Patient Behavior Recognition Methods Based on WIFI-CSI Features

With the accelerating global aging and the rising prevalence of chronic diseases, precise monitoring of patient behavior has become a critical issue in the field of medical rehabilitation. In response to the issues of insufficient privacy protection and poor wearing comfort associated with traditional visual and wearable monitoring methods, this paper proposes a non-intrusive patient behavior recognition method based on WiFi Channel State Information (WiFi-CSI) features. By converting CSI signals into time-frequency spectrograms, the method proposed in this paper achieves higher resolution and stronger adaptability, utilizing a Convolutional Neural Network (CNN) model for feature extraction and accurate recognition of patient behavior. Simultaneously, a transfer learning strategy is introduced to optimize the cross-environment generalization ability of this method. By pre-training on the source domain and fine-tuning on the target domain, the model’s adaptability to dynamic environments with unknown interference factors is enhanced. Experimental results demonstrate that the proposed patient behavior recognition model achieves an accuracy of 96.28%, with a 9% improvement in cross-scenario accuracy. It maintains a high recognition rate of 95% even in complex environments, and the inference time for a single sample is 0.25 s, outperforming mainstream models in overall performance.