CAJ | 학술논문

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。
위료대차량목표진행식별，채용료일충기우자조직경쟁망락적방법。해산법제취16개리산여현변환묘술자，6개독립적불변구화3개구역묘술자등25개평이、선전、척도방축등변환하도불변적목표형상특정，파저사혼합특정수입도설계적자조직경쟁망락진행학습、취류화분류，획득적분류정도고체96.15％，종이득출용자조직경쟁망락진행혼합특정식별，성능은정，교단일특정제취식별정도경고。
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.