The last decade the impulse-radio ultra-wideband (IR-UWB) radar sensors have tended to substitute the gold standard, polysomnography (PSG), for the acquisition of the primary vital signs of the human body. However, the main drawback of the radar sensor is its sensitivity to the environmental noise and body motion. In this work, we aim to investigate whether radar sensor recordings of the chest and heart motion can be accurate substitutes to the PSG chest and heartbeat signals measurements. In terms of this scientific issue, we develop an innovative pipeline of handling the radar-based recordings, which includes: (a) the motion detection, (b) the extraction of the respiration, heartbeat and activity vital signs and (c) the estimation of the respiratory and heartbeat rates, i.e., the RR and HR, respectively. Our experimental results, applying our proposed methodology to 5 subjects during their sleep, showed that the radar sensor’s measurements can be comparative to the those producted by the PSG. Moreover, the corresponding RR and HR frequencies were proven to have Pearson’s correlation, between the radar and the PSG, greater than 0.9 and 0.78, respectively. Finally, the relative errors between the PSG, and the radar-based RR and HR rates were proven to be less than 6% and 10%, respectively.

Musical instrument recognition is a relatively unexplored area of machine learning due to the need to analyze complex spatial-temporal audio features. Traditional methods using individual spectrograms, like STFT, Log-Mel, and MFCC, often miss the full range of features. We propose a hierarchical residual attention network using a scaled combination of multiple spectrograms, including STFT, Log-Mel, MFCC, and CST features (chroma, spectral contrast, and Tonnetz), to create a comprehensive sound representation. This model enhances focus on relevant spectrogram parts through attention mechanisms. Experimental results with the OpenMIC-2018 dataset show significant improvement in classification accuracy, especially with the "Magnified 1/4 Size" configuration. Future work will optimize CST feature scaling, explore advanced attention mechanisms, and apply the model to other audio tasks to assess its generalizability.