Stroke is a major neurological disease that often results in long-term motor impairments requiring continuous rehabilitation. However, due to the shortage of rehabilitation professionals and the high cost of therapy, many patients attempt self-rehabil...
Stroke is a major neurological disease that often results in long-term motor impairments requiring continuous rehabilitation. However, due to the shortage of rehabilitation professionals and the high cost of therapy, many patients attempt self-rehabilitation at home, often with limited effectiveness. This study proposes an AI-based hemiplegic upper limb rehabilitation assistance systemcapable of evaluating the quality of rehabilitation movements using integrated IMU (Inertial Measurement Unit) and EMG (Electromyography) sensors. The system collects signals, applies preprocessing filters, and utilizes a 1D Convolutional Neural Network (1D-CNN) model to classify user movements into three categories: normal, inaccurate, and error. Data were collected from four healthy male subjects, each performing 1,440 upper-limb motions. Notch and Complementary filters were applied for signal denoising and alignment. The classification accuracies were 94.4% for normal, 93.7% for inaccurate, and 88.8% for error movements. The results demonstrate that, even under limited clinical data conditions, the proposed IMU–EMG fusion with a 1D-CNN architecture provides high recognition accuracy and potential applicability to intelligent rehabilitation and smart healthcare platforms.