Research On Target Recognition Of Multi-rotor UAV Based On Kernel Method

In recent years,small UAVs have been widely used in military reconnaissance equipment detection,civilian aerial performances,forest fire monitoring,agricultural planting and fertilization,etc.But at the same time,unauthorised flying of drones into the aviation field to disrupt airport order has caused harm to people’s safety and the national economy.Therefore,accurately judging the type of UAV is of great significance to the national air defense security.At present,combined with radar technology,the UAV identification method based on linear features can achieve good results.However,in practical applications,due to the complexity of the target and the influence of environmental noise,there is obvious nonlinearity in the target data distribution.This thesis by introducing the kernel method,the research on the nonlinear feature recognition method is carried out.The main contents are as follows:（1）An identification method of multi-rotor UAV based on the maximum class correlation feature of kernel principal components is proposed.The method uses canonical correlation analysis to extract features that are highly correlated with category information in the kernel principal component features,fuses the category information of the data,and improves the target recognition rate.The simulation results show that the average recognition rate of this method is about 1.18% and 6.1% higher than that of the kernel principal component feature extraction method and the linear principal component feature extraction method under the condition of a signal-to-noise ratio of 5d B,respectively.（2）An identification method of multi-rotor UAV based on the kernel weighted maximum distance discriminant feature is proposed.This method calculates the euclidean distance between the center vectors of different classes in the kernel space,selects a monotonically decreasing function of euclidean distance to weight the inter-class divergence matrix,and strengthens the role of similar classes in finding the best projection direction.The ability to distinguish similar classes is enhanced,and the recognition rate is improved.The simulation results show that the average recognition rate of this method is better than that of the kernel maximum distance discriminant feature extraction method,the conventional kernel discriminant feature extraction method and the linear discriminant feature extraction method under the condition of a signal-to-noise ratio of10 d B.The rate increased by about 2.13%,3.13% and 9.99% respectively.（3）An identification method of multi-rotor UAV based on optimal two-dimensional kernel principal component features is proposed.Firstly,the kernel two-dimensional principal component eigenvectors of the spectral samples in the high-dimensional feature space are obtained,and the eigenvectors with larger classification information are extracted from them to form a projection matrix,and the identification features with the best classification performance are obtained.At the same time,this feature preserves the spatial structure information of the image,which further improves the recognition rate.The simulation results show that the average recognition rate of this method is better than the average recognition rate of the kernel two-dimensional principal component feature extraction method,the kernel one-dimensional principal component feature extraction method and the linear two-dimensional principal component feature extraction method under the condition of a signal-to-noise ratio of 5d B.The rate increased by about 1.34%,2.16% and 4.14% respectively.