is a need for designing alternate low-cost reliable arc detectors to detect and report the electrical arcing. Therefore, a series arc detection system is necessary to the low voltage distribution system (LVDS) for the reliable and efficient operation of the LVDS. In this paper, empirical mode decomposition (EMD) and support vector machine (SVM) based series arc detection in LVDS has been proposed. The arc faults have been generated in the lab with an experimental setup. When arc faults are generated, the conducted electromagnetic radiation (EMR) is present in the cable. The conducted EMR as an arc fault signature is acquired with the current transformer (CT) and digital storage oscilloscope (DSO). The arc fault signature has been extracted by the EMD analysis and, with the help of the SVM classifier, the arc fault has been detected.

Certainty based reduced sparse solution (CARSS) reduces the solution space of the source localization problem to the most certainly active sources. CARSS estimates the sources by utilizing (i) the extrema and the source distribution in the measurement vector, and (ii) the neighborhood continuity among measurement channels and sources. The near active sources will be interfering with each other causing the shift of extrema. The interference of the sources may eliminate the active source. In this paper, (1) The method is extended to densely and less sparse problems. (2) The reformulation of the problem to the mixed norms is proposed. (3) The methodology to de-interfere the interfered source signatures is proposed. The extension is found to be robust to sparsity as well as to the unknown additive noise.

To study the effect of total sleep deprivation on brain activity using EEG. Methodology: In this paper, the EEG of sixteen subjects (Mean = 21.5; SD = ±2.5) is recorded just after they woke up from sleep. Then, they are sleep deprived for the next twenty-four hours, and the EEG is recorded after sleep deprivation. The raw EEG data is pre-processed. The source localization is performed by a recently developed method- CARSS and sLORETA.Results: (1) The prominence of the right lobe increased from pre to post. (2) The increase in frontal source activation powers and a decrease in parietal is observed from the pre to the post (3) The δ, θ, and α powers increased from pre to post in pre-frontal leads. The left frontal lead showed an increase in α, but a decrease in the right frontal lead is observed. Conclusions: The changes in the source activations and the powers show that there is a change in the way in resting-state neural networks post sleep deprivation.

This paper presents a stochastic optimization approach to quickly and accurately estimate the model parameters such as amplitude, phase, frequency, and damping factor for the fault signals generated in an underground distribution cable. The CUMSUM algorithm is implemented for change detection. As soon as the change is detected, the parameter can be estimated for monitoring the system. In addition, Adaptive gradient descent (ADAGRAD), root mean square prop (RMSPROP), and adaptive moment estimation (ADAM) have been implemented to estimate the parameters of several instants of fault currents taken at the sending end of the underground cables. Further, the estimated parameters of simulated signals have been validated by conducting an experiment in the laboratory. A brief comparison has been presented among all the mentioned algorithms, which demonstrates the superior performance of the ADAM method.

Lie-detection test is fast becoming a defacto standard for the police department to testify the accused. Till now polygraph test remains the gold standard for the purpose. This test is considered to be invasive, time consuming and requires an expert opinion. In this paper we propose alternative non invasive methods based on portable and hidden thermal imaging to detect deceit and falsehood. The major constraint in this research is the availability of proper database emulating deceit. In the present study a real life database has been created using a mock experiment on human subjects. The experiment has been conducted at a government hospital where the subjects have been invited under the plea of health checkup. The subjects are asked to be alone and have been tempted to pick up currency notes appearing to be carelessly dropped on the floor. Subsequently during interrogation some of them lied. During this interrogation their facial thermal video have been captured with a hidden thermal camera. These experiments have been conducted with prior approval of the ethical committee of the hospital. The temperature of the forehead and periorbital region are analyzed. It is observed that there are remarkable changes in the temperature profile in these two areas. Different models have been used to find out the difference between thermal video sequence for persons lying and not lying.

Lithofacies extraction is an essential step for mapping the earth’s subsurface, which is crucial for effective hydro-carbon extraction. The identification of lithofacies in borehole location is frequently handled as a part of the core data analysis. The exact knowledge of the various lithological units is also provided by the downhole well log data. In our present work, we use the Walsh low pass filter to the wire-line log signal responses and the bed boundary identification algorithm to spot the bed boundaries at their corresponding depth with an automated approach. Initially, we construct a stepped version of the well log data using the Walsh domain filter with a fixed step width. Afterwards, each version of the well log data is given a different weight according to their influence on lithological change. The fluctuation at a particular location is compared with a reference value to generate an automatic set of lithological boundaries. Within a complicated environment of sedimentary strata, the Walsh low pass filtering can correctly locate the thin lithological units. Further, the proposed work is an efficient way of understanding the subsurface inhomogeneity and identification of the lithofacies boundaries. The approach is tested successfully on the well log data obtained from the Krishna-Godavari basin.

