Yuh-Shyan Chen

Radar-based human motion recognition is gaining attention due to its inherent resistance to lighting conditions, especially in healthcare and safety applications concerning the personal privacy. In this paper, we propose a novel cross-modal bi-contrastive learning model named "BiCLR," which utilizes a Transformer-based network for temporal modeling to conduct instance discrimination in both single-modal and cross-modal modalities in a self-supervised learning manner. To enhance data density, a new radar data format, the "Radar Combination Map (RCM)," is presented to seamlessly integrate Range-Doppler Map (RDM), Range-Azimuth Map (RAM) and Range-Elevation Map (REM) into a single map. The objective of this paper is to address the inherent sparsity of radar data through cross-modality and newly introduced RCM, offering a transferable framework for various kinds of downstream tasks, advancing understanding through radar-based recognition. After a comprehensive evaluation, the pre-trained encoder demonstrates effectiveness in a new human motion recognition task using only radar data, despite being trained on a significantly smaller dataset. The experimental results clearly demonstrate BiCLR’s capability to utilize cross-modal and contrastive learning methods, as well as the improved performance in downstream tasks.

Due to the increasing severity of the COVID-19 pandemic, timely screening and diagnosis of infections are essential. Since cough is a common symptom of COVID-19, an AI-assisted cough classification scheme is designed in this paper to diagnose COVID-19 infection. To reduce the labeling efforts by human experts, a semi-supervised learning with voting scheme using a triple-classifier model is proposed for the COVID-19 cough classification. This work aims to improve the accuracy of the classification. Initially, the data pre-processing scheme is executed by performing data cleaning, resampling, and data enhancement so as to improve the audio quality before training. The pre-training scheme is then performed by using a few numbers of COVID-19 cough data with labeling. Then we modify a well-known self-supervised learning model, SimCLR, to a semi-supervised learning-based SimCLR-like model, which uses three different loss functions to fine-tune three training models for cough classification. Finally, a voting scheme is performed based on the classification results of the three cough classifiers so as to enhance the accuracy of the cough classification for COVID-19. The experiment results illustrate that the proposed scheme can achieve 85% accuracy, which outperforms the existing semi-supervised learning-based classification schemes.

Doppler-radar-based continuous human motion recognition (CHMR) recently has attracted extensive attention, which is a favorable choice for privacy and personal security. Existing results of CHMR using mmWave FMCW Radar not consider the continuous human motion with the high similarity problem. In this paper, we proposed a new CHMR algorithm with the consideration of the high similarity (HS), is namely CHMR-HS, by using the modified Transformer-sequential-decoder model, such that our Transformer-sequential-decoder model can recognize the continuous human action with the high similarity. The experimental results show that the accuracy of our proposed CHMR-HS scheme is 94.5%, is better than that of existing CHMR schemes.

Doppler-radar-based continuous human motion recognition recently has attracted extensive attention, which is a favorable choice for privacy and personal security. Existing results of continuous human motion recognition (CHMR) using mmWave FMCW Radar are not considered the continuous human motion with the high similarity problem. In this paper, we proposed a new CHMR algorithm with the consideration of the high similarity (HS) problem, called as CHMR-HS, by using the modified Transformer-based learning model. As far as we know, this is the first result in the literature to investigate the continuous HMR with the high similarity. To obtain the clear FMCW radar images, the background and target signals of the detected human are separated through the background denoising and the target extraction algorithms. To investigate the effects of the spectral-temporal multi-features with different dimensions, Doppler, range, and angle signatures are extracted as the 2D features and range-Doppler-time and range-angle-time signatures are extracted as the 3D features. The 2D/3D features are trained into the adjusted Transformer-encoder model to distinguish the difference of the high-similarity human motions. The conventional Transformer-decoder model is also re-designed to be Transformer-sequential-decoder model such that Transformer-sequential-decoder model can successfully recognize the continuous human motions with the high similarity. The experimental results show that the accuracy of our proposed CHMR-HS scheme are 95.2% and 94.5% if using 3D and 2D features, the simulation results also illustrates that our CHMR-HS scheme has advantages over existing CHMR schemes.