Disturbed sleep has become more common in recent years. To improve the quality of sleep, undergoing sleep observation has gained interest in an attempt to resolve possible problems. Additionally, the classification of real-time sleep states without interference is needed in nursing and medical settings. In this paper, we discuss a non-restrictive and non-contact method for estimating real-time sleep stages and report its potential applications and problems. The system we used measured body movements and respiratory signals of a person while sleeping using dual 24-GHz microwave radars placed underneath the mattress. We first determined a body movement index to identify wake and sleep, and fluctuation indices of respiratory intervals to indicate sleep stages. The data of movements and respiratory signals derived from 11 healthy university students were used for feature selection and training of the classification model parameters. For identifying wake and sleep, the rate of agreement between the body movement index and the result using the R & K method as reference was 83.5±6.3%. Five-minute standard deviation, a fluctuation index of respiratory intervals, had a high degree of contribution and showed a significant difference (p<0.001) among three sleep stages (REM, LIGHT, DEEP). Th e degree of contribution of the 5-min fractal dimension of respiratory intervals, a fluctuation index, was not as high as expected but showed a significant difference (p<0.05) between REM and DEEP. For validation, we applied canonical discriminant analysis to classify wake or sleep and to estimate the three sleep stages in two additional university students. The accuracy was 79.3% and 59.5% for classification of wake/sleep and 71.9% and 64.0% for estimation of three sleep stages. The novelty of this study includes measurements of body movements and respiration-induced body surface movements withh ighsensitivities using microwave radars, and systematic analysis of the indices of body movement and respiratory interval. The method allows easy measurement of sleep stages in nursing care and home settings and may be employed to increase the quality of sleep. Although we successfully estimated three sleep stages, the method requires further study to determine the five stages of sleep.
This paper describes a novel post-processing algorithm for electromyographic (EMG) pattern classification, for use with myoelectric prosthetic hands. Amputees have difficulties controlling multiple degrees of freedom, but there is an increasing number of prosthetic hands with multiple degrees of freedom. Generally, an increasing number of classes decreases the classification accuracy. Artificial neural networks have been used for EMG pattern classification in previous studies. The proposed post-processing algorithm stores the temporal sequence of classifications from the EMG pattern classification algorithm, and runs a second classification based on the sequential patterns. We compared the accuracy of the output before and after the post-processing step. In our experiment, we set the training time of the EMG pattern classification algorithm to 1 s for each class, and used three channels of surface EMG signals. We selected 7 and 9 classes of EMG patterns, and recorded the output every 10-20ms. The classification accuracy improved by 11.5% with 7 classes, and 17.7% with 9 classes. The overall accuracy of the proposed system was 82.5% for 9 classes and 92.9% for 7 classes. With the adequately high classification accuracy and other features (small number of EMG channels and short training time), the proposed method is potentially suitable for practical use with prosthetic hands.