Multi-Modal Dataset for "Towards Robust Surface Electromyography for Upper Limb Protheses using Machine Learning "

Data collection method:

The data were collected using 2 custom-built sEMG sensors from the participant's forearm, placed on the flexor carpi ulnaris and the extensor carpi radialis of each participant. The sEMG data was collected at 500 Hz, with an onboard 1st-order Bandpass filter at 4.82 and 241.1 Hz applied before digital conversion. A gain of 162.5 is applied to the data. The data were recorded in an unregulated environment. The sEMG data provided are as recorded, no further filtering has been applied. For each participant, 13 recording trials were performed, in each the participant performed 2 repetitions of 6 different hand gestures, picking up appropriate objects to perform the gestures. Image depictions of the gestures can be found alongside the data. Participants performed the gestures for approximately 5 seconds, followed by approximately 11 seconds of rest. Over the first 5 seconds of the rest period the sEMG sensors do not record, and instead a skin temperature and a skin-electrode impedance recording are made by the custom sensor units. The 12 temperature and impedance data are stored within the same .mat file as the sEMG data for each trial. Temperature data is recorded in Celsius, impedance in ohm and phase angle pairs per channel, and the sEMG data is recorded in Volts. Accompanying each set of data is additionally vectors indicating the gesture being performed (or rest where appropriate) and whether it is repetition one or two, these allow for the generation of training and testing datasets for pattern recognition applications. The 13 recording trials include: 2 control trials, 2 trials in which the participant arm temperature was varied, 2 trials in which the skin-electrode impedance was changed, 3 trials in which the participant varies the arm's position, and 4 trials in which the electrodes position is shifted relative to the base position. Each trial is stored within its own recording file in the dataset, and the details of the trial order and name are presented in the accompanying text file.