Obstacle Classification Of UAV Based On Millimeter Wave Radar

With the development of aviation technology,UAV is more and more widely used in military and civil fields,many industries need UAV auxiliary operation.Because of the complexity of low altitude environment and many obstacles,there are hidden dangers in the performance of UAV tasks,so it i s necessary to study the obstacle avoidance technology of UAV.Among many obstacle avoidance sensors,millimeter wave radar has obvious advantages,such as relatively low price,high detection accuracy,small size,light weight,and can work all day.Based on the above advantages,this paper uses millimeter wave radar to study the classification technology of typical obstacles encountered by UAV in the course of line patrol flight.Firstly,the millimeter wave radar is used to collect the data of three typi cal obstacles in the course of UAV’s line patrol flight: trees,electr ic poles and power lines.Then,the target echo data frame is analyzed into multi-dimensional array to lay the data foundation for the following feature extraction and classification wor k.Secondly,the radar echo data of three kinds of obstacles are prepr ocessed.One dimension and two dimension FFT are used to get the range velocity response,and the range velocity graph is drawn.Then the singular value decomposition(SVD)method is used to extract the range velocity feature of the target.On the basis of one-dimensional FFT,conventional beamforming is used to get the range angle response of echo data,and the range angle graph is drawn.On this basis,texture features based on gray level co-occurrence matrix(GLCM)are extracted,and the idea of color coherence vector(CCV)is used to extract connected region features.The concept of RRCS(root radar cross section)is introduced,and the RRCS formula is derived by RCS formula.The rationality of using target amplitude to represent the echo intensity characteristics of obstacles is demonstrated and the echo intensity characteristics of targets are extracted.Finally,based on the features of range velocity,range angle and echo intensity,a support vector machine with Gaussian kernel function as its core is constructed,SVM model and parameter optimization are carried out to classify the target based on distance velocity feature,distance angle feature and RRCS feature respectively,and compared with the target classification based on joint feature.The exp erimental results show that the classification accuracy of joint feature is basically the same as that of distance velocity feature because of the large dimension of distance velocity feature.In order to ensure the reliability and accuracy of classification results,the method of assigning different weights to different feature classification results is adopted to reduce the dominance of high-dimensional features.