Detection Instead Of Classification In Brain-Computer Interfaces

The electroencephalogram (EEG) based method is the main interest of current Brain-computer interface (BCI) research due to its advantages of convenient operation, low cost and non-invasiveness. Current studies about EEG processing focused on classification. These properties of brain-computer interface (BCI) based on classification methods suggest that better recording and feature extraction will not necessarily continue to improve performance; In addition the current BCI system should rely on experts to set up and continuously adjust. The situation obstructs BCI widely used in the area of controlling and communication with needs of severely disabled. The method of Signal detection overcomes the issues described on classification. The approach is to detect any change that occurs in an appropriate set of brain signal features rather than to detect a specific set of changes in specific features. So the new approach overcomes the potential obstacle that brings BCI from the laboratory into the clinical application.EEG detection based-GMM, Gaussian component of GMM to represent some of the EEG spectral shape, with the ability to simulate any distribution, and is very effective in computing. the new samples of EEG features can be calculated its likelihood based on the established model to detect whether it is the model.EEG detection based-SVDD, the problem is to target EEG training set as a whole a class, to find a closed boundary hypersphere around the data, demand that the hypersphere contains all the described target objects as much as possible, while minimize the chance of non-target objects falling into the sphere.EEG detection based- sparse representation, The signal detection uses the identification of sparse representation. We represent the test subject in an overcomplete dictionary whose base elements are the training subjects themselves. If sufficient training subjects are available from each class, it will be possible to represent the test subjects as a linear combination of just those training subjects from the same class. Sparse representation also provides a simple and surprisingly effective means of rejecting invalid test subjects not arising from any class in the training database.