In this paper, an automatic multi-stage seismic fault detection framework is introduced. Initially, the seismic data is pre-processed using structure oriented anisotropic diffusion filtering. Next, we propose to estimate a target seismic trace using a linear combination of neighborhood traces. For improved estimation, the neighborhood traces are augmented. The squared residual trace between the target trace and the estimated trace contains the required fault information. We refer to this error trace as the augmented collaborative representation based fault extractor. At the fault location, the error magnitude is larger than usual. This helps in localization of seismic faults. Finally, the fault path is estimated using Hough transform. The performance of the proposed framework is evaluated on both complex synthetic and real-time datasets. The superior performance of proposed fault extractor over the existing algorithm is also demonstrated here.

The paper aims at proposing a non-invasive method for monitoring of the quality determining factors of Indian mangoes. The framework shows an IoT architecture in global mango cold chain using edge computing. Positioned at the inside of the reefer container, the system performs the task of sampling physio-chemical data from its integrated sensors and records the dynamic changes in continuous data streams. These records are treated in batches and incase of sensor events, the device triggers an alarm to a remote administration system hosted in Google cloud. GSM (during land transport) and on board Wi-Fi (during sea transport) are used as communication protocols during the phases of the journey. The data generated from the devices have been calibrated with reference to a standard Industry grade meter and compared with other similar systems.

In affective computing, identifying the true emotion of an individual is still a significant concern. Under real-life conditions, muscle movements are found to be unreliable to identify the behavior of a person. Researchers have used thermal modality to recognize real emotions; however, features used were originally handcrafted for visible modality and were directly adopted for thermal modality. Since visible and thermal images are built with different principles and have distinct characteristics, adopted features do not perform well. This paper presents an algorithm to classify six basic emotions from thermal facial images. The primary aim is to find the thermal modality-specific filters using a subset of the NVIE dataset. For this, an optimal set of local region-specific filters are generated using convolutional sparse coding. The optimal set of filters is used for feature extraction in which the idea of a histogram-based feature descriptor known as binarized statistical image features (BSIF) is used. Further, a supervised dimensionality reduction algorithm acknowledging the correlation between classes is employed based on the idea of discriminant correlation analysis (DCA). Finally, six emotions are classified using a linear support vector machine (SVM). The improved accuracy validates the performance of the proposed method with respect to previous works.

Intensity estimation of genuine emotion is a challenge for an inexpressive face or deceiving emotion. Thermal modality is experimentally seen to have the capability to reflect true emotion. However, emotion intensity in thermal images is not studied much due to the lack of an annotated database. In this paper, we propose a thermal domain-specific feature-based approach to estimate the intensity of emotion sequence images. The method uses labeled apex images of all six emotions from the NVIE dataset. For extracting domain-specific features, firstly, linear filters are learned from apex images using convolutional sparse coding. Then we obtain features using those learned filters. Further, a distance-based emotion intensity method is proposed without any knowledge of the actual intensities. For a specific emotion, this is done by finding the distance between features of sequence images and clusters of apex features. Proper clustering is ensured by using a supervised dimensionality reduction method and further verifying using SVM classification. Also, a way is suggested to calculate the error in the absence of the actual intensities. The experimental results on the standard NVIE dataset validate better performance compared to previous methods.

In this paper, we propose a fault extraction method using curvature attribute and log -Gabor filter with the help of saliency technique, which is the improvement of the dip, edge, and azimuth attributes. In these attributes, sometimes pieces of information are confusing so interpretation can be diverted from the aim. Different types of curvatures are calculated, but the most positive and most negative curvatures are taken into account because these are more sensitive to the edges. These curvature attributes ascertain the fault lineaments, which are easy to contain within the seismic section. The visual saliency approach is incorporated for probable fault point highlighting on the curvature attribute rather than conventional coherence techniques. Further, the log-Gabor filter is used for the enhancement of the fault edges. The proposed approach results in fault detection in the seismic volume. Experimental results validate the robustness of the algorithm.

This paper presents new image descriptors for heterogeneous face recognition (HFR). The proposed descriptors combine directional and neighborhood information using a rotating spoke and concentric rings concept. We name the descriptors as multi‐directional local adjacency descriptors (MDLAD). This family of descriptor captures the directional information through successive rotations of a pair of orthogonal spokes. Likewise, they capture the adjacency information through a comparison against the central pixel of a window with concentric rings around the central pixel. The MDLAD is found to describe the face images well for recognition purposes, which when matched using the chi‐squared distance. The face recognition performance with MDLAD improves with its use as a layer in a deep neural network, which yields a robust classification for heterogeneous face recognition with respect to the state‐of‐the‐art methods. The MDLADNET deep network is easily trainable with few hyperparameters and limited data samples as compared to existing similar deep networks. We have experimented on different heterogeneous modalities viz. Extended Yale B, CASIA, CUFSF, IIITD, LFW, Multi‐PIE, and CARL, and have found proficient results.

