Japan Society for the Promotion of Science:Grants-in-Aid for Scientific Research
Date (from‐to) : 2003 -2004
Author : FUKUMI Minoru; TAKEDA Fumiaki; YAMAMOTO Toru; MITSUKURA Yasue
Recently, information terminals such as a cellular phone have been widely used. According to this, industrial standard of radio communication such as Bluetooth has been established. As a result, it would be possible to combine and to perform various interfaces. However, now a day, the device that has various operations of portable machines and tools and can control networks (call "total operation device" for short) has not been provided yet. Moreover, the wristwatch type is preferable in the viewpoint of operationality. Therefore, we investigate ElectroMyoGram (EMG) which is a signal generated from a living body with movement of a subject.
First, time series data for EMG is measured by electrodes in the input part. Second, this data is amplified and A/D transform is performed in the signal processing part. Next, this amplified data is converted to Fourier power spectra. Finally, we evaluate various data in the learning-evaluation part.
We aim for construction of a high-speed and high-accurate EMG recognition system which can do on-line learning using DSP training board. In order to achieve high accuracy, we used Fast Fourier Transform (FFT) for feature extraction, Simple-PCA (SPCA) for feature compression, and a neural network (NN) for recognition. In particular, we presented a novel method based on Multiple PCA to improve recognition accuracy for EMG. From results of computer simulation, it is shown that our approach is effective for improvement in recognition accuracy and speed.
Furthermore, we used a genetic algorithm for condteracting a rule generation system which can improve recognition accuracy for EMG. This method yielded mathematical functions using input attributes selected by the genetic algorithms. These functions can achieve a high accuracy compared to conventional approach.
Finally we tested noize elimination performance using wavelet transform. In this method, small components after the wavelet transform are eliminated and then signals are inversely transformed. These signals were used for EMG recognition and evaluated its accuracy.