In order to solve the problem that traditional unmanned aerial vehicle (UAV) can not be commanded and controlled by air traffic controllers in the fusion airspace, a deep learning based UAV air-to-ground dialogue command understanding technology is proposed. The bi-directional long short-term memory (Bi-LSTM) network and the conditional random fields (CRF) are used to extract the key information of the instruction. The structured instructions that can be directly executed by the UAV are obtained, and the direct interaction between the air traffic controller and the UAV is realized. The experimental results show that this method can solve the problem of traditional interaction mode to some extent, and achieve the purpose that air traffic controllers control UAV directly by voice.