This work proposes a physics-guided unified deep learning architecture for single image restoration targeting dehazing, deraining, and low-light enhancement. The architecture first estimates the transmission map and airlight under an atmospheric scattering model, and then refines the result with a grayscale prior. A DiffConv feature extractor is proposed which combines vanilla and difference convolutions with a Laplacian branch (to capture high-frequency features). During inference, its branches are re-parameterized into a single kernel for reducing computational complexity. The grayscale prior replaces the Y channel in the YCbCr space to suppress noise and color artifacts, while a refinement stage uses Spatial Feature Transform (SFT) to inject structural features from this grayscale prior into the RGB domain. Experiments on standard benchmarks show consistent improvements in PSNR and SSIM at lower computational cost.

Drowsy driving is a leading cause of traffic accidents, often linked to delayed reaction times (RT). While EEG-based methods are effective in detecting drowsiness, they struggle with the lack of adaptive mechanisms for selecting optimal time windows that capture neural synchrony, and the limited generalizability due to trial-to-trial variability. This study proposes Adaptive Windowed Evolutionary Synchrony Analysis (AWESA), which uses evolutionary algorithms to dynamically optimize EEG time windows. AWESA combined with Convolutional-LSTM model achieves state-of-the-art performance in RT prediction on the Lane Keeping Task dataset, reducing the time window by a factor of 100. Despite the shorter window, AWESA delivers comparable prediction accuracy (MAE and RMSE), reducing complexity and offering a robust approach for enhancing drowsiness detection and for capturing relevant synchrony patterns for other neuro-cognitive tasks while drastically reducing complexity.

The Optimal Synchrony Window Detection (OSWD) method addresses the challenge of capturing dynamic temporal patterns in non-stationary EEG data for reaction time (RT) prediction. Traditional fixed-window approaches often fail to capture critical neural dynamics due to inter trial variability, leading to suboptimal performance in time-series modeling. OSWD introduces an adaptive window selection mechanism based on neural synchrony, dynamically optimizing time windows to capture task-relevant features that vary across trials and individuals. By integrating OSWD with Cov-CNN-LSTM architectures, the model achieves state-of-the-art accuracy while reducing computational complexity. Our experiments on the Lane Keeping Task dataset demonstrate that OSWD significantly enhances RT prediction by improving feature extraction and reducing prediction errors. This adaptability makes OSWD an effective solution for real-time EEG-based applications, with broader implications for time-series forecasting and signal processing.

Electroencephalogram (EEG) signal classification for neural activity monitoring and cognitive state assessment using Machine Learning (ML) models is a well-established strategy. The usage of ML models with traditional Euclidean vector-based feature extraction is efficient and effective if the behavior and distribution of the activity are normal. In specific abnormal behavioral disorders like depression, schizophrenia, and bipolar disorder, a dedicated non-euclidean manifold implementation can be the coherent strategy to handle the non-stationarity and underlying complexity. Riemannian manifold and Tangent space mapping is one of the efficient differentiable non-euclidean manifold techniques in analyzing higher dimensional data like EEG signals. To introspect the potential of Riemannian manifold learning in understanding neural behavior, this paper introduced the novel Geodesic Mean Threshold Channel (GMTC) selection scheme. The proposed GMTC framework has been validated with publicly available depression and schizophrenia datasets, which are collected with different specifications and strategies. Performance metrics of the proposed methodology have shown a 7% gain in accuracy with stabilized standard deviation.