In this paper, the focus is on the analysis of a scalp electroencephalography (EEG) database of human subjects using the electrophysiological source imaging or source localization and the classification of normal and sleep-deprived states. The EEG collection was carried out while the subjects were driving in simulated condition in a laboratory, where the fatigue level propagates through 11 different stages of fatigue, to achieve the sleep deprivation of a total period of 36 h. Standardized low-resolution brain electromagnetic tomography (sLORETA) algorithm has been used here for estimating the source activations on the surface of the neo-cortex. sLORETA transforms the surface or scalp EEG data to the corresponding corticular dipole sources at each voxel on a simulated neo-cortex. For the characterization of the underlying neural patterns, approximate and sample entropies in voxels nearest to specific electrodes for different subjects and varying fatigue levels have been computed. Approximate entropy, sample entropy, and modified sample entropy are used here as the measures of complexity, similarity, and regularity in the sources. As a further investigation, these measures computed over all the stages are used to train a support vector machine, which classifies the measured values between alert and extremely fatigued states. As a result, several observations on the nature of change of the chaotic entropies are provided, and up to 86% classification accuracy is obtained.

Recently, near-field source localization has received more attention due to its capability of estimating the source direction of arrival (DOA) as well as the distance of the source from sensors. In the near-field, the waves are spherical and form a second-order model. A two dimensional (2D) method such as 2D MUSIC is used to solve this second-order problem. Theoretically, the 2D MUSIC has infinite resolution, but practically its resolution is limited and depends on various parameters. In this paper, we have investigated the resolution limit of 2D MUSIC. It has been seen that for closely spaced sources, the resolvability of the MUSIC depends upon the parameters such as the number of sensors, and the distance between the sensors. Mathematically this relationship is derived based on the orthogonality property of the autocorrelation matrix's eigenvectors. Simulation results varying above mentioned parameters validate the mathematical analysis. This study is important for practical implementation, where closely spaced sources are present.

Insulated gate bipolar transistors (IGBT) are used broadly in DC power transmission, power converters, and drives. These applications are cost-sensitive and require high module reliability. Researchers have been found that IGBT degradation over time is directly dependent on its junction temperature. Normally the temperature is measured with a temperature sensor close to the IGBT case or module. This requires the predesigning of the converter or results in a cost-effective and inconvenient process. In this paper, we are proposing a non-invasive method of junction temperature extraction, which do not suffer from the difficulties mentioned above. The method is based on the analysis of electromagnetic radiation (EMR) generated by the IGBT. It has been found that the radiation is a function of switching delay of the IGBT, and this delay is directly proportional to the junction temperature. Thus the junction temperature of IGBT is extracted from the EMR. The method requires only a single loop antenna to capture the radiation. Experimentally to increase the junction temperature of IGBT, an accelerated aging method is adopted. Then the junction temperature, the delay time, and the electromagnetic radiations are measured. The junction temperature is extracted from the inverse relation of these three parameters.

In this paper, we propose a fault detection method using a multiscale fusion of dip attribute in seismic dataset with the help of Hough transform and unconstrained optimization. Dip is the geometric attribute which corresponds to the dip of the seismic events. First, the dip is estimated by complex trace analysis and then it is multiscale analyzed using Gaussian or Laplacian pyramid to enhance the geometric dip and reduce noise. Dip attribute is calculated at each level of the pyramid and is fused using statistical properties such as mean, median and weighted mean. Followed by dip estimation, semblance attribute is calculated to accentuate likely fault points which involves local multiscale fused dip information. Discontinuity map is calculated from the semblance attribute and Hough transform is applied after thresholding the discontinuity map to extract faults. Finally with the information of local maximum in the discontinuity map and initial detected fault lines, an optimized version of fault is calculated using unconstrained optimization with brute force algorithm. Illustration of experimental results validates the robustness of the proposed algorithm on various seismic sections.

With the knowledge of adverse effects of harmonics in power systems, parameters (amplitude, frequency, phase, and harmonic contents) estimation of a signal have attracted a rising interest. In recent years sparse representation of a signal has shown a growing interest over the existing techniques that follow Shannon/Nyquist sampling theorem, which requires large storage space, huge computational time, and moreover, the overall process is cost-effective. Keeping in mind the flaws of the conventional methods, this paper presents a novel sparse power system domain based cubature Kalman filter (SPSD-CKF) algorithm, which is the completely new framework that exploits the orthogonal matching pursuit (OMP), a sparse coding algorithm, and K-SVD, a learned dictionary, for sparse representation of a signal. The K-SVD contains the prototype signal-atoms where the signals are described by sparse linear combinations of these atoms, is flexible one and can work with any pursuit method. On the other hand, the OMP is easy to implement and takes less time. The CKF is utilized to estimate the amplitude, phase, frequency, and harmonics using fewer measurements in the sparse domain. To examine the efficacy of learning-based dictionary obtained using the K-SVD, some well-known static dictionaries such as Gabor dictionary (GD) and an overcomplete hybrid dictionary (OHD) have been adopted and their results are compared. Various simulation results suggest that the proposed algorithm provides an efficient mechanism to estimate the parameters and also robust against the noise.

