The invention discloses a polarization code belief propagation decoding method based on a dynamic check matrix to solve the problems of the prior art that performance is low and time complexity is high when a traditional BP decoding algorithm is used. The method comprises the steps of 1, setting module parameters; 2, initializing parameters; 3, adjusting the check matrix to be in a low-density form; 4, executing belief propagation decoding; 5, judging whether a codon estimation vector is an effective vector; 6, calculating the minimum value of an absolute log likelihood; 7, calculating a threshold value; 8, judging whether decoding succeeds; 9, outputting a decoding result. By adjusting the polarization code check matrix and reducing the number of short loops of the check matrix, rapid convergence of the BP algorithm is achieved, and low time complexity is maintained while polarization code decoding performance is improved by means of the early-stage stop standard.