Breast Cancer Postoperative Rehabilitation Nursing Action Recognition Based on Millimeter-Wave Radar

Rehabilitation exercise after breast cancer surgery is very important for the recovery of female patients. Unfortunately, the shortage of professional nursing staff has resulted in the inability to monitor patients exercise in real-time. Human activity recognition (HAR) technology based on visual sensors can alleviate the current situation, but the technology has the risk of disclosing the privacy of patients. Wireless human action sensing (esp. mmWave radar) technology can protect patients privacy, which is a potential solution. However, the rehabilitation nursing exercises after breast cancer surgery are mainly completed by the upper limbs, and the research on recognizing such actions based on a radar sensing system is still blank. Thus, we propose a high-precision breast cancer postoperative rehabilitation nursing action recognition system using commercial millimeter-wave radar, termed mm-BCRAR. In this paper, we first extract the range-time map (RTM) and micro-Doppler map (MDM) of rehabilitation action. Then, a multi-feature enhancement fusion network is constructed, termed MFEF-Net, which extracts the key features of rehabilitation actions by a deep neural network and attention mechanism. In addition, we design a joint loss function to distinguish the similarity between samples to improve the recognition accuracy of the system. Finally, the experimental results show that the overall accuracy of the mm-BCRAR system in recognizing 9 different rehabilitation actions is 98.44%, and it has good generalization. The method proposed in this paper has good application potential in helping medical staff to monitor the rehabilitation exercise of patients with breast cancer after surgery.