This paper proposes a novel, faster unmixing approach based on a convex geometric approach and slow feature analysis, and extends it to perform efficient library pruning based semi-blind unmixing. The former algorithm performs complete blind unmixing, whereas the subsequent algorithm performs exact library pruning for semi-blind unmixing. Slow feature analysis algorithm impels the pure pixels towards the exterior region. The work identifies the endmembers by detecting the extreme points. On the other hand, the proposed dictionary pruning method augments each library element with the data, extracts the extreme points and calculate a volume of the transformed data. The augmentation of actual image endmember changes the structure of the simplex. The library pruning method proposes an index to capture the change in the volume of the simplex and identify the actual image endmember. We evaluated unmixing performance as well as the runtime on real images, which ratifies the computational edge and proficiency of our proposed method.

In this paper, Sparse layer inversion using linear programming (LP) approach is introduced as an improvement to the Basis Pursuit Decomposition (BPD) method. In the BPD method, the seismic trace is represented as a superposition of dictionary patterns. A dictionary is built using the functions of odd and even reflection coefficients which represent a reflectivity series. The seismic data is reconstructed as the convolution of a seismic wavelet and the reflectivity series. For this convolution based model, the model parameters are estimated by minimizing a loss function which consists of the L2 norm of error and an L1 norm of model parameters. However, in this paper, it is suggested to replace the L2 norm of error with that of L1 norm. These L1 norms of error and the model parameters are rewritten as the inequality constrained optimization problems and this is solved using linear programming approach. Application of this method on a real data set from onshore of the Western part of India shows improved resolution of thin layers in seismic sections.

In this paper, we introduce an online facial expression recognition (FER) model, which infers the emotional states in real time. This model enables the computer to interact more intelligently with the user. Our proposed mechanism identifies the frontal face along with the region of interest (ROI), extracts discriminating features from suitable facial landmarks, and classifies the facial expressions. Histogram of oriented gradient (HOG) is implemented to extract features and facial landmark positions from active facial regions, which enhances the system performance against all the possible scale and pose variations. The system speed improved with appropriate integration of detection and tracking algorithms. Further, support vector machine (SVM) classifier is used to classify the detected face into neutral or six universal emotions. For achieving the best results with new user faces, the system extracts the neutral features of the user during the time of execution and uses them to train the classifiers. To validate the performance of the proposed algorithm is validated using CK+ and RafD databases.

Band selection of hyperspectral images is an optimal feature selection method, which aims at reducing the computational burden associated with processing the whole data. The significant and informative bands identified by the band selection process lead to efficient, compact representation of the image data and produce a satisfactory performance in the succeeding applications viz. classification, unmixing, target detection and so on. In this study, the authors present an unsupervised manifold clustering approach for band selection, which accounts for different types of scenarios. Unlike other band selection approaches, the authors’ proposed manifold clustering framework identifies the informative bands by utilising the interrelation between the bands and accounts for the multi‐manifold structure prevalent in some real images. The proposed band selection framework identifies the optimal number of clusters by cluster validity index, clusters the bands by manifold clustering and select representative bands from each cluster according to graph weight. Their proposed manifold clustering approach is a generic clustering approach, which produces a satisfactory result even when the data contains non‐linearity. The information theoretic performance measures, classification and unmixing performance on real image experiments demonstrate the proficiency of their proposed band selection algorithm.

Availability of a large number of application‐specific spectral libraries has generated a great deal of interest in semi‐blind unmixing of the hyperspectral image in both remote sensing and signal processing community. This study presents a novel, semi‐supervised, parameter‐free algorithm which employs sparsity measures for library pruning. The overall algorithm includes sparsity criteria based library pruning and sparse inversion method for abundance computation. In the pruning process, each library element is removed from the spectral library and the corresponding sparse abundance matrix is computed. The library elements which lead to higher sparsity are adjudged as image endmembers, based on the assumption that elimination of actual image endmember enhances sparsity level. The authors also present a detailed exploration of standard sparsity measures. They calculate the abundance of the pruned library by maximising Gini index or pq ‐norm sparsity, which satisfies the desirable sparsity properties and is easier to compute. The abundance calculation task is solved using the adaptive direction method of multipliers. The experimental results on several real and synthetic image datasets demonstrate the computational efficiency and proficiency the authors’ method in the presence of noise and highly coherent spectral library.

Hyperspectral sparse unmixing methods estimate abundance of endmembers, assuming spectral library as an overcomplete set of endmembers. In this letter, we present a novel, fast and efficient dictionary pruning approach for hyperspectral unmixing. We quantify the change in the latent structure of data due to augmentation of spectral library element using covariance similarity measure. Since the covariance matrices form a nonlinear manifold, choosing an appropriate similarity measure is a nontrivial task. We explored prevalent similarity measures, which motivated us to employ Jeffrey's Kullback-Leibler divergence measure due to its tighter bounds and better noise performance. We also present analytical formulations for faster implementation of covariance similarity. We evaluate the performance of dictionary pruning algorithms on several synthetic and real hyperspectral images and demonstrate the proficiency of our proposed work in diverse scenarios.

