In the Cognitive Radio (CR) networks, the detection performance of Primary Users (PU) randomly arriving during the sensing period is poor. So, an improved spectrum sensing method exploiting the cyclostationary feature is proposed, which adds the second half sampling values of sampling signal to the first half. The method improves the detection performance by enhancing the test statistic even though the sensing time is not added. Then, detection probability, false probability and throughput are analysed theoretically. Simulation results show that the proposed method performs better on detection performance and throughput than the conventional spectrum detection and the conventional cyclostationary spectrum sensing.