• Formulation of fully complex-valued building blocks for complex-valued CNNs, ensuring consistent operation in the frequency domain. • Lightweight, computationally efficient, fully complex-valued residual CNN operating on complex data in the frequency domain. • Novel log-magnitude activation function that maintains phase information while introducing effective non-linearity, along with a detailed comparative analysis with the complex ReLU variant and the Cardioid activation function. • We have shown that complex-valued CNNs can outperform real-valued CNNs, enhancing performance and generalization while reducing computational demands. Complex-valued convolutional neural networks have demonstrated promising results in reducing space, time, and computational complexity compared to real-valued models, particularly in signal and image processing. Despite their strong representational capacity and theoretical benefits, complex-valued CNNs remain limited due to theabsence of simplified theoretical and practical formulations for fully complex-valued building blocks. Existing studies often depend on fast Fourier transforms (FFT/IFFT) for domain transitions between layers due to the lack of well-established complex-valued activation functions or filter parameters initialization. Additionally, many earlier works adapt complex versions of the real-valued activation functions in a split-type manner, which might distort phase information and weaken generalization. To overcome these challenges, we propose a lightweight fully complex-valued residual CNN that operates entirely on complex data in the frequency domain. Our design simplifies fully complex building blocks and introduces a Log-Magnitude activation function that preserves phase information, outperforming traditional complex ReLU variants and the Cardioid activation function. Experimental validation across diverse multi-modal datasets, including MNIST, SVHN, MIT-BIH Arrhythmia, PTB Diagnostic ECG, DIAT- μ RadHAR , and DIAT- μ SAT , demonstrates the superior performance of our fully complex-valued CNNs over real-valued models.

In this paper, we introduce a framework for multi-class novelty detection using structural vibration signals. Structural vibration-based person identification is a promising soft-biometric approach with potential applications in elderly care and access control. However, current research faces two key challenges. The first challenge is the lack of large-scale datasets necessary for thorough evaluation in structural vibration gait recognition. To address this, we created a new dataset with recordings from fifty individuals. The second challenge lies in the limited exploration of deep learning methods for large-scale multi-class novelty detection in structural vibration data. To fill this gap, we propose the energy-shifted contrastive loss function, specifically designed for this task. Our results demonstrate that the proposed framework achieves 96.57% accuracy in multi-class classification. For novelty detection, it achieves an Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) score of 89.15% for single footsteps, which improves to 93.83% with five consecutive footsteps.

Structural vibration-based gait recognition has emerged as a promising soft-biometric modality, particularly for privacy-sensitive monitoring and access control. Despite its potential, current research is largely limited to proof-of-concept studies that rely on hand-crafted features, with minimal exploration of deep learning methodologies. This gap reduces the potential for integrating structural vibration-based gait recognition with existing modalities, such as camera-based systems. In this study, we propose a multi-modal gait recognition system that integrates both vision and structural vibration modalities. We address two key challenges: (a) lack of studies exploring outdoor gait recognition using both vision and structural vibration, and (b) absence of a multi-modal training scheme that combines these two modalities. To tackle the first challenge, we curated a dataset comprising five minutes of walking data from ten individuals captured simultaneously by two cameras and a geophone sensor. To address the second challenge, we developed a joint training framework that uses data from both modalities. Our methods achieve an accuracy of 96.03% (±1.12) using structural vibration signals alone, and this improves to 98.27% (±0.06) when both modalities are combined.

Gait recognition based on structural vibration signals is an emerging area in soft biometrics and healthcare. It is valued for its privacy-preserving features, making it ideal for continuous monitoring in healthcare environments. In this work, we propose a method for simultaneous person identification and gait-abnormality detection using structural vibration signals. We have experimented on a dataset of eight people. The system uses shared feature extraction layers with separate branches for identification and abnormality detection. Our framework achieves a person identification accuracy of ~89.00% and ~90.00% accuracy in detecting abnormality in gait patterns.

