For cognitive users with low SNR at transceiver front-end, a cooperative spectrum sensing algorithm based on truncated sequential probability radio test (SPRT) is proposed to improve the sensing performance. A plit-phase processing method is employed firstly. A split-phase processing method is employed by all collaboration cognitive radio（CR）nodes firstly, which divided the receipt signals into a number of segments. The cognitive radio nodes do a count of its local energy-based sensing in their various segments and compared with the specified threshold consequently. Instead of sending all the sensing results of cooperative CR user, only the number of sampling points bigger than the threshold is sent to fusion center serving as local detection statistics. It greatly lowers transmission overhead on common control channel. Gaussian distribution is approximately obtained based on DeMoivre-Laplace theorem and central limit theorem, so the computation complexity and the deriving process of likelihood ratio function are reasonably simplified. Theoretical analysis and simulation results indicate that compared with existing energy-detection sensing method, the proposed algorithm respectively reduces the average sample number around 55.86% and 37.51% under the conditions of H0 and H1 while the average SNR of CR users is -15 dB.