基于遗传算法的目标声信号特征选优

Target Acoustic Characteristics Optimization Based On Genetic Algorithms

采用各类特征提取技术处理目标声信号,至少可以获得数十种不同的特征量,对于战场声识别系统而言,选用哪几种特征量作为分类器的输入向量是一个非常重要的问题。经分析,目标声信号特征选择可以描述为一个带约束条件的优化问题,在目标函数和约束条件确定后,为了设计准确、高效的搜索算法成为特征选择的关键。根据目标声信号和分类器的特点,设计了目标声特征选优的遗传算法,将可获得的各类特征组成一个基因链码,在保持特征向量维数不变的条件下,随机选择基因链码交叉截断点和变异点。算法具有搜索快、效率高的特点,经计算机仿真证明,搜索结果准确、可信,根据搜索结果组成的输入向量,分类器可以准确高效地识别目标。

More than tens of characteristics can be acquired in acoustic signal processing of targets using various DSP technologies. Importantly, for a battlefield acoustic reconnaissance system, it is critical to select what kinds of characteristics to make up the input vector for the classifier. According to our analysis, the selection of the target acoustic characteristics can be described as an optimization problem with constraints. The key step of target acoustic characteristics selection is the design of an accurate and effective algorithm if the objective function and the constraint of the optimization problem have been determined. Based on the feature of both the acoustic signal and the classifier, the genetic algorithm of target acoustic characteristics selection is developed. The genetic code is formed by all available characteristics, and the crossover point and mutation point of genetic code arc random selected, meanwhile, the dimension of the feature vector is fixed. The developed algorithm seeks out results quickly and effectively. According to the computer simulation, the search results are accurate and credible. Finally, using the input vector composed of the search results, the classifier can recognize targets accurately and effectively.