Research On UAV Anomaly Detection Algorithm Based On Semi-supervised Learning

With the rapid development of unmanned aerial vehicle(UAV),it has been widely used in various fields of society.The function and structure of drones become more and more complex with the deepening of tasks,and at the same time,the probability of drone failures is increased.UAV anomaly detection is a necessary technology to ensure that UAV can perform various tasks safely and reliably.At present,data-driven methods have good application prospects.Among them,supervised algorithms have good performance,but are overly dependent on the use of labeled data;while unsupervised algorithms do not need to label the used data,they have no test results.Can meet expectations.In response to this problem,this paper proposes a semisupervised learning(SSL)algorithm that uses both labeled data and unlabeled data,and then improves the active semi-supervised learning algorithm based on weighted sliding sampling that is more suitable for the characteristics of UAV abnormal data.First,use wavelet packet transform to obtain energy features,and train to build a support vector machine regression(SVR)model.For UAV,there is less labeled data and higher cost of labeling.At the same time,there is a large amount of unlabeled data that is not fully utilized.The SVR algorithm is used to obtain an anomaly detection algorithm based on semi-supervised learning SVR,which is used to detect anomalies data.Then,in view of the unsatisfactory effect of the anomaly detection model based on semisupervised learning SVR,the analysis is due to the unbalanced distribution of normal and abnormal data in the marked data of drones.Therefore,a semi-supervised SVR anomaly detection algorithm based on weighted sliding sampling is proposed.In the process of data sliding sampling,dynamic sampling weight adjustment is carried out,so that the sampled data distribution has more obvious learning characteristics.Use the improved model to detect abnormal data.Finally,due to the presence of low-quality data in the unlabeled data,the performance of the anomaly detection model will be reduced.Use the sample selection strategy based on sample information in active learning(AL)to label the data with larger sample information in the unlabeled data,and add it to the training set of the semi-supervised learning model based on weighted sliding sampling.Active semi-supervised learning model of weighted sliding sampling.Active learning uses less labeled data while querying unlabeled data,and uses a large amount of high-quality unlabeled data to improve the model detection effect.The number of labeled samples required by the anomaly detection model to achieve the same average detection accuracy is obtained through experiments.Comparative analysis shows that the active semisupervised learning algorithm based on weighted sliding sampling proposed in this paper has achieved better detection results while reducing reliance on labeled data.The anomaly detection algorithm based on active semi-supervised learning of weighted sliding sampling reduces the pressure of labeling data when detecting anomalies in UAV flight data,and has a better anomaly detection effect.It is not only of great significance for UAV anomaly detection,but also can provide reference for other anomaly detection fields under the data-driven.