The invention discloses an electrocardiosignal classification method and
system based on multi-domain
feature learning, and the method comprises the steps: carrying out the preprocessing of an original electrocardiosignal, i.e., extracting an
RR interval sequence and P-wave region data of the electrocardiosignal, and carrying out the time-
frequency conversion of the P-wave region data, and obtaining a P-wave region time-frequency graph; performing multi-domain
feature extraction on the electrocardiosignals to obtain heart
rhythm feature representation, atrial activity feature representation and global spatial-temporal feature representation of the electrocardiosignals; and fusing the heart
rhythm feature representation, the atrial activity feature representation and the global spatio-temporal feature representation to obtain fused features of the electrocardiosignals, and inputting the fused features into a classification layer to obtain a
classification result of the electrocardiosignals. According to the method, collection and fusion of multi-domain features are achieved, local features and global features are combined, more complete patient representation is obtained, and then the precision of a model
classification result is improved. Particularly, when the method is applied to
atrial fibrillation classification, the
classification result is of great significance for clinically assisting doctors to make decisions.