一种用阈值提取子空间的多步匹配场反演方法

A multi-step matched field inversion approach using a threshold for subspace extraction

在分析已有的匹配场反演方法的基础上,构造了一种用阈值提取子空间的多步匹配场反演方法。它根据一定反演环境下参数的不同敏感性将参数划分为子集(子空间),并依次在各敏感子空间内反演。反演时用一定的阈值将目标函数优于阈值的参数区域提取出,最后在提取出的已相对缩减的区域和最后一个子空间(通常是不敏感参数子空间)内联合反演全部参数,求得最优值。这样既可减少反演参数空间又能可靠地保证精确度,避免了已有的子空间方法反演结果受非反演参数失配影响的问题。仿真研究结果表明,本算法比已有的两类算法性能上有明显提高。

Based on the analysis of existing matched field inversion methods,a new approach of multi-step matched field inversion is proposed by using a threshold for subspace extraction.With this approach the acoustic parameters are firstly divided into several subsets（or subspaces） according to their sensitivities under a certain inversion environment,and then the inversions are in turn carried out for sensitive parameters in their subspaces.For each inversion a sub-subspace is extracted out where the values of objective functions are lower than a certain preset threshold.Finally over all the extracted sub-subspaces combined with the subspace of insensitive parameters,an inversion is performed for all parameters to find their optimal values.Thus,not only is the reduction of space realized,but also a reliable precision can be gained.This method effectively avoids the problem that the result of the previous subspace methods is much affected by mismatches of parameters that are not to be inverted.Experiments based on synthetic data show that the performance is significantly improved in contrast to other methods.