基于人工神经网络的机械工具的识别方法

提出一种快速的、鲁棒性的人工神经网络目标识别方法。针对工业环境中常用机械工具的识别，通过正规化目标图像和正交复值的Ｚｅｒｎｉｋｅ矩变换提取目标的平稳我、比例及旋转不变性特征，应用具有２个隐层的ＢＰ网络学习与识别这些特征敌意一。对４类具有一个自由度的机械工具进行识别实验，表明该方法优于最近分类决策规则，对噪声及循环矢量变化具有鲁棒性，并达到９５％的识别率。

Object Recognition of Mechanical Tools Based on Neural Networks

In this paper, a robust and fast approach to recognise industrial tools based on neural networks is developed. The feature vector that is invariant to translation, rotation and size scaling are first extracted by normalization and complex orthogonal Zernike moments. The next stage,a BP neural network with two hidden layers classifies the extracted features. The recognition experiment involving four kinds of tools with one DOF shows our method outperforms the nearest-neighbor rule widely used in pattern recognition. Performances are also compared under noisy environments and for some untrained objects.