The research of spectrum sensing based on compressed sensing mostly needs to reconstruct the original signal, but the accuracy of reconstruction directly affects the effect of spectrum detection. To reduce the low detection rate and computational complexity caused by this factor, a non-reconstructed spectrum detection algorithm based on compressed sensing framework is proposed, the channel is divided into multiple small channels, each small channel is sensed in turn, and the sampling covariance matrix of the divided channel group is obtained, and then combined with the energy detection algorithm to finally obtain each small channel occupation. The different compression ratios, signal-to-noise ratios, and number of users are simulated respectively. The results show that the detection rate is better when the compression ratio is above 30% and the signal-to-noise ratio is above -6dB, and the detection probability is higher in the case of multi-users. Compared with the reconstructed algorithm, the algorithm has higher detection probability, lower computational complexity and shorter detection time.