一种基于自组织竞争网络的车辆识别方法

Research on vehicle recognition based on self-organizing competitive network

为了对车辆目标进行识别,采用了一种基于自组织竞争网络的方法。该算法提取16个离散余弦变换描述子,6个独立的不变矩和3个区域描述子等25个平移、旋转、尺度放缩等变换下都不变的目标形状特征,把这些混合特征输入到设计的自组织竞争网络进行学习、聚类和分类,获得的分类精度高达96.15%,从而得出用自组织竞争网络进行混合特征识别,性能稳定,较单一特征提取识别精度更高。

A vehicle recognition method based on self-organizing competitive network is proposed. 16 DCT descriptors,6 independent invariant moments and 3 region characterizations are extracted to identify vehicle targets. These features are invariant under the conditions of translation,rotation,and scale change of targets. After these mixed features were input into the self-orga-nizing competitive network for learning,clustering and classification,96.15% classification accuracy was obtained. Compared with single-feature extraction method,the self-organizing competitive network based on the mixed features is faster and has higher recognition rate.