混浊介质的光学参数提取在光动力学疗法及光学无创诊断中有着重要意义。提出了一种实数遗传算法，结合逆蒙特卡罗方法及图形处理单元加速技术，从蒙特卡罗模拟得到的介质表面漫反射光的空间分布中提取光学参数。设计了差值平方和适应值函数、随机竞争选择算子、带扩展半径的均匀随机交叉算子、均匀变异算子、冠军变异算子，保证了算法的收敛性和种群的多样性。在0≤μa≤100 cm-1和0≤μs≤1000 cm-1范围内μa和μs提取的平均相对误差分别为0.25%和0.58%，均方根误差(RMSE)分别为0.32 cm-1和1.68 cm-1，表明实数遗传算法提取混浊介质光学参数是可行和准确的。

Determination of optical parameters of turbid media is quite useful in the photodynamic therapy and optical noninvasive diagnostics. A kind of real coded genetic algorithm incorporated with inverse Monte Carlo method and graphics processing unit based acceleration technology is proposed, which can determine optical parameters from the spatially resolved diffuse reflectance of turbid media by Monte Carlo simulation. Fitness function of accumulated square differences, random tournament selection operator, uniform random crossover operator with extended radius, uniform mutation operator, champion mutation operator are designed to guaranty the algorithm to converge with good population diversity. In the range 0≤μa≤100 cm-1 and 0≤μs≤1000 cm-1, the average relative errors are 0.25% and 0.58%, and the root mean-square errors (RMSEs) are 0.32 cm-1 and 1.68 cm-1 for the absorption coefficient and for the scattering coefficient, respectively, which means that this algorithm is feasible and accurate enough for determination of optical parameters of turbid media.