Dimensionality reduction (DR) methods have attracted extensive attention to provide discriminative information and reduce the computational burden of hyperspectral image (HSI) classification. However, the DR methods face many challenges due to limited training samples with high-dimensional spectra. To address this issue, a graph-based spatial and spectral regularized local scaling cut (SSRLSC) for DR of HSI data is proposed. The underlying idea of the proposed method is to utilize the information from both the spectral and spatial domains to achieve better classification accuracy than its spectral domain counterpart. In SSRLSC, a guided filter is initially used to smoothen and homogenize the pixels of the HSI data in order to preserve the pixel consistency. This is followed by generation of between-class and within-class dissimilarity matrices in both spectral and spatial domains by regularized local scaling cut and neighboring pixel local scaling cut, respectively. Finally, we obtain the projection matrix by optimizing the updated spatial-spectral between-class and total-class dissimilarity. The effectiveness of the proposed DR algorithm is illustrated with two popular real-world HSI data sets.

In high-stake situations, the micro-expressions reveal the hidden emotions of a person and it has potential applications in many areas. The recognition of such short-lived subtle expressions is a challenging task. The literature proposes several spatio-temporal features to encode the subtle changes on the face during a micro-expression. The spatial changes are almost indistinguishable as the facial appearance does not change appreciably. However, these changes possess a temporal pattern. This paper explores the temporal features associated with facial micro-movements and proposes fuzzy histogram of optical flow orientation (FHOFO) features for recognition of micro-expressions. The FHOFO constructs suitable angular histograms from optical flow vector orientations using histogram fuzzification to encode the temporal pattern for classifying the micro-expressions. We have also discussed the effect of inclusion and exclusion of the motion magnitudes during FHOFO feature extraction. It has been demonstrated by repeated experiments on the publicly available databases, that the performance of FHOFO is consistent and close or at times even better than the state-of-art techniques.

Hyperspectral images (HSIs) contain a wealth of information over hundreds of contiguous spectral bands, making it possible to classify materials through subtle spectral discrepancies. However, the classification of this rich spectral information is accompanied by the challenges like high dimensionality, singularity, limited training samples, lack of labeled data samples, heteroscedasticity, and nonlinearity. To address these challenges, we propose a semisupervised graph-based dimensionality reduction (DR) method named “semisupervised spatial spectral regularized manifold local scaling cut” (S3RMLSC). The underlying idea of the proposed method is to exploit the limited labeled information from both the spectral and spatial domains along with the abundant unlabeled samples to facilitate the classification task by retaining the original distribution of the data. In S3RMLSC, a hierarchical guided filter is initially used to smoothen the pixels of the HSI data to preserve the spatial pixel consistency. This step is followed by the construction of linear patches from the nonlinear manifold by using the maximal linear patch criterion. Then, the interpatch and intrapatch dissimilarity matrices are constructed in both spectral and spatial domains by RMLSC and neighboring pixel MLSC, respectively. Finally, we obtain the projection matrix by optimizing the updated semisupervised spatial-spectral between-patch and total-patch dissimilarity. The effectiveness of the proposed DR algorithm is illustrated with publicly available real-world HSI data sets.

This paper presents a smartphone-based system for the detection of drowsiness in automotive drivers. The proposed framework uses three-stage drowsiness detection. The first stage uses the percentage of eyelid closure (PERCLOS) obtained through images captured by the front camera with a modified eye state classification method. The system uses near infrared lighting for illuminating the face of the driver during night-driving. The second step uses the voiced to the unvoiced ratio obtained from the speech data from the microphone, in the event PERCLOS crosses the threshold. A final verification stage is used as a touch response within a stipulated time to declare the driver as drowsy and subsequently sound an alarm. The device maintains a log file of the periodic events of the metrics along with the corresponding GPS coordinates. The system has three advantages over existing drowsiness detection systems. First, the three-stage verification process makes the system more reliable. The second advantage is its implementation on an Android smart-phone, which is readily available to most drivers or cab owners as compared to other general purpose embedded platforms. The third advantage is the use of SMS service to inform the control room as well as the passenger regarding the loss of attention of the driver. The framework provides 93.33% drowsiness state classification as compared to a single stage which gives 86.66%.

In present academic system, regular class attendance of students' plays a significant role in performance assessment and quality monitoring. The conventional methods practised in most of the institutions are by calling names or signing on papers, which is highly time-consuming and insecure. This article presents the automatic attendance management system for convenience or data reliability. The system is developed by the integration of ubiquitous components to make a portable device for managing the students' attendance using Face Recognition technology.

Rich spectral information of hyperspectral images provides a non-invasive way to characterize the skin tissues and thereby improves hyperspectral face recognition accuracy. However, the increased computational complexity is reduced by efficient feature selection method. In this paper, we amalgamate pixel selection with spectral discrimination. The pixel selection process choses the informative pixels, which improves the computational performance, whereas, covariance similarity encompasses the complete spectral information. We compare the covariance matrices formed from the selected pixels obtained by fiducial points and edge. A detailed study of the covariance similarity measures has been conducted. This leads us to use Jeffrey's KL divergence measure because of its tighter bounds and better noise robustness. We have evaluated our proposed framework on two popular hyperspectral face recognition datasets.

