Abstract: WiFi human respiratory sensing technology has low-cost and non-contact characteristics, which can solve the problems of high cost and poor comfort in wearable respiratory monitoring and thus great development potential. Aiming at the existing respiratory monitoring technology due to the perception of the "blind spot" which leads to the decrease of respiratory perception range and perception accuracy, a new non-contact respiratory frequency monitoring algorithm WiK-Breath based on WiFi is proposed. K-Means clustering algorithm to cluster 90 subcarriers into several sets with similar characteristics, and control the number of subsequent subcarriers to be fused by a threshold. Secondly, a respiratory feature sequence was constructed by linearly combining the amplitude and phase of the subcarriers, and the respiratory frequency results of each subcarrier were weighted and fused to achieve respiratory frequency monitoring using the BNR values as the weights. Experimental results show that the proposed respiration monitoring algorithm can achieve a respiration error as low as 3.4% when the human body and the transceiver device are 7 metres apart. Compared with existing algorithms, WiK-Breath's breathing estimation error is reduced by 15.7%.