论文部分内容阅读
提出了一种基于多层前馈神经网络的二维不变性目标识别方法。利用傅里叶描述器提取具有旋转、平移及尺度不变性的目标形状特征。由于所识别的工业工具具有一个自由度， 它们的形状有一定的动态变化范围， 导致同一目标的形状特征矢量的不唯一性。文中采用含有两个隐层的多层前馈网络学习及识别这些特征矢量。在实验中， 对四类机械工具进行测试， 并将所提出方法与最近邻分类器进行比较。结果表明， 具有反向传播（ Ｂ Ｐ）学习算法的多层前馈网络对噪音和形状特征变化具有鲁棒性， 且它还能判断未训练样本。
A two-dimensional invariant target recognition method based on multi-layer feedforward neural network is proposed. The Fourier descriptor is used to extract target shape features with rotation, translation and scale invariance. Due to the one degree of freedom of the identified industrial tools, their shapes have a certain dynamic range, resulting in the non-uniqueness of the shape eigenvectors of the same target. In this paper, a multi-layer feedforward network with two hidden layers is used to learn and identify these eigenvectors. In the experiment, four types of mechanical tools were tested, and the proposed method was compared with the nearest neighbor classifier. The results show that the multi-layer feedforward network with backpropagation (BP) learning algorithm is robust to changes in noise and shape features, and it can also determine un-trained samples.