Recent advances in affective state recognition have leveraged non-invasive electroencephalography (EEG) sensor signals to decode neural activity, where EEG-based biomarkers are critical for affective computing. While various deep learning models have been applied to classify affective states based on EEG sensor signals, challenges persist due to limited sample sizes, high inter-subject variability, and insufficient integration of cross-regional brain dynamics. To address this, we propose a persistent topology-driven graph learning framework for EEG sensor network analysis, exploiting temporal correlations between brain regions to model connectivity. By employing persistent homology, our method captures spatiotemporal dynamics through topological invariants, enabling robust representation of EEG sensor networks for affective states classification. Experimental validation on a publicly available GAMEEMO dataset demonstrates the proposed framework’s efficacy, achieving 98.73% average accuracy in binary classification tasks (positive and negative emotion) and 92.59% in multi-class classification (LALV, LAHV, HALV, and HAHV), and similarly, on FEEL dataset demonstrates enhanced classification performance outperforming existing state-of-the-art methods. This work advances affective computing by integrating topological data analysis with graph-based EEG interpretation, offering novel applications in human-computer interaction (HCI) and human-in-the-loop adaptive system studies.

Ambient factors such as fog, rain, and haze significantly degrade images, adversely impacting the performance of computer vision systems in autonomous vehicles. Numerous image restoration architectures have been developed to address this challenge, with local and non-local attention-based methods gaining significant traction for their promising results. However, existing approaches predominantly focus on either local or non-local attention mechanisms, which limits their ability to comprehensively restore the image quality. Additionally, nonlocal attention methods, while leveraging the self-similarity of natural images, often struggle to accurately model long-range dependencies due to excessive degradation and limited information in spatial domain. To overcome these limitations, this work proposes SFCola-Net, a novel spatial-frequency collaborative attention multiscale network that combines local and non-local features for effective image restoration. The proposed SFColaNet architecture restores the image features in both spatial and frequency domains, effectively handling areas with complex textures. Furthermore, patch-wise non-local attention model is proposed, enabling the network to capture long-range features. The proposed network demonstrates significant improvements across various image restoration tasks, including image dehazing, deraining, and low light enhancement, thereby enhancing the robustness and reliability of computer vision systems in autonomous vehicles.

Cervical cancer is a highly prevalent and clinically significant malignancy impacting women, with escalating incidence and mortality rates. Early diagnosis is paramount for enhancing patient prognoses. Recent breakthroughs in natural language processing (NLP), particularly transformer-based architectures, have exhibited remarkable efficacy in medical image classification, frequently surpassing conventional convolutional neural networks (CNNs). However, the challenges of interpretability and computational overhead persist, limiting their direct applicability in real-world clinical scenarios. This study introduces a novel lightweight InTranCeNet transformer architecture incorporating semantic specialization and an attention-distributed map to enhance medical image analysis. A convolutional token embedding with projection layers is devised to effectively encapsulate spatial context and localized feature representations while maintaining translation invariance. The semantic specialization mechanism is integrated within the multi-head self-attention (MHSA) framework, where distinct attention heads are explicitly optimized to capture critical pathological attributes. This approach reinforces feature discrimination by ensuring robust hierarchical extraction of diagnostically significant features. Deeper layers in the proposed model iteratively refine these specialized feature representations, thereby enhancing classification performance for cervical cancer images. The hierarchical feature aggregation mechanism systematically merges tokenized patches, reduces dimensionality, and extends the receptive field to efficiently capture long-range dependencies. Evaluations conducted on the two publicly available SIPaKMeD and Mobile-ODT datasets demonstrate the superiority of the proposed framework, achieving average classification accuracies of approximately 96.01% and 92%, and precisions of 97.58% and 95.46%, respectively, under a rigorous 5-fold cross-validation protocol. Comparative experimental analyses substantiate that the proposed model surpasses state-of-the-art deep learning methodologies by approximately 0.5 to 1%. Additionally, the attention-distributed map, enriched with semantic features, enhances model interpretability by elucidating the spatial allocation of attention across critical pathological regions, thereby facilitating transparent and interpretable decision-making in medical image classification.

