模糊融合分类器在水雷识别中的应用

Application of fuzzy fusion classifier in mine recognition

当目标的类别多时会使分类器的精度和稳健性大受影响,用神经网络分类器去完成复杂的目标分类任务是难以保证其可靠性的。引入信息融合条件下一种新型的分类器,即模糊融合分类器,该分类器可以自动“过滤”无效和冗余特征的负面影响,实现有效的特征层融合识别。采用4种特征提取方法,利用三个模糊融合神经网络作为分类器进行特征层的融合,再将分类器的输出作为决策层的融合,提高系统的分类性能。通过处理水雷实体回波数据得出的识别率表明,所选取的特征提取和分类器算法是有效的。

To desi target recognition, gn a single classifier based on a single characteristic vector is the idea of traditional but in general it cannot take essential target characteristics. Target recognition requires as many targets as possible. But, with an increased number of targets, stability and precision of the classifier is greatly affected. Therefore, traditional neural networks cannot guarantee reliability in target recognition. This paper introduces an information fusion technique and studies a fussy fusion classifier. The classifier can automatically remove undesired effects of invalid and redundant features and realize efficient feature fusion recognition. Four feature extraction methods and three fuzzy fusion neural networks are used as classifiers to carry out fusion of the characteristic layers. The classifier＂ s output is then used for fusion of the decision layer so that the performance of classification is improved. The obtained data are processed in experiments. The recognition rate shows that the algorithms of feature extraction and classifier are effective.