To amplify the accuracy and the reliability in HCI Based on EMG it required advanced methods for detection, decomposition, processing, and classification.
If I got your question right, you need to train a classifier, and then use it for decomposing (segmenting) and also tagging each part of a new EMG signal (continues sequence data).
FYI, The words in brackets are the equivalents of phrases you have used in my field. In Machine Learning this is in the category of temporal sequence labeling and segmentation, That includes a lot of important problems like HCI, as you rightly pointed out.
I think this article which I have attached is very similar to your subject.
Article Discriminative Methods for Classification of Asynchronous Im...
My former PhD student, Dr. Li. addressed this issue in his promotion research. Please start with the following publication:
Li, C., Rusak, Z., Horvath, I. and Ji, L. (2016) Implementation and Validation of Engagement Monitoring in an Engagement Enhancing Rehabilitation System, Article in IEEE Transactions on Neural Systems and Rehabilitation Engineering, 25(6):726 - 738·
Can you specify the type of Human Computer Interaction you are planning to investigate? I modified double threshold method for gesture detection in EMG envelope signal. I also use neural network for different gestures classification (i implemented FANN library in QT/C++) . Before classification you need to specify some parameters like "zero-crossing", average amplitude and so on, then normalized them (for example in regards to MVIC value) and then trained your network. Let me know what you want to do. Maybe I can help you more.
Thank you @Amir Atashin & @Imre Horvath for your kind response.
@Tomasz,
We are trying to translate commands from EEG for BCI application. We selected the features but while implementing in hardware we faced some complexity due to the artifacts as well as the nonlinear behavior of EEG. We are working on eliminating the artifacts from EEG with a new approach and in the meantime, we choose EMG to translate some commands.
Conference Paper User Independency of SSVEP Based Brain Computer Interface Us...
All process of classification of EMG signals comes before a feature extraction stage. There are many methods to extract information about signals in time, frequency, time-scale domain. It´s important review the papers that address this topic to select the appropriate method in base for application.
Segmentation in EMG may be of two types if I understood your question right.
If you want to find the features of the EMG signals in the feature extraction step before classification then the best way is to find the Time domain, Frequency domain or T-F domain features with the help of MATLAB. However if you want o find the SMUAP of the EMG signal then I would like to suggest you to use EMGLAB. I have used the previous methods and found very good results.
Stashuk, D. (2001). EMG signal decomposition: How can it be accomplished and used? Journal of Electromyography and Kinesiology, 11(3), 151–173. https://doi.org/10.1016/S1050-6411(00)00050-X