Early and optimal identification of cardiac anomalies, especially Myocardial infarction (MCI) can aid the individual in obtaining prompt medical attention to mitigate the severity. Electrocardiogram (ECG) is a simple non-invasive physiological signal modality, that can be used to examine the electrical activity of heart tissue. Existing methods for MCI detection mostly rely on the temporal, frequency, and spatial domain analysis of the ECG signals. These conventional techniques lack in effective identification of cardiac cycle inter-dependency during diagnosis. Hence, there is an emerging need for incorporating the underlying connectivity of the intra-sessional cardiac cycles for improved anomaly detection. To address this gap, this article proposes a Topological Signal Processing (TSP) based novel framework for ECG signal analysis and classification. Persistent homological features have been extracted through Čech Complex generation with homotopy equivalence check. Homological features like persistent birth-death rates, Betti curves, and persistent entropy provide transparency of the regional and cardiac cycle connectivity when combined with Machine Learning (ML) models. The proposed framework is assessed using publicly available datasets (MIT-BIH and PTB), and the performance metrics of machine learning models indicate its efficacy in classifying Normal Sinus Rhythm (NSR), MCI, and non-MCI subjects, achieving a 2.8% mean improvement in AUC (area under the ROC curve) over existing approaches.

Usage of Machine Learning (ML) models has been trending for automated screening of mental health. Electroencephalogram (EEG) signals, due to their non-invasive nature and ease of availability with low cost, are mostly recorded and used for diagnosis. Such signals however are non-stationary and also lie on a nonlinear manifold. Therefore, ML models may then struggle to discover the underlying connectivities in neurological disorders diagnosis using EEG. Topological studies can help in this context. However, there have been limited studies conducted on the application of Topological Data Analysis (TDA) for the purpose of characterizing and classifying EEG data. For the very first time, a Semantic Topological Ensemble approach is proposed through Graph Mapping and Vietoris-Rips Complex (GM-VRC) framework to improve the robustness of TDA features for depression classification. The proposed framework is assessed with the publicly available datasets containing Healthy Controls (HC) and Major Depressive Disorder (MDD) subjects. By comparing the test accuracies of the traditional TDA features analysis with baseline Deep Neural Network (DNN) and Graph Neural Network (GNN), improved performance with a mean gain of 9% is noticed with the proposed GM-VRC framework.

Meditation methods, which have their origins in ancient traditions are gaining popularity as a result of their potential mental and physical health advantages. EEG neural correlates underlying enhanced cognitive abilities such as sustained attention and working memory need to be analyzed scrutinizingly to evaluate the effects of meditation practices. In this article, we thus provide an analysis of EEG features such as various band powers and connectivity based features to evaluate the meditation effects. Also, we provide a classification framework to classify the meditation states from the baseline EEG states. We report our results on an in house dataset of 20 participants(10 experienced and 10 novice) who underwent a two-week long mantra meditation practice. Strikingly we have found out that, as the novice participants practice meditation overtime, the accuracies of machine learning classification between the baseline EEG versus meditative EEG of the novice increase significantly, as an indication of their enhanced meditation experiences. Also, we found out that even such short and regular meditation practices, the cognitive abilities of novice meditators get enhanced which are evaluated through Brain-Based Intelligence Test (BBIT) psychometric tests and the results that got reflected in their EEG correlates.

Passive electroencephalogram (EEG) brain–computer interfaces (BCI) have common usage in the area of Driver Drowsiness Detection. The approach presented herein identifies the cognitive state of the user while no mental action is required. Data recorded in EEG-based BCI experiments are generally noisy, nonstationary, and contaminated with artifacts that can deteriorate any analyzer’s performance. Recently, common spatial patterns (CSPs) have been adapted with EEG state-space incorporating spatiospectral optimization using fuzzy time delay (FTD-CSSP). Temporal phase disparity sequence (TPDS) is used to measure synchrony between EEG signals. The output of Linear transforms operating on the TPDS constitute useful features for EEG regression problems. On similar lines, this article proposes spatiospectral optimized fuzzy-independent phase-locking value (SSO-FIPLV) representations (exploiting the spatiospectral information from TPDS) for EEG signals to monitor a user’s cognitive states. Specifically, we analyze changes in EEG synchronization for a car driver as she/he drifts between alert and drowsy states. We use neural networks (NNs) for prediction. This article also proposes a cutting-edge method for training NN using the Euler–Lagrangian formulation. A stability proof is provided for the intended training approach alongside, and the performance is corroborated on the EEG reaction time prediction task, both within and across subjects, using a publicly available dataset. The NN trained by the proposed approach performs better than other competitive approaches in terms of minimizing root-mean-squared error and maximizing correlation coefficient. Channelwise feature importance in terms of average relevance values calculated from NN feature representations is visualized in the form of Topoplots using layerwise relevance propagation for regression.

