My initial purpose is very simple: invented a device can help people just use mind to control objects. To achieve this goal, both off-line and on-line experiments are needed.
1. Off-line experiment
Data collection: we use 20-channel brain electric cap to collect two-class motor imagery brain wave signals.
Feature extraction: we first use Hilbert-Huang transform (HHT) as preprocessing. Then we use Empirical Mode Decomposition (NA-MEMD) algorithm as feature extraction algorithm. To further improve the effect, Common Spatial Patterns (CSP) follows as the last step.
Feature classification: we test Support Vector Machine (SVM) and Random Forest algorithms, finding that SVM is better, so we choose SVM as our feature classification algorithm.
2. On-line experiment
The signal processing pipeline is HHT --> NA-MEMD --> CSP -->SVM. All of the signal processing algorithms are encapsulated. We implement our whole pipeline by collecting brain wave data, algorithms processing and return the classification result. According to the classification result, program can know what people is thinking and take action. For example, if our goal is letting people control the turning of a car toy, when the man is thinking 'turn right', our algorithms can identify this signal and make the car turn right. Result has shown that the classification accuracy of our method is up to 87%.
The best news is that one of my partners has set up a company based on this project. The aim of the medical product is to help rehabilitation of cerebral apoplexy. Although my dream is also starting my own company, as a sophomore, I don’t regard it is the best time because I think I need to further hone my research and engineering skills.
For more details about this project and product, please see the video below. (Chinese edition)