Research On Signal Processing Technology Of Full Waveform Airborne Lidar

The high sensitivity and high precision of laser radar make it widely used in military,aerospace and civilian fields.The development of digital echo technology has added new vitality to laser radar.Compared with traditional detection methods,digitized echo technology helps in the detailed understanding and classification of objects through the extraction and processing of object information,and the detailed features of object features.This thesis takes the echo signal as the research object,and carries out the theoretical and experimental research on the key problems that need to be solved in the digitized processing of the echo signal.The article summarizes the application of laser radar in various industries and the research status at home and abroad,and analyzes and compares the commonly used data post-processing technologies at home and abroad Firstly.Then,based on the introduction of airborne laser radar's principle of ground positioning and principle of distance measurement,the key factors influencing the ranging accuracy are deduced through radar equations,and echo digital technology is analyzed and optimized accordingly.Finally,a set of applicable airborne laser radar echo processing algorithms was proposed.The processing of noise is a key part of digital echo technology.In traditional denoising methods,frequency domain low-pass filtering inevitably results in distortion of the signal's detail waveform while removing high-frequency signals.However,although the empirical modal decomposition obtains the basic modal components adaptively,simply removing the high frequency components can also cause distortion,and additional operations are required to retain the detailed features and increase the complexity of the algorithm.In order to retain the details of the waveform while minimizing the noise and ensure the effect of wavelet denoising,this manuscript chooses the method of wavelet denoising,and chooses the wavelet parameters that are most suitable for this data through comparison analysis of SNR,RMS error and smoothness.The selection of function models in waveform decomposition is another key issue of digital echo techniques.By comparing the Gaussian function,the generalized Gaussian function,and the skewed normal distribution function,This manuscript selects a more generalized skewed normal distribution function.In order to estimate the component number and component parameters,this manuscript compares the most common inflection point method and layer stripping strategy at present.The analysis results show that the inflection point method is more susceptible to the filtering effect,there are many pseudo inflection points,and it is often necessary to apply the inflection point.The smoothing and thresholding are performed before the method,and then the pseudo-inflection point is judged,thereby ensuring the accuracy of the estimation.The layer stripping strategy is not so high,only one component is stripped from the original data each time until the requirements are not satisfied,thus simplifying the algorithm and obtaining better results.Finally,the optimization parameters are optimized by the optimization method.The LembergMarquardt(LM)optimization algorithm is an initial value-sensitive algorithm.In order to ensure the non-singularity of the Hessian matrix,so as to ensure the iterative operation of the LM optimization,this manuscript uses the swarm intelligence algorithm to pre-optimize the estimated characteristic parameters.And using the crowd search algorithm to perform well in the multi-peak fitting problem,using the crowd search algorithm to perform the pre-optimization on the basis of the LM optimization algorithm.At the same time,in order to ensure that LM optimization is a global convergence algorithm,Armijo imprecise search is added in the iterative process to ensure the global convergence ability and effectively avoid falling into a local optimum.By using data verification for the algorithm,the correlation coefficient between the fitting waveform and the original data is as high as 0.9951,thus verifying the effectiveness of the algorithm.