Neurons usually converse through electrochemical signals and pooled neuronal firings feasibly be recorded on the scalp through the medium of electroencephalogram (EEG). EEG waveforms are recorded, analysed and categorized across directives concerning a Brain-Computer Interface (BCI). Deteriorated signal to noise ratio and non-stationarities stand as a paramount obstacle in steady decoding of EEG. Appearance of non-stationarities across EEG patterns notably upset the feature waveforms thus worsening the functioning of detection block and as a whole the Brain Computer Interface. Stationary Subspace schemes bring to light subspaces within which data distribution persists stably over time. Current work focuses on the development of a novel spatial transform based feature extraction scheme to address nonstationarity in EEG signals recorded against a drowsiness detection problem (a machine learning regression scenario). The presented approach: F-DIV-IT-JAD-WS derived features distinctly surpassed DivOVR-FuzzyCSP-WS based standard features across RMSE and CC performance criteria pair. We construe that the propounded feature derivation approach based on F-DIV-IT-JAD-WS will usher a significant attention in researchers who are developing algorithms for signal processing, specifically, for BCI regression scenarios.

EEG signals can be processed and classified into commands for brain-computer interface (BCI). Stable deciphering of EEG is one of the leading challenges in BCI design owing to low signal to noise ratio and non-stationarities. Presence of non-stationarities in the EEG signals significantly perturb the feature distribution thus deteriorating the performance of Brain Computer Interface. Stationary Subspace methods discover subspaces in which data distribution remains steady over time. In this paper, we develop novel spatial filtering based feature extraction methods for dealing with nonstationarity in EEG signals from a drowsiness detection problem (a machine learning regression problem). The proposed method: DivOVR-FuzzyCSP-WS based features clearly outperformed fuzzy CSP based baseline features in terms of both RMSE and CC performance metrics. It is hoped that the proposed feature extraction method based on DivOVR-FuzzyCSP-WS will bring in a lot of interest in researchers working in developing algorithms for signal processing, in general, for BCI regression problems.

The ubiquitous presence of non-stationarities in the EEG signals significantly perturb the feature distribution thus deteriorating the performance of Brain Computer Interface. In this work, a novel method is proposed based on Joint Approximate Diagonalization (JAD) to optimize stationarity for multiclass motor imagery Brain Computer Interface (BCI) in an information theoretic framework. Specifically, in the proposed method, we estimate the subspace which optimizes the discriminability between the classes and simultaneously preserve stationarity within the motor imagery classes. We determine the subspace for the proposed approach through optimization using gradient descent on an orthogonal manifold. The performance of the proposed stationarity enforcing algorithm is compared to that of baseline One-Versus-Rest (OVR)-CSP and JAD on publicly available BCI competition IV dataset IIa. Results show that an improvement in average classification accuracies across the subjects over the baseline algorithms and thus essence of alleviating within session non-stationarities.