Fast and accurate estimation of harmonics of a typical power system signal is very much desirable for power quality assessment. This paper proposes an application of the cubature Kalman filter (CKF)for estimating the parameters of harmonics, sub-harmonics, and inter-harmonics of a signal in presence of noise. CKF utilizes a third-degree spherical radial cubature rule to estimate the probability density functions of the states as well as the measurements. At the same time, this technique does not require any linearization and saves the computational time. The effectiveness of CKF has been compared with UKF by performing various test cases. It is observed from the simulation results that CKF exhibits superior performance in estimating the parameters of harmonics, inter-harmonics, and sub-harmonics of a distorted power system static as well as the dynamic signal by virtue of execution time and accuracy.

This paper proposes a pre-processing scheme based on variational mode decomposition (VMD) to regularize target lithological log from seismic signals for reservoir characterization (RC) using support vector regression (SVR) algorithm. The regularization scheme has been incorporated by a systematic framework on a real hydrocarbon field dataset from India. The performance of SVR algorithm is quantified in terms of four metrics -correlation coefficient (CC), root mean square error (RMSE), absolute error mean (AEM), and scatter index (SI). The tuned SVR parameters are used to predict synthetic porosity logs over the study area from three-dimensional (3D) seismic volumes. The predicted porosity logs are spatially filtered using a two-dimensional (2D) median filter for better visualization. The designed framework can be used for RC in future to predict and visualize lithological properties from seismic attributes over a hydrocarbon field.

The lack of proper class discrimination among the hyperspectral (HS) data points poses a potential challenge in HS classification. To address this issue, this letter proposes an optimal geometry-aware transformation for enhancing the classification accuracy. The underlying idea of this method is to obtain a linear projection matrix by solving a nonlinear objective function based on the intrinsic geometrical structure of the data. The objective function is constructed to quantify the discrimination between the points from dissimilar classes on the projected data space. Then, the obtained projection matrix is used to linearly map the data to more discriminative space. The effectiveness of the proposed transformation is illustrated with three benchmark real-world HS data sets. The experiments reveal that the classification and dimensionality reduction methods on the projected discriminative space outperform their counterpart in the original space.

In this paper, we introduce a system named Portable Personality Recognizer (PPR), which classifies the personality of an individual using his/her transitions of affective states. This work attempts to reveal the latent relationship between emotions and personality of a person. Here, we train a hidden Markov model (HMM) with observable emotional states viz. Happiness (H), Anger (A), Surprise (S) and Disgust (D) and the hidden traits viz. Psychoticism (P), Extraversion (E) and Neuroticism (N). Based on the model, the system estimates the personality as Psychotic, Extravert or Neurotic. It does so by capturing the facial images of an individual using a visible and a thermal camera to decide the present affective state of the person. The emotion classification is carried out using fused eigenfeatures from the visible and blood perfused thermal images. The emotional state changes are observed using the trained HMM to estimate the personality. The proposed hardware prototype consists of a Banana Pi board with a seven inch LCD screen having a thermal and a visible camera add-ons. The system achieves an emotion classification accuracy of 87.145 percent, while an accuracy of 87.87 percent is achieved for personality recognition.

In social interactions, Aggression is a behavior towards another individual with the motive of causing physical or psychological damage. In humans, anger or disgust due to obstructs in achieving certain goals causes aggression. This article proposes an automatic method for detection of aggression using features extracted from pressure distribution in vocal tract during voiced speech. The variations in air pressure distribution across different sections of vocal tract have been computed from the speech signal by inverse estimation. A Hidden Markov Model has been trained to classify these air pressure variations as Aggression or Calm. The system has been tested on a set of audio clips extracted from interviews of political personalities on television shows. A total of 120 audio clips with an overall duration of around 30 minutes were collected from three different speakers based on human perception of Aggression and Calm. The clips were rated by human raters to discard perceptually ambiguous clips. The speaker dependent system was trained using 40 percent of the data of each speaker and was tested with the remaining 60 percent of the data of the same speaker. The system was able to detect aggression with about 93 percent accuracy.

The metric learning problem is concerned with learning a distance function tuned to a particular task, and has been shown to be useful when used in conjunction with nearest-neighbor methods and other techniques that rely on distances or similarities. This paper proposes an ensemble learning technique which combines the efforts of multiple metric learning algorithms like Large Margin Nearest Neighbours (LMNN), Local Fisher Discriminant Analysis (LFDA), Logistic Discriminant Metric Learning (LDML) and a few others to solve the problem of face recognition. In the ensemble learning technique, we propose and study 4 kinds of weighting schemes, namely (1) hard voting, (2) equally weighted soft voting, (3) adaptive soft weighting, and (4) decision tree/neural network based soft voting. In this paper, we present our results compared to Support Vector Machines (SVMs). Experiments show that our proposed method attains state-of-the-art results on the challenging Labeled Faces in the Wild (LFW) dataset [1].

