Unmanned aerial vehicle sensor data anomaly detection using kernel principle component analysis

Unmanned aerial vehicle (UAV) has attracted more and more attention for its unique advantages and has been widely used in the military and civilian areas. With expansion of functions of UAV and increased improvement of technical complexity, UAV reliability and security are particularly essential, especially large amount of fleets and insufficient redundancy cause high operating risk and more failures. Sensors are important components in UAVs. Detecting the sensing data can monitor UAV flight condition effectively and analyze the operating status, thus, it is necessary and feasible to detect anomaly with sensing data. Due to the large number of UAV sensors and the high dimensionality of sensing data, it poses a great challenge to the sensor anomaly detection of UAVs. While kernel principal component analysis (KPCA) can effectively deal with large numbers of samples and high dimensional data, and has advantages of simple model and high efficiency of dimensionality reduction. Thus, this work proposes a KPCA based sensor data anomaly detection method. Experimental results with UAV simulated data indicates that using the proposed method to detect the UAV sensing data can obtain satisfied performance.