基于径向基概率神经网络的工程图纸图形符号识别

Graphic symbol recognition of engineering drawings based on radial basis probabilistic neural networks

基于径向基概率神经网络,提出一种扫描工程图纸图像分割后的图形符号识别方法.针对已分割的扫描工程图纸图形符号图像,首先进行二值化处理,然后对二值图形符号图像进行Hu不变矩特征提取,再使用一种新型的径向基概率神经网络进行分类,从而实现图像识别.为加快径向基概率神经网络的收敛速度,采用递归最小二乘算法进行训练.实验结果表明,径向基概率神经网络在识别性能与速度等方面非常适合于工程图纸的图形符号识别.

A novel graphic symbol recognition approach of engineering drawings based on radial basis probabilistic neural networks （RBPNN） is proposed. The Hu invariant moment method is applied to extract the shape features of the segmented graphic symbol image of scanned engineering drawings. The experimental results show that the RBPNN achieves a higher recognition rate and better classification efficiency with respect to radial basis function neural networks （RBFNN） and multi-layer perceptron networks （MLPN） for the graphic symbol recognition task.