The application of radial basis function neural network in the defect classification is studied. A training algorithm for RBF network used in pattern recognition is proposed. fabric defect characteristic parameters such as mean value. variance and entropy are extracted. and discriminate defect category using neural network, the result is that the every kind of defects of real distribution is reflected accurately as accuracy is more than ninety percent. And then the effect of defect classification using another kind of neural network -learning vector quantization network LVQ is analyzed, their training speed and accuracy of classification are compared. The experimental results show that faster classification speed and higher accuracy of RBF neural network than LVQ neural network and it is more effective to apply in the fabric defect classification.%对径向基函数神经网络在疵点分类中的应用进行了研究;提出了一种应用于模式识别的RBF训练算法,提取织物疵点的特征参数如均值、方差和熵,再利用神经网络进行疵点类别的判别,精确度高达百分之九十多,准确地反映了每一类瑕疵特征的真实分布情况;然后分析了另一种神经网络——学习矢量量化网络LVQ对疵点分类的效果,将它们的训练速度和分类精度进行了比较;实验结果表明,采用RBF神经网络比LVQ神经网络的分类速度更快、精度更高,更有效地被应用于织物疵点分类中.
展开▼
