Contactless breathing rate (BR) estimation using ultra-wideband (UWB) radar offers unobtrusive health monitoring without subject cooperation. However, existing methods assume a fixed posture, and performance degrades when sleeping individuals change position, an event that occurs 20-40 times per night. This paper presents PostureBreathNet, a lightweight multi-task architecture that jointly classifies sleeping posture (facedown, lateral, supine) and estimates BR from UWB radar signals. The network uses Feature-wise Linear Modulation (FiLM) to condition the BR estimation branch on detected posture, allowing posture-adaptive signal interpretation. We extract a 62-dimensional feature vector spanning four domains: breathing dynamics, range profile geometry, range-time texture, and cross-domain interactions from 2D range-time matrices. On a public UWB breathing dataset of 431 recordings (one excluded for out-of-range breathing rate) across three postures and three radar bandwidths, PostureBreathNet achieves 80.74% posture classification accuracy [95% CI: 77.0%-84.5%] and a BR mean absolute error (MAE) of 1.66 BPM [95% CI: 1.23-2.17 BPM], a 53.6% reduction over the FFT baseline (3.58 BPM). Leave-one-posture-out (LOPO) validation confirms generalization to unseen postures (2.34 BPM average MAE). At 66,820 parameters and 65.3 KB INT8, the model requires approximately 66.3 KB SRAM on an ESP32 microcontroller, enabling autonomous edge deployment with sub-millisecond model inference; measured end-to-end pipeline latency including feature extraction is 142.98 ms. A simulation-based sensitivity analysis using a low-cost DWM3000 positioning chip reveals a large domain gap (37.92% posture accuracy under zero-shot transfer), indicating that radar-specific calibration is needed for multi-hardware deployment. Prospective target applications include infant monitoring for Sudden Infant Death Syndrome (SIDS) prevention, home sleep apnea screening, and elderly care.
The growing environmental issues regarding plastic waste have necessitated the need for sustainable and biodegradable alternatives to traditional plastic pots. Present study focuses on the production of eco-friendly biodegradable plant containers using various bio composites consisting of natural fibers such as pineapple leaves (PC), water hyacinth (WC), dried leaf litter (DC), banana fibers (BC), and coco peat with cornstarch acting as a natural adhesive. The main aim is to produce a nature-friendly alternative to conventional plastic pot maintaining structural integrity and durability. To assess the viability of the bio composite, several mechanical and environmental tests were carried out, such as compression, flexural strength, impact resistance, water absorption, and biodegradability tests. A direct planting test with the money tree (Epipremnum aureum) was conducted to qualitatively assess the practical applicability of the biopots under real-use conditions. All experimental data were analyzed statistically, and differences among the samples were considered significant at p < 0.05. Water absorption analysis showed that BC exhibited the highest absorption (96.40 ± 3.59
Digital micromirror devices (DMDs) are a popular choice for wavefront-shaping applications. We describe the restructuring of commercial digital light projectors to utilise their DMD for consistently replicable and budget-friendly diffraction and wavefront-shaping application experiments. Techniques for validating the optical parameters of DMDs are described. The experimental validation of the concept of the Fresnel zone plate using the set-up is discussed. We verified the experimental and simulated reconstruction of two dots and their Fourier transforms. We also describe the generation of orbital angular momentum states of light using the experimental set-up.
We present StegoMed, a deep learning-based framework intended to facilitate secure transmission of medical images in teleradiology environments. Drawing inspiration from GAN-based representations, StegoMed uses a lightweight two-stage encoder-decoder architecture without a discriminator which enables stable training and computational efficiency. The proposed framework embeds MRI brain images within natural cover images and subsequently reconstructs them with high fidelity. To enhance both imperceptibility and recoverability, the model is optimized by combining pixel-wise reconstruction, perceptual, and structural similarity losses. Experimental results demonstrate that StegoMed achieves high average reconstruction quality (PSNR = 47.7 dB, SSIM = 0.9976, MSE = 0.00006) and outperforms several existing baselines. The method also shows robustness to common distortions such as JPEG compression and Gaussian noise. These results demonstrate StegoMed’s potential as an effective and privacy-preserving solution for secure medical image transmission in modern teleradiology workflows.
The covalent functionalization of the graphene surface is successfully employed using a silane coupling agent, 3-aminopropyl trimethoxy silane (APTMS). The silane modification increased the interlayer spacing between the rGO layers. The composite electrode loaded with 1.5 g APTMS displayed 97.3 % pseudocapacitance, implying the successful silylation of the oxygenated functional groups on the graphene structure. The silylation imparts a sheet-like internal morphology with sharp edges to rGO's morphology. Among binary electrodes, Si-rGO with 20 wt% loading of acid-treated multi-walled carbon nanotubes (A-CNT) (PSRC20 binary electrodes) show excellent electrochemical properties and the highest specific capacitance. PSRC20 binary electrode displayed 82.6 %
