基于GA-RBF神经网络及边界不变特征的车辆识别  

Vehicle recognition using boundary invariants and a genetic algorithm trained radial basis function neural network

作　　者：张涛[1] 费树岷[1] 李晓东[1] 

机构地区：[1]东南大学复杂工程系统测量与控制教育部重点实验室,江苏南京210096

出　　处：《智能系统学报》2009年第3期278-282,共5页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金重点资助项目(60835001)

摘　　要：修正的边界不变矩在目标旋转、缩放和平移过程中能保持不变性.将其作为车辆目标的识别特征,并且利用遗传算法(GA)优化径向基函数(RBF)神经网络参数,能很好地实现对车辆目标的识别.实验表明,该方法在复杂背景下对目标的识别具有很强的鲁棒性,能快速准确地识别车辆类型;并且边界不变特征的引入,减少了数据运算量,提高了识别效率.A method for vehicle recognition using the modified boundary invariant moments and a genetic algorithm trained radial basis function （GA-RBF） neural network was developed. The modified boundary invariant moments have the accustomed invariance for rotation, scaling and translation of targets, which can be used as the invariant characteristic vectors. Using these features as the inputs of a neural network, the vehicle targets can then be recognized accurately. In order to improve recognition accuracy and speed, the genetic algorithm （GA） was used to optimize the RBF parameters： centers, variance, and numbers of hidden nodes. Experimental results indicated that this method, which introduces invariants based on boundaries, yields robust target recognition with greatly reduced computation time and improved efficiency.

关 键 词：车辆识别 遗传算法 径向基函数网络 边界矩不变量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]