The paper proposes a novel method for health monitoring and early stage fault diagnosis of IGBT by analyzing the electromagnetic radiation (EMR) pattern. The main cause behind IGBT failure is the evolving stress due to thermal cycle. The change in the operating characteristic of IGBT affects the EMR signature. The paper establishes the relation between different IGBT characteristic with EMR pattern. A standard near-field antenna has been used to capture the EMR near the power circuit board housing the IGBT circuitry. Electrical overstress method has been adopted for accelerated aging of the IGBT. Experimental results demonstrate the effectiveness of the proposed method to identify faults, aging and thermal stress.

Rampant power theft at the low voltage consumer end is a growing concern for the power distribution companies. This paper proposes an effective method for detection of power theft at low voltage consumer end. The proposed method is designed to reliably detect hooking in service line cable and bypassing of electric energy meter. In this method, a low magnitude, high-frequency, non-interfering signal has been injected into the power line. Two LC traps have been designed and placed on either side of the energy meter to restrict the flow of the injected component from reaching the load end. In either case of bypassing or hooking, the power of the high-frequency component will deviate from its value under normal operating condition. The proposed algorithm utilizes this fact for the power theft detection. In order to attain the purpose, power spectral density (PSD) coefficients of the acquired line current are evaluated, magnitude of the PSD coefficients corresponding to the injected frequency are identified and subsequently thresholding technique is applied on it for detection of power theft. The proposed algorithm has been validated in simulation environment as well as with real-world data. This algorithm can play a significant role in arresting power theft.

Automatic recognition of spontaneous facial expressions is a major challenge in the field of affective computing. Head rotation, face pose, illumination variation, occlusion etc. are the attributes that increase the complexity of recognition of spontaneous expressions in practical applications. Effective recognition of expressions depends significantly on the quality of the database used. Most well-known facial expression databases consist of posed expressions. However, currently there is a huge demand for spontaneous expression databases for the pragmatic implementation of the facial expression recognition algorithms. In this paper, we propose and establish a new facial expression database containing spontaneous expressions of both male and female participants of Indian origin. The database consists of 428 segmented video clips of the spontaneous facial expressions of 50 participants. In our experiment, emotions were induced among the participants by using emotional videos and simultaneously their self-ratings were collected for each experienced emotion. Facial expression clips were annotated carefully by four trained decoders, which were further validated by the nature of stimuli used and self-report of emotions. An extensive analysis was carried out on the database using several machine learning algorithms and the results are provided for future reference. Such a spontaneous database will help in the development and validation of algorithms for recognition of spontaneous expressions.

Iris centre (IC) localisation in low‐resolution visible images is a challenging problem in computer vision community due to noise, shadows, occlusions, pose variations, eye blinks etc. This study proposes an efficient method for determining IC in low‐resolution images in the visible spectrum. Even low‐cost consumer‐grade webcams can be used for gaze tracking without any additional hardware. A two‐stage algorithm is proposed for IC localisation. The proposed method uses geometrical characteristics of the eye. In the first stage, a fast convolution‐based approach is used for obtaining the coarse location of IC). The IC location is further refined in the second stage using boundary tracing and ellipse fitting. The algorithm has been evaluated in public databases such as BioID, Gi4E and is found to outperform the state‐of‐the‐art methods.

This paper presents a new technique for glottal inverse filtering using a distributed model of the vocal tract. A discrete state space model has been constructed for the speech production system by combining the concatenated tube model of the vocal tract and Liljencrants-Fant (LF) model of the glottal flow derivative waveform. An adaptive system identification technique, based on extended Kalman filtering, has been used for estimation of the states and model parameters from continuous speech. The glottal signal, represented by the LF model, is piecewise differentiable in one glottal cycle. Hence, the hybrid system has been characterized by separate models during two different modes. Multiple model estimation has been performed by switching between the two models at the mode jumps. The open phase of the glottal cycle has been considered as Mode 1; whereas, the return phase and closed phase combined has been taken as Mode 2. The starting point of Mode 1, also known as glottal opening instant, was estimated by observing formant modulation, which remains negligible during closed phase, and starts to increase at the onset of opening. The starting point of Mode 2, also known as the glottal closing instant, was computed by peak-picking from linear prediction (LP) residual signal. The proposed method estimates the glottal waveform as well as changes in flow occurring at different sections of the vocal tract during speech production. This technique has been found to be accurate and robust to variations in pitch as compared to other LP-based methods in the literature. The method also estimates the air pressure distribution at different sections of the vocal tract.

Detectability of closely spaced sinusoids in a noisy signal using MUltiple SIgnal Classifier (MUSIC) depends to a great extent on the sampling frequency (Fs) and the size of the autocorrelation matrix (N). Improper choice of any of these may result in increased computational burden and/or unresolved frequency components. This paper presents an analytical approach to determine expressions of lobe width using Fsand N at lobe base (Δfb) and half of the lobe height (Δfh). The required values of Fsand N can be derived from the expression of Δfbfor distortion-less lobe heights of two closely spaced sinusoids. A tighter bound can be found using the expression of only Δfhto resolve two distinct peaks. Probability of resolution using reciprocal of MUSIC peaks is determined for various N and it's limit for full resolvability was verified with the derived analytical expressions.

