基于Zernike矩特征的FCM-RBF神经网络图像分类器

针对交通监控图像识别精度较差的问题,设计一种基于径向基(radial-basis)函数神经网络的图像分类器.该分类器利用Zernike矩噪声敏感度较小、形状特征稳定性好的特点,构建四阶矩的特征向量,用于特征提取;利用自适应模糊聚类方法,解决径向基函数神经网络隐层节点数不确定的问题.仿真分析表明,该分类器与基于改进的快速模糊C均值聚类算法的Back Propagation网络分类器和径向基函数神经网络分类器相比具有更高的识别率,与改进的粒子群优化模糊C均值聚类算法的径向基函数神经网络分类器相比具有相近的识别率,但其计算复杂度较低.仿真实验结果表明,该方法具有较好的分类能力及较高的计算效率.

Image FCM-RBF Neural Network Classifier Based on Zernike Moment Features

In order to solve the problem of nonhigh image recognition accuracy for traffic monitoring, an image classification was proposed based on radial basis function (RBF) neural network. Zernike array less noise sensitivity, shape features and good stability were considered to build a fourth order array feature vector for feature extraction; and an adaptive fuzzy clustering method fuzzy C means was used to solve hidden neurons uncertain of RBF neural network. The simulation analysis shows that the classifier has a higher recognition rate than the classifier based on fuzzy C means clustering algorithm of
BP and RBF neural network, a lower computational complexity than RBF neural network classifier with particle swarm of fuzzy C means clustering algorithm, though they have similar recognition rate. Simulation and experiments show that this method has better classification capabilities and higher computational efficiency.