Study on Filter Shape for Bio-Signals Training Data in CNN-Based HAR
Gerelbat Batgerel, Chunki Kwon · 2023
Convolutional neural network(CNN) has been also a powerful tool to recognize human activity using bio-signals. Unlike camera-captured images, bio-signals have row matrix-shaped training data after their imagification process. This is because sampling data for a second outnumber the number of bio-signal sources. In CNN using image data, there is no doubt that a square matrix shape is a good choice for filters to extract features from image training data. However, in CNN using bio-signal training data, a choice of filter shape would be a big issue for CNN's performance. This study addresses the issue and offers motivation for further research on the issue.