Low-orbit satellite networks play an important role in the global communication system, and their spectrum resources are increasingly tight. Traditional spectrum sensing methods face the problem of insufficient detection accuracy in low signal-to-noise ratio environments. In this paper, a low-orbit satellite spectrum sensing model based on cooperative multi-satellite beamforming and machine learning is proposed, which aims to improve the spectrum sensing performance and resource utilization. In the system construction, a binary hypothesis model is used to determine the signal, and the boda angle is optimized by distributed satellite cluster array to mitigate the performance degradation caused by the angle mismatch. Aiming at the deficiencies of energy detection, the study designs a clustering algorithm based on K-mean and Gaussian mixture model to classify the sensed energy signals into two categories of spectrum idle and occupied, so as to realize spectrum sensing under the condition of no a priori information. Simulation results show that when the number of cooperative sub-users is increased from 4 to 6, the classification of feature points is clearer and the sensing performance is significantly improved; in the case of an offset angle of 27 degrees, compared with the -19.87dB attenuation of the traditional LCMV algorithm, the collaborative beamforming method proposed in this paper is able to form a wave peak at the offset angle to ensure the quality of the signal processing; in the case of a normalized load of 1 and a packet copy of 3, the Under the normalized load of 1 and packet copy of 3, the maximum normalized throughput of this model using three satellites collaboration reaches 1.05bits/symbol, which is significantly better than 0.91bits/symbol in ACA scheme and 0.71bits/symbol in CRDSA scheme.The study proves that the proposed spectrum-aware model can effectively improve the efficiency of spectrum utilization and system throughput in the LOSNET.
