基于遗传编程的SMT机器视觉检测特征提取

在印刷电路板贴片安装的机器视觉检测中,贴片元件的型号识别和缺陷检测都是基于CCD采集的图像,数据量大、维数太多,该图像特征的提取是正确检测识别的关键技术之一.遗传编程通过遗传优化可以从原始数据或传统的高维特征中提取出更能反映类别本质的有效特征,降低特征维数、减少分类器的计算成本,同时提高分类识别精度.设计基于遗传编程的特征提取方案用于该机器视觉检测,并改进了特征评价指标.对比实验验证了本方案提取的特征分别用于ANN和SVM的良好分类识别效果.

Feature extraction of SMT machine vision inspection based on genetic programming

Feature extraction is one of the major key technologies in pattern recognition.In automatic machine vision inspection for PCB SMT assembly,components recognition and default inspection are both based on digital images,which contain too large data or too high dimensions.Effective features extracted from raw data or existing features based on genetic programming,which utilizes genetic opti-mization,can reduce features dimensions,decrease classifier's computing cost,meanwhile improves discrimination capability.A feature extraction approach based on genetic programming is developed for SMT machine vision inspection and fast feature evaluation method based on the well-known Fisher criterion is improved.Contrast experiments show effective classification results using GP ex-tracted features for ANN and SVM classifiers.