Electroencephalogram (EEG) signals are one of the most widely used noninvasive signals in brain-computer interfaces. Large dimensional EEG recordings suffer from poor signal-tonoise ratio. These signals are very much prone to artifacts and noise, so sufficient preprocessing is done on raw EEG signals before using them for classification or regression. Properly selected spatial filters enhance the signal quality and subsequently improve the rate and accuracy of classifiers, but their applicability to solve regression problems is quite an unexplored objective. This paper extends common spatial patterns (CSP) to EEG state space using fuzzy time delay and thereby proposes a novel approach for spatial filtering. The approach also employs a novel fuzzy information theoretic framework for filter selection. Experimental performance on EEG-based reaction time (RT) prediction from a lane-keeping task data from 12 subjects demonstrated that the proposed spatial filters can significantly increase the EEG signal quality. A comparison based on root-mean-squared error (RMSE), mean absolute percentage error (MAPE), and correlation to true responses is made for all the subjects. In comparison to the baseline fuzzy CSP regression one versus rest, the proposed Fuzzy Time-delay Common Spatio-Spectral filters reduced the RMSE on an average by 9.94%, increased the correlation to true RT on an average by 7.38%, and reduced the MAPE by 7.09%.

Brain-Computer Interfaces (BCI) are systems that enable users to use neural signals, typically Electroencephalogram (EEG) to direct an application or an external device. Motor imagery (MI) based BCI detects subject motor intentions which could be further used as control signals. Due to spatial lateralization of different MI tasks, spatial filtering followed by band power extraction is the most commonly used algorithm for classification tasks in MI-based BCIs. Unfortunately, the spatial filtering approach significantly incorporates Amplitude characteristics when compared to Phase characteristics of the EEG signal. Single trial Phase locking value (sPLV) has been a popular statistics to extract phase based information for classification task in MI-based BCI. To utilize the phase characteristics for MI classification, this paper proposes a novel approach based on instantaneous phase difference (IPD) sequence to extract phase features that explicitly use the phase synchronization information between EEG sensors. We maximized the discriminability of IPD sequence using linear transformation calculated from common spatial pattern algorithm (CSP) on the IPD sequence. An evaluation of our method on BCI competition dataset led to around 15% increase in mean classification accuracies compared to sPLV approach and comparable accuracies to power feature based CSP algorithm. Furthermore, incorporating phase features from our method and power features from traditional algorithms using sparse feature selection technique increased the classification accuracy over both CSP and CSP on the IPD sequence.

A control theoretic approach is presented in this paper for both batch and instantaneous updates of weights in feed-forward neural networks. The popular Hamilton-Jacobi-Bellman (HJB) equation has been used to generate an optimal weight update law. The main contribution in this paper is that a closed form solutions for both optimal cost and weight update can be achieved for any feed-forward network using HJB equation. The proposed approach has been compared with some of the existing best performing learning algorithms. It is found as expected that the proposed approach is faster in convergence in terms of computational time. Some of the benchmark test data such as 8-bit parity, breast cancer and credit approval, as well as 2D Gabor function have been used to validate our claims.

In the past decade, improvements in the production of in-expensive PC equipment and software has permitted more refined real-time signal processing in BCI systems. In the literature, Deep learning concepts have not been applied to EEG data analysis in a systematic manner. This paper applies various existing Deep learning architectures and algorithms for the classification of EEG data applied to eye state detection. The deep learning based classifier systems presented in this work are comparable to the state of the art classifiers devised by Roesler and Suenderman (2013), and Cameron et al. (2015). The goal of this work is to construct a system producing accuracies comparable to Roesler's K* classifier, Cameron et al.'s (RRF+K*) classifiers and at the same time providing enough speed to be used in an online BCI framework. In order to meet the constraints, following architectures were designed: A Multi layered neural network with ReLU and drop-out, deep belief networks based unsupervised learning, drop-out masks on deep neural networks. Specifically, we compare our results with K*, RRF, (K*+RRF), ada(RJ48F) classifiers. Also an in-depth analysis of binary class features has been done using t-SNE based visualizations while fitting elliptical contours to the features. Prior research suggests that instance-based/lazy learners like the K* algorithm are likely to be too slow to be used in a BCI framework, with ada(RJ48F) model performing decently well. But our chosen deep neural network architectures produce higher classification accuracies and have lower convergence times making them even faster within the time specifications of real-time classification and control applications.