Abstract
UpperMyo is an open six-muscle surface electromyography (sEMG) dataset for upper-limb action recognition, sEMG-referenced temporal classification, and cross-dataset transfer. Analyses use 264 recording blocks from 24 healthy male participants, sampled at 1 kHz with 12-bit resolution; the nine-class dataset contains 2640 four-second samples spanning eight actions and background. Internal evaluations cover recognition, channel ablation, and elbow-flexion/rest temporal classification. Six classifiers underwent nested leave-one-subject-out validation; a raw-waveform convolutional neural network (CNN) achieved macro-F1 of 0.9860 and accuracy of 99.09%, versus 0.9830 and 98.71% for a support vector machine. Geng et al. reported 99.0% accuracy for eight gestures using high-density sEMG and 40-frame voting under within-subject testing. Offline elbow-flexion/rest temporal classification yielded feature-based CNN accuracies of 89.5% and 75.3% at 0 and 2000 ms backward offsets from an sEMG onset reference. Four external experiments compare neural and statistical transfer with target-only learning at identical labeling budgets for matched actions, new actions, activities without class mapping, and continuous grip-force estimation. WristMotion balanced accuracy increased by 14.21 percentage points with four labeled trials; mean flexor carpi ulnaris (FCU) grip-force increased from 0.6431 to 0.6899 at the full budget. UpperMyo provides traceable data and evaluation evidence for sEMG action analysis and cross-dataset learning.