Abstract: ‍ ‍Unmanned aerial vehicle （UAV） communication systems are vulnerable to various forms of jamming due to the openness of air-to-ground wireless channels. These jamming signals will seriously degrade the performance of UAV communication and lead to reduced communication quality and potential interruptions. Therefore， to ensure the reliable and secure communication of UAVs in complex environments， the effective detection and accurate identification of various jamming are significant. Existing UAV communication jamming detection methods often assume a closed-set jamming scenario， meaning they can only recognize interference patterns that have been used during the training process and cannot effectively detect unknown jamming. However， within a complex electromagnetic environment， it is impossible to know all jamming patterns. Consequently， UAV communications are subjected to the effects of both known and unknown jamming， which is an open-set jamming scenario. In this study， we propose an open-set jamming recognition method for UAV communication based on adversarial reciprocity learning to address the jamming detection problem in a real open-set environment. First， we design a residual neural network （ResNet）-based adversarial reciprocity learning framework to effectively extract jamming features from I/Q data， which separates the known and unknown jamming well in the feature space. Subsequently， considering the differences in the feature distribution under different jamming signals and different jamming-to-signal ratios （JSRs）， we design a joint adaptive threshold classifier to accurately identify known and unknown jamming. To validate the effectiveness of the proposed method， various factors of UAV air-to-ground communication are considered， and a UAV communication interference dataset is generated based on a two-ray propagation model. Simulation results demonstrate that the proposed method outperforms the baseline methods under different JSRs. When the JSR is 10 dB， the normalized recognition accuracy of the proposed method is 88.1%， realizing effective open-set recognition of UAV communication jamming.