The invention discloses a fetal electrocardiogram blind separation method based on a low signal-to-noise ratio. The method includes the steps that firstly, comb filtering is performed on multiple paths of source signals to eliminate 50 Hz power frequency interference, higher harmonic interference of 50 Hz and baseline drifts, and anti-aliasing low-pass filtering is performed on the multiple paths of source signals to remove high-frequency signal components with the frequency larger than half of sampling frequency; then, wavelet transformation is performed on the filtered source signals to obtain high frequency coefficients and low frequency coefficients of various dimensions, peak value function processing is performed on the high frequency coefficients of various dimensions to obtain high frequency coefficients containing electrocardiogram information of an expectant mother, and wavelet reconstruction is utilized for obtaining electrocardiogram signals of the expectant mother; then, the reconstructed expectant mother electrocardiogram signals containing noise are subtracted from the filtered source signals to obtain fetal electrocardiogram signals containing noise; finally, a blind separation algorithm is utilized for performing blind separation, and the multiple paths of fetal electrocardiogram signals containing noise are processed to obtain multiple paths of signals generated after blind separation. The method has great significance in monitoring perinatal fetuses, the fetal electrocardiogram extraction method is simple, calculation is fast, and accuracy is high.