Advice

"Common mans" guide to Machine learning

posted Dec 28, 2016, 4:52 PM by Bodo Hoenen

James Newton: "..Not sure it will help, but I've written a "common mans" guide to machine learning which is available here:
http://techref.massmind.org/techref/method/ais.htm.."

Processing EMG signals through an artificial neural network

posted Nov 28, 2016, 9:26 AM by Bodo Hoenen

Pamela Hardaker "... There are a couple of things that might be of interest to you.  

 

The first is my own research on processing EMG signals by extracting features and passing them through an artificial neural network to determine what the user is doing.  You can find both my papers here:  http://www.dmu.ac.uk/about-dmu/academic-staff/technology/pamela-hardaker/pamela-hardaker.aspx and my work has moved on again since then.

Myoelectric Pattern Recognition

posted Nov 28, 2016, 9:18 AM by Bodo Hoenen

"... We see your biggest needs in this phase of development to be in 2 areas: 

1) a pattern recognition algorithm and 
2) high quality myoelectric sensing electronics. 

For the algorithm part, we do not think your requirements are as advanced as what the Coapt system does. There are a lot of novel machine learning approaches out there and we would suggest using the most basic. I anticipate that you could use a signal-amplitude based fuzzy logic or basic neural network or similar. That is, I don't believe that you will need the most advanced signal processing to achieve your goals and the basic building blocks here could be just what you need. 

For the EMG sensing, an approach using basic electrodes, but more of them, may be all that you require. Any implemented recognition algorithm will take advantage of subtle differences from multiple electrodes instead of comparing signals from just two. I believe that a lot could be achieved simply by adding an array of low cost, off-the-shelf electrodes to your approach. 

Machine Learning for Myoelectric Signal Recognition

posted Nov 28, 2016, 8:21 AM by Bodo Hoenen   [ updated Nov 28, 2016, 8:22 AM ]

William Smith "... 

My advice would be to look into support vector machine learning, it is widely used for your application and seems to be effective, for example: http://ieeexplore.ieee.org/document/4463647/. Particular advantages of this method are the robustness and low computational load, all key if you want to make this mobile. 


Unfortunately I don't know of any plug and play software for this, but http://www.svms.org/ is an excellent place to get started and the LIBSVM software library would be a potentially good place to start pulling something together as they have some simple examples in Python which is useful if you want to work on the Arduino. 


Arduino SVM in particular might be useful for direct usage

1-6 of 6

Comments