基于分层异构模型的机载激光测深波形拟合算法

Fitting algorithm for airborne laser bathymetric waveforms based on layered heterogeneous model

波形拟合是机载激光测深数据处理的关键环节,能够为水下地形测量、海底底质分类和水体浑浊度分析等应用领域提供数据基础。针对传统机载激光测深波形拟合算法受噪声干扰严重、对复杂波形形状拟合不准确的问题,提出一种基于分层异构模型的机载激光测深波形拟合算法。针对波形不同组成部分的相应特性,采用异构函数(水面-高斯函数、水体-双指数函数及水底-B样条函数)构建分层异构模型,分别进行拟合,从而实现对各部分波形信号的拟合。采用南海实测数据对所提算法进行了验证,结果表明:该算法拟合波形的平均运行时间T为0.019 4 s,相比于RL(Richardson-Lucy)去卷积算法提高0.328 6 s;平均均方根误差(Root Mean Square Error,RMSE)为6.222 4,相比于双高斯函数拟合算法平均均方根误差RMSE、平均决定系数(Coefficient of determination,R^2)、平均相关系数(Correlation Coefficient,CORR)和相关系数标准差(Standard Deviation,STD)分别提高65.11%、2.83%、1.01%和86.61%,保证了拟合效率和拟合精度。算法具有良好的鲁棒性,能够有效满足机载激光测深科学研究和工程应用的技术需求。

Waveform fitting is a key point in data processing of airborne laser bathymetry(ALB), which can provide the foundation data for water depth calculation, submarine sediment classification and water turbidity analysis. Traditional waveform fitting algorithms are often disturbed by noise. In addition, the problem of the present algorithms is that the fitting of complex waveform is not accurate. Therefore, a new waveform fitting algorithm for ALB based on layered heterogeneous model was proposed in this paper. According to the corresponding characteristics of different components of waveform, the ALB waveforms were fitted to a combination of three functions: a Gaussian function for the water surface contribution, a B-spline function for the water bottom contribution, and a Double-exponential function to fit the water column contribution. The performance of the proposed fitting model was verified by the measured data from the South China Sea,compared with three classical waveform processing algorithms: the Double-Gaussian, Generalized-Gaussian,and Richardson-Lucy(RL) deconvolution. The experimental results demonstrate that the proposed fitting model performs best in terms of waveform fitting accuracy and efficiency. The average running time T of the proposed fitting model is 0.019 4 s, saving 0.328 6 s than RL deconvolution. The proposed fitting model performs significantly better than the Double-Gaussian algorithm by reducing 65.11%, 2.83%, 1.01% and 86.61% of their average root mean square error(RMSE), average coefficient of determination(R^2), average correlation coefficient(CORR) and average correlation coefficient standard deviation(STD), respectively. The proposed fitting model has the great robustness and can effectively meet the technical requirements of the scientific research and engineering application for ALB.