Extraction of discriminative features from salient facial patches plays a vital role in effective facial expression recognition. The accurate detection of facial landmarks improves the localization of the salient patches on face images. This paper proposes a novel framework for expression recognition by using appearance features of selected facial patches. A few prominent facial patches, depending on the position of facial landmarks, are extracted which are active during emotion elicitation. These active patches are further processed to obtain the salient patches which contain discriminative features for classification of each pair of expressions, thereby selecting different facial patches as salient for different pair of expression classes. One-against-one classification method is adopted using these features. In addition, an automated learning-free facial landmark detection technique has been proposed, which achieves similar performances as that of other state-of-art landmark detection methods, yet requires significantly less execution time. The proposed method is found to perform well consistently in different resolutions, hence, providing a solution for expression recognition in low resolution images. Experiments on CK+ and JAFFE facial expression databases show the effectiveness of the proposed system.

In oil exploration, studying the similarity between patterns of the same geophysical properties in different wells is essential for making early decisions on future planning as well as for assessing the lithology of the area under survey. Geoscientists either rely on visual tools or resort to correlation studies between the different wells to match portions of the well logs. This is a tedious process involving several trial and error runs, which includes shifting, stretching, and sometimes preprocessing of the well logs by experienced geoscientists. However, this can be simplified by automating the process of matching. The well logs, a measure of the lithology, fall under the class of nonlinear signals. Therefore, linear methods are inadequate for matching these sequences. In this letter, we introduce similarity measures based on the concept of synchronization as used in matching nonlinear signals such as chaotic time series data. Two recently proposed methods, i.e., synchronization likelihood (SL) and visibility graph similarity (VGS), have been applied on the gamma-ray and porosity logs along different wells. These are considered as depth sequences, which can also be converted to suitable time series with the availability of the velocity profile. The data for this study originate from 12 existing wells in the western coast of India. The values of SL and VGS as well as the correlation are computed between these wells. Higher values indicate the existence of similarities. This has also been verified from the overlapped plots of well-log data.

Onboard monitoring of the alertness level of an automotive driver has been challenging to research in transportation safety and management. In this paper, we propose a robust real-time embedded platform to monitor the loss of attention of the driver during day and night driving conditions. The percentage of eye closure has been used to indicate the alertness level. In this approach, the face is detected using Haar-like features and is tracked using a Kalman filter. The eyes are detected using principal component analysis during daytime and using the block local-binary-pattern features during nighttime. Finally, the eye state is classified as open or closed using support vector machines. In-plane and off-plane rotations of the driver's face have been compensated using affine transformation and perspective transformation, respectively. Compensation in illumination variation is carried out using bihistogram equalization. The algorithm has been cross-validated using brain signals and, finally, has been implemented on a single-board computer that has an Intel Atom processor with a 1.66-GHz clock, a random access memory of 1 GB, ×86 architecture, and a Windows-embedded XP operating system. The system is found to be robust under actual driving conditions.

This paper presents the functional interdependences among electroencephalograph (EEG) signals collected from human subjects undergoing a controlled experiment over a period of 36 h of sleep deprivation. The EEG signals were recorded from 19 electrodes spread all over the scalp. The interdependence among the signals was measured using synchronization likelihood (SL), which measures the dynamical (both linear and nonlinear) interdependence between two or more nonstationary time series. A network structure was evolved based on these SL values. The EEG signal being nonstationary, instead of the frequency bands, the connectivity was evaluated at various intrinsic modes known as intrinsic mode functions (IMFs). These IMFs were generated using empirical mode decomposition. It was observed that the connectivity of the networks exhibits definite patterns at specific IMFs with increase in sleep deprivation at successive stages of the experiment. The results were validated using subjective assessment and audiovisual response tests.

The state of brain and its rapid transition from one state to the other is responsible for various activities and cognitive functions. These brain states are the result of balanced coordination between integrating and segregating activities of different lobes through rhythmic oscillations. Such coordination has been studied in recent times through synchronization of EEG signals generated from different lobes. In this paper, the authors have considered Synchronization Likelihood (SL) to measure the synchronization or integration between the lobes. The synchronization information is stored in SL matrix and the principal components of an SL matrix have been used to represent the state of brain at any instant. Finally, the time series of weight vectors corresponding to the principal components of SL matrices at each time point has been used to classify different states of brain at different stages of a sleep deprived experiment.

This correspondence proposes a novel nonlinear adaptive algorithm named as filtered-s least mean square (FSLMS) algorithm for multichannel active control of nonlinear noise processes. A reduced complexity FSLMS algorithm using filter bank approach is also suggested. The performance of the proposed algorithm is validated through computer simulations for nonlinear noise processes. It is demonstrated that the proposed method outperforms the conventional filtered-x least mean square algorithm and second-order Volterra filtered-x LMS (VFXLMS) algorithm for control of nonlinear noise processes. Computational complexity analysis shows the proposed method involves lesser number of computations as compared to second-order VFXLMS algorithm