Kayla’s Status Update for December 5th
This week, and including last week, I worked on narrowing down the feature extraction and classification pipeline to be used for classifying EMG signals in our project. After discussing with Tarana to understand the expected data input to the SVM classifier, I created a 1×15 matrix for the feature vector to which included the max value for each electrode, five binary signals about whether or not a particular electrode fired during the recorded signal, and the peak frequency from the Fourier transform. I originally have used a bandpass filter to keep data between 50-150 Hz but upon discussing with Tarana, we realized that we lost meaning in the signal when filtering out the low frequencies because I didn’t realize the DC offset was accounted for in this region. In order to resolve this, I ended up using a lowpass filter at 150 Hz and the resulting data was much more useful for our classification purpose. Another major success from this week was completing integration with the whole game. It was very exciting to see everything put together and to see us as a team working together toward preparing our final presentation and report.
In preparation for our final report, I wanted to have a way of visualizing the decision boundaries of the classifier. I created two plots to visualize this, shown below.
Figure 1A: https://ibb.co/9rgxs5X Figure 1B: https://ibb.co/pjQjm3J
(Our website seems to have exceeded the storage quota of 75 MB so I am using links)
Figure 1. A (left). This plot shows the decision boundaries of the first two features in the feature vectors of the simulated data. This classification was limited to only two features in order to fit on a 2-Dimensional plot.
B (right). This plot shows the decision boundaries of the first two principal components of all the feature vectors.
Overall, the progress in the past two weeks has been steady for our team. We have all been learning from each other in this integration process. Everything is on schedule for final presentations and deliverables this upcoming week. I look forward to presenting our final product.