Multi node cooperative spectrum sensing improves the sensing performan ce of the primary user by fusing the detection information of nodes in different geographical locations.However,there is a nonlinear relationship between the sensing performance and the number of sensing nodes.The increase of sensing nod es leads to the increase of energy consumption,while the nodes with less inform ation participate in cooperative sensing,which is not conducive to improving th e sensing performance,but increases the additional energy consumption.In order to improve the sensing efficiency and reduce the energy consumption,a node sel ection algorithm is proposed.The algorithm only schedules nodes with high relia bility to participate in cooperative sensing and excluding nodes with poor perfo rmance.The fusion center continuously interacts information with external envir onment through machine learning mechanism, evaluates the performance of nodes in real time,eliminates nodes with reduced reliability in time,and selects high reliability dynamically In order to maintain the dynamic balance of high reliabi lity perception and improve the robustness of cognitive network,the nodes with high reliability participate in cooperative sensing.The experimental results sh ow that the detection probability of the proposed algorithm can reach 99% under six energy consumption units,which can effectively reduce the energy consumptio n and significantly improve the sensing performance,which is far better than th e traditional sensing methods.