This study proposes a hand motion intention recognition framework integrating surface electromyography (sEMG) decomposition with the addition of a residual spiking neural network (Res-SNN). A group of 7 chronic stroke and 14 neurotypical participants performed 35 wrist and hand movements while their sEMG signals were recorded. Extraction of motor unit spike trains (MUSTs) from sEMG data is performed using convolution  blind source separation (BSS) to decode motor unit firing patterns can address the altered recruitment dynamics in stroke survivors and increase accuracy. Incorporation of residual connections into the baseline convolution spiking neural network (CSNN) forms the Res-SNN architecture. The model was compared against ResNet, a traditional sEMG-based residual network and the baseline CSNN.  The Res-SNN model achieved classification accuracies of above 0.95 for stroke survivors and neurotypical participants. Res-SNN outcompetes ResNet in both participant pools and outcompetes CSNN in the stroke survivor group.