.png){kind=icon}
.png){kind=icon}
Return
CSI-Based Respiratory Pattern Recognition Using Diffusion Model and Deep Learning
.png){kind=icon}
PRE
AI
DOI：10.1109/TNSE.2025.3638894.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Respiratory patterns are important indicators of human health, and using AI models to analyze channel state information (CSI) for non-contact respiratory detection shows great potential. However, these methods suffer from limited data, which reduces both the detection accuracy and the model's generalization. Generative AI offers a novel solution to this issue, which can efficiently generate high-quality respiratory data, alleviate the problem of data scarcity and improve model performance. This study proposes an innovative approach that combines a diffusion model with a convolutional-transformer (CNN-Transformer) network model to achieve reliable recognition of respiratory patterns under data-scarce conditions. We first design a set of data preprocessing methods, which converts CSI amplitude and phase into 1D time series representing respiratory patterns. Then, we apply the diffusion model for data augmentation to generate diverse and high-quality RF data. Subsequently, the preprocessed data are mapped into 2D images using the gramian angular field (GAF) method and then fed into a CNN-Transformer network model to classify three types of respiratory patterns. Experimental results demonstrate that the proposed approach can effectively enhance the performance of respiratory pattern recognition. When combined with the CNN-Transformer model, the accuracy, precision, recall, and specificity are 98.3%, 98.3%, 98.0%, and 98.8%, respectively, which are improved by 3%-4% compared with the CNN, ResNet-18, and long short-term memory (LSTM) baseline models. This approach validates the feasibility of respiratory pattern recognition and provides robust support for the early diagnosis and monitoring of related diseases.
Keywords:
Respiratory pattern
data augmentation
channel state information (CSI)
deep neural network
Journal
I
IF:
7.9
Papers:
2.5K
Citations:
10.0K