Aerosols significantly affect the transmission of optical signals in the atmosphere, necessitating accurate atmospheric models for the performance evaluation of electro-optic devices. Aerosol size distribution is a critical parameter in these models, and the lognormal function is commonly used to mathematically represent it. This study aims to handle the lack of a solid criterion for determining the number of lognormal modes and introduces an improved scheme that leverages the characteristics of the second derivative (SD) of the Gaussian curve to identify the mode amount and to initialize mode parameters for fitting. The optimization problem is solved using a genetic algorithm, incorporating a goodness-of-fit index to determine the presence of spurious modes. For aerosol size distributions characterized by a single Gaussian peak, mode parameters such as mode radius and width can be straightforwardly identified through the positions of peaks and roots on the SD curve. However, the original mode dissociation method may overlook potential modes in distributions composed of superimposed Gaussian peaks. Numerical tests indicate that such oversights can result in substantial errors in calculating the aerosol extinction coefficient, with relative errors exceeding 100%. The proposed scheme significantly enhances the accuracy of mode dissociation in aerosol size distribution, reducing errors in aerosol extinction coefficient calculations by approximately 40% when applied to data from the Aerosol Robotic Network (AERONET). An additional benefit of this method is its ability to constrain the number of lognormal modes in an aerosol size distribution. Results from applying this scheme to data from selected AERONET sites reveal that over half of the size distributions consist of more than two lognormal modes, highlighting the effectiveness of the proposed approach in capturing complex aerosol behaviors.