Aiming at the communication reconnaissance problem of detection and parameter estimation of direct sequence spread spectrum(DSSS) signals in HF Channel, a new algorithm is proposed to extract cyclic frequency of cyclic cumulants by cluster analysis which named broad first search neighbors. The algorithm not only can remove the irrelevant and singular signals by comparing the differences of the intercepted signal peaks, but also overcome the ISI effects which is commonly caused by multipath and attenuation in HF Channel, thus achieves the detection and carrier frequency blind estimation of DSSS signals in the low SNR conditions. The algorithm is verified by theoretical derivation and simulation experiments, which has higher robust in low SNR conditions.