基于改进孤立森林算法的无人机异常行为检测

To further improve the control ability of low altitude UAV and break through the key technologies of accurate monitoring of cooperative UAV operation state and rapid judgment of the type of non-cooperative UAV task, a method for UAV abnormal behavior detection is studied. Firstly, the abnormal behaviors in the operation of cooperative and non-cooperative UAVs are defined. The operation parameters of the two types of UAVs are analyzed, and the extraction methods of various operation characteristics and judgment parameters are determined. Then, a UAV type determination method is proposed based on the Sobel Operator-CNN algorithm. Finally, the improved isolated Forest (iForest) algorithm based on the method of dynamic maximum growth height is proposed to distinguish the abnormal behavior of cooperative and non-cooperative UAVs. The task characteristics and abnormal types of UAVs are determined according to data nodes. The results of test based on the ardupilot-airsim experimental platform show that the improved iForest algorithm has the characteristics of fast convergence speed and high accuracy, and abnormal-behavior recognition accuracy of the algorithm is 96.4%, which is 3.2% higher than that of the traditional algorithm.