In order to solve the problem of low radio frequency spectrum utilization, a long short-term memory network (LSTM) is proposed to predict the radio frequency spectrum information over a period of time.Enables drones equipped with cognitive radios to access spectrum holes opportunistically to improve the utilization of radio spectrum.First, establish a prediction model Auto-Regressive Integrated Moving Average(ARIMA), Time Delay Neural Network (TDNN) and LSTM.Secondly, construct the corresponding prediction algorithms for the three prediction models.Finally, use the proposed LSTM prediction model to compare the prediction errors with ARIMA and TDNN, which also have time series prediction models.Prove that the LSTM network has better predictive performance in predicting time-series UAV spectrum information