Browsing by Author "Yeng, Chia-Hong"

Now showing 1 - 2 of 2

Convolutional denoising autoencoder based SSVEP signal enhancement to SSVEP-based BCIs
(1970-01-01)
Chuang, Chia-Chun
;
Lee, Chien-Ching
;
Yeng, Chia-Hong
;
So, Edmund-Cheung
;
Lin, Bor-Shyh
For steady state visually evoked potential (SSVEP) based brain computer interfaces (BCIs), the elicited SSVEP signals always contain noises and then the performance of SSVEP-based BCIs would be greatly degraded in practical applications. Therefore, to develop an SSVEP signal enhancement would be able to increase the accuracy of SSVEP-based BCIs. In this study, a convolutional denoising autoencoder based SSVEP signal enhancement is proposed to suppress the noise components. The convolutional denoising autoencoder is applied to estimate and suppress the noise components. To effectively estimate the noise components, a sinusoid wave is designed as an ideal SSVEP signal. To ignore the effects of phase, cross correlation is adopted to estimate the phase in the training stage. The experimental results evaluated by using signal-to-noise ratio and canonical correspondence analysis showed that the proposed approaches can effectively suppress the noises components. Therefore, the proposed approach can be applied to develop robust SSVEP-based BCIs.
Noise Suppression by Minima Controlled Recursive Averaging for SSVEP-Based BCIs With Single Channel
(2017-12-01)
Lee, Chien-Ching
;
Chuang, Chia-Chun
;
Yeng, Chia-Hong
;
Chen, Yeou-Jiunn
;
Lin, Bor-Shyh
Subjects with amyotrophic lateral sclerosis (ALS) consistently experience decreasing quality of life because of this distinctive disease. Thus, a practical brain-computer interface (BCI) application can effectively help subjects with ALS to participate in communication. In practices, the noise would greatly reduce the performance of BCIs. In this study, minima controlled recursive averaging is applied to suppress noise and improve the performance of practical BCI applications. Minima controlled recursive averaging is used to correctively track the noise. To suppress these noises, a log-spectral amplitude estimator is selected as the gain function and used to effectively estimate the power spectrum of the noises. Eight subjects were asked to attend a performance test of the proposed approach and the canonical correlation analysis (CCA) was adopted to compare the proposed approach. The average recognition rates based on single channel are 69.57% and 74.63% for CCA and proposed approach, respectively. The experimental results demonstrated that our approach is able to improve performance in practice.