Wind Field Images Generation And Feature Indentification Based On Coherent Lidar

As a sudden atmospheric wind field phenomenon,wind shear will seriously affect the normal driving of aircraft and is one of the main reasons that threaten the safety of civil aviation.Because the wind speed information in the wind field can be reflected in the form of a pseudo-color image,researchers can analyze the characteristics of the wind field image to determine whether there is wind shear in the wind field and identify its type,thereby assisting the pilot to make correct operations.Therefore,it is extremely important to study wind field image generation and feature analysis methods.In this paper,based on coherent lidar,the research of radar data preprocessing,wind field inversion,wind field image generation,wind shear detection and type recognition are studied.First,as the basis of wind field image generation,the radar echo data preprocessing method is analyzed.This paper first studies the denoising method based on range gate coherent lidar echo data.The denoised coherent lidar echo signal can be used to calculate the radial wind speed of the wind field.The solution of radial wind speed belongs to the problem of frequency estimation of weak signals.In this paper,the method of combining maximum likelihood discrete spectral peak estimation and frequency domain incoherent cumulative estimation is used to calculate the radial wind speed in different measurement directions and form a wind field radial wind speed matrix..Subsequently,the wind field inversion method based on linear and nonlinear interpolation is analyzed,and it is considered that a dense radial wind velocity matrix is obtained,and the wind field image is generated using software.In order to solve the problem of insufficient coherent lidar measured data and the lack of wind shear types in real data,this paper uses simulation methods to construct coherent lidar measurement data to prepare for the training of subsequent wind shear recognition models.Finally,in terms of wind shear type recognition,this paper proposes three wind shear recognition models based on machine learning algorithms,among which the wind shear recognition model based on semi-supervised learning to generate adversarial networks has the best performance and its recognition rate When the number of training set samples is 500,it reaches 97.86%,which proves that the model can accurately identify the type of wind shear in the simulated wind field.Aiming at the problem that the training set samples can only be obtained by random sampling,and the most valuable samples cannot be selected for labeling,this paper adds an active learning algorithm to the training process of the wind shear recognition model.Active learning will automatically select unlabeled samples according to the corresponding scoring rules and perform manual marking work.In this paper,different active learning algorithms are involved in the training process of the wind shear recognition model,so that the recognition ability of the wind shear recognition model is further improved.The experimental results show that the recognition rate of wind shear recognition model based on semi-supervised GAN network is further improved to 98.97% with the help of active learning algorithm.This paper believes that the proposed algorithm has a good performance in identifying simulated wind shear data,and provides a reference and reference value for subsequent related research based on the measured data of coherent lidar.