In this paper, we propose a novel system prototype for human activity recognition using a low-cost, low-power millimeter-wave (mmWave) frequency-modulated continuous wave (FMCW) radar. Our approach applies the Fast Fourier Transform on the slow time axis and employs a Capon filter to generate range-Doppler, range-azimuth, and range-elevation maps, respectively. It can also effectively mitigate noise and multipath effects. We then use principal component analysis for feature reduction, reducing the dimensionality of the feature vectors extracted from these maps, which can be used to train conventional machine learning classifiers. This approach aims to achieve a balance between computational complexity, accuracy, and overall system performance. Our proposed system demon-strates promising recognition rates and robustness across varying levels of activity granularity, achieving recognition rates from 90.28% for four activities up to 70.97% for seven fine-grained ac-tivities. These findings highlight the potential of millimeter wave radar and suggested range maps combined with conventional machine learning classifiers for noninvasive, privacy-preserving activity recognition, with significant implications for healthcare, elderly care, and ambient assisted living.

In biometrics, footstep-based seismic signals for person identification have become increasingly popular. Person identification is a crucial aspect because it serves as the foundation for personalized services like elderly monitoring, tailored access control, etc. However, these systems may not have enough data for specific individuals, especially those with limited mobility. This lack of data can be a significant challenge, making achieving accurate and reliable results difficult. We used Generative Adversarial Networks (GAN) to develop a novel technique for generating a footstep-based seismic signal of a person. In this paper, we have addressed two challenges. The first one involves generating synthetic footstep signals that closely mimic the significant statistical properties of an individual's footstep signal. The second challenge is filtering the generated data to achieve high classification accuracy. To overcome the first challenge, we have developed a Wasserstein distance-based One-Dimensional Generative Adversarial Network(1DGAN) with gradient penalty, to generate synthetic footstep data of a person. To address the second challenge we formulated a screening process based on statistical error metrics. Our experimentation on ten people achieved an overall test identification accuracy of 98.44%. Furthermore, it achieved an average identification accuracy of 91.23% from just ten footsteps when tested with unseen real data

Protective measures against small unmanned aerial vehicles (UAVs) are vital from a national security perspective. As a result, the importance of surveillance systems that automatically identify and classify low radar cross section (RCS) aerial targets increases. In this work, an indigenously developed continuous wave (CW) (X-band: 10 GHz) radar is used to build a diversified “DIAT-$\mu $SAT” dataset comprising 4849 micro-Doppler signature images of five different small aerial targets. We also proposed a transfer learning-based deep convolutional neural network (DCNN) approach for classifying low RCS aerial targets. We demonstrated the classification accuracy of 95% and 97%, with VGG16 and VGG19 as feature extractors, respectively, with minimal false-negative and -positive results. The open-field experimental classes covered in this work are: 1) a two-blade rotor; 2) a three-short-blade rotor; 3) a three-long-blade rotor; 4) a quadcopter; 5) a bionic bird; and 6) a two-blade-rotor and bionic bird. We also observed a good classification accuracy (>97%) when more than one target is operated simultaneously.

The coronavirus COVID-19 pandemic is today's major public health crisis, we have faced since the Second World War. The pandemic is spreading around the globe like a wave, and according to the World Health Organization's recent report, the number of confirmed cases and deaths are rising rapidly. COVID-19 pandemic has created severe social, economic, and political crises, which in turn will leave long-lasting scars. One of the countermeasures against controlling coronavirus outbreak is specific, accurate, reliable, and rapid detection technique to identify infected patients. The availability and affordability of RT-PCR kits remains a major bottleneck in many countries, while handling COVID-19 outbreak effectively. Recent findings indicate that chest radiography anomalies can characterize patients with COVID-19 infection. In this study, Corona-Nidaan, a lightweight deep convolutional neural network (DCNN), is proposed to detect COVID-19, Pneumonia, and Normal cases from chest X-ray image analysis; without any human intervention. We introduce a simple minority class oversampling method for dealing with imbalanced dataset problem. The impact of transfer learning with pre-trained CNNs on chest X-ray based COVID-19 infection detection is also investigated. Experimental analysis shows that Corona-Nidaan model outperforms prior works and other pre-trained CNN based models. The model achieved 95% accuracy for three-class classification with 94% precision and recall for COVID-19 cases. While studying the performance of various pre-trained models, it is also found that VGG19 outperforms other pre-trained CNN models by achieving 93% accuracy with 87% recall and 93% precision for COVID-19 infection detection. The model is evaluated by screening the COVID-19 infected Indian Patient chest X-ray dataset with good accuracy.