This paper explains the feasibility of applying support vector machine based on cognitive wireless sensor network. Under condition of the wireless environment of low SNR and complex noise, aimed at the problems that single identification method fails to reach relatively accurate results, based on Hidden Markov Model, HMM, this paper optimizes the traditional spectrum sensing algorithm of SVM by adopting multiple classifiers ensemble to reduce identification error and strengthen identification robustness, and by adopting least square method to turn linear inequality constraints into linear constraints so as to get optimal hyperplane to distinguish primary signal from noise and then decide primary user state. Finally, its performance is compared with traditional energy detecting algorithm. The simulation results show that the spectrum sensing performance based on SVM is closer to the theoretical value, is more reliable and accurate than that of the energy detection, the error rate is 16％, the detection probability is 18 percent higher than the energy detection under condition of low SNR, and has more favorable detection performance and robustness.