几何光学交互遮蔽模型（GOMS）是一种重要的遥感前向模型，它较好解释了“热点”现象，具有较强的前向模拟能力。但由于其固有的非线性性，给反演带来困难。本文尝试采用近年来兴起的并行随机全局寻优算法－遗传算法对GOMS进行反演，并针对传统遗传算法的不足进行了改进。在使用相同先验知识的条件下，将该算法与目前最有效的约束非线性最优化确定性搜索算法－逐步二次规划法对GOMS模型的反演效果进行了比较，结果表明，逐步二次规划法搜索效率较高，但结果受初值的影响很大，初值选择不当，易收敛于局部最优解，而遗传算法具有全局最优的收敛效果，但局部搜索效率较差。在某些对精度要求不高，而对搜索效率要求较高的场合，可以采用遗传算法与确定性搜索算法相结合的混合遗传算法，以提高算法的搜索效率，获得较为满意的效果。

The Li-Strahler Geometry Optical Mutual Shadow(GOMS)model is a simple, yet efficient mechanism for modeling forest canopies as arrays of three-dimensional objects. In GOMS model, the signal received by the sensor is modeled as consisting of reflected light from tree crowns, their shadows and the background within the field of view of the sensor. The model is intrinsically bound to the influence of variation in viewing and illumination geometry, and may be inverted to recover biophysical parameters. However,because the GOMS model is a nonlinear model,difficulties exist to invert it. In this paper, a Modified Genetic Algorithm(MGA) are introduced for the inversion. Compared with the deterministic search method-Sequential Quadratic Programming(SQP),MGA can quickly find promising regions of the search space, but may take a relatively long time to reach the optimal solution. In contrary, SQP can converge to an extreme value quickly, but whether the result is optimal or not depends greatly on the initial value. For this reason, a mixed method is used to invert GOMS model in some cases. The result obtained by MGA is inputted to SQP as initial value. This method significantly increases the power of MGA in terms of solution quality and speed of convergence to the optimal.