1989 年 17 巻 2 号 p. 148-157
In this paper, we propose an application of a neural network to solve the Fredhol integralequation of the first kind to yield an aerosol size distribution from multi-wavelength laser radarmeasurements. The neural network reproduces the aerosol size distribution function or its parametersfrom the aerosol volume extinction and back scattering coefficents in the laser radar equation.We used two wavelengths data (λ1=0.53μm, λ2=1.06μm) of a Nd: YAG laser for thesimulation studies. The neural network of three layers was used in two stages. The first stage isfor the classification of the type of aerosol and the second one is for the identification of a distributionfunction or parameters. Both cases were studied by simulation, and the results showedthe success in reconstructing the distribution functions under the existence of measurement errors. The effect of the initial setting for the coefficients of the neural network and that of averaging theresults from different networks were also examined.