EMG-Classification

This repository contains code for basic finger and hand gesture classification using EMG signal as input and some pretrained deep neural networks such as -googlenet, Alexnet etc. Dataset has been taken from Rami Khusaba Electromyogram repository [1].

Steps for running this code

Step 1

First download the dataset from this link and extract in the same folder where you will run the codes. This is the first dataset of Rami Khusaba Electromyogram repository. It contains 10 classes of individual finger movements, i.e.,Thumb (T), Index (I), Middle (M), Ring (R),Thumb–Index (T–I), Thumb–Middle (T–M), Thumb–Ring (T–R), Thumb–Little (T–L), hand close (HC), Little (L), Thumb–Little (T–L) from 10 subjects. We only used first 8 classes.

Step 2

Open MATLAB and click on the file named Hd.mat (A Bandpass Butterworth filter object) to import it in MATLAB workspace.

Step 3

Run the function read in MATLAB command window in the following style -

EMGData = read(sub, Hd);

Here sub = 1,2,3,..etc. the desired number of subjects you want to read.

Step 4

Now run the file script section by section using in the following style -

parentDir = './';
dataDir = 'EMGD1All';
helperCreateEMGDirectories(EMGData,parentDir,dataDir)

You can give your own name for dataDir. Plot the EMG data for first 1000 samples using the following section -

helperPlotReps(EMGData)

Create a scalogram for 1st 1000 samples and visualize it-

Fs = 4000;
fb = cwtfilterbank('SignalLength',1000,...
    'SamplingFrequency',Fs,...
    'VoicesPerOctave',12);
sig = EMGData.Data(1,1:1000);
[cfs,frq] = wt(fb,sig);
t = (0:999)/Fs;figure;pcolor(t,frq,abs(cfs))
set(gca,'yscale','log');shading interp;axis tight;
title('Scalogram');xlabel('Time (s)');ylabel('Frequency (Hz)')

Create RGB images for all EMG signals-

helperCreateRGBfromTF(EMGData,parentDir,dataDir)

This may take a while, be patient. Store all the images in allImages variable -

allImages = imageDatastore(fullfile(parentDir,dataDir),...
    'IncludeSubfolders',true,...
    'LabelSource','foldernames');

Split the images in training and validation sets (here ratio is 0.7) -

rng default
[imgsTrain,imgsValidation] = splitEachLabel(allImages,0.7,'randomized');
disp(['Number of training images: ',num2str(numel(imgsTrain.Files))]);
disp(['Number of validation images: ',num2str(numel(imgsValidation.Files))]);

Now you can run the deep neural network googlenet-

net = googlenet;

Visualize the layers -

lgraph = layerGraph(googlenet);
numberOfLayers = numel(lgraph.Layers);
figure('Units','normalized','Position',[0.1 0.1 0.8 0.8]);
plot(lgraph)
title(['GoogLeNet Layer Graph: ',num2str(numberOfLayers),' Layers']);

net.Layers(1)

Remove the last 4 layers and add 4 new layers -

lgraph = removeLayers(lgraph,{'pool5-drop_7x7_s1','loss3-classifier','prob','output'});

numClasses = numel(categories(imgsTrain.Labels));
newLayers = [
    dropoutLayer(0.6,'Name','newDropout')
    fullyConnectedLayer(numClasses,'Name','fc','WeightLearnRateFactor',5,'BiasLearnRateFactor',5)
    softmaxLayer('Name','softmax')
    classificationLayer('Name','classoutput')];
lgraph = addLayers(lgraph,newLayers);

lgraph = connectLayers(lgraph,'pool5-7x7_s1','newDropout');
inputSize = net.Layers(1).InputSize;

Specify training options-

options = trainingOptions('sgdm',...
    'MiniBatchSize',15,...
    'MaxEpochs',20,...
    'InitialLearnRate',1e-4,...
    'ValidationData',imgsValidation,...
    'ValidationFrequency',10,...
    'Verbose',1,...
    'ExecutionEnvironment','cpu',...
    'Plots','training-progress');

Now you can train-

rng default
trainedGN = trainNetwork(imgsTrain,lgraph,options);

Follow almost same procedure for alexnet and other networks in the subsequent sections.

References

[1] R. N. Khushaba, M. Takruri, S. Kodagoda, and G. Dissanayake, "Toward Improved Control of Prosthetic Fingers Using Surface Electromyogram (EMG) Signals," Expert Systems with Applications, vol 39, no. 12, pp. 10731–10738, 2012.