炭素制品X射线图像的缺陷特征选择

针对炭素制品X射线检测图像的特点,对缺陷特征提取与选择技术进行了研究。为排除噪声干扰的影响,采用数学形态学和迭代阈值分割相结合的方法从原始图像中提取缺陷区域。在此基础上,从缺陷样本中提取19个特征值。为提高缺陷模式识别对各种噪声及干扰的鲁棒性,提出以特征组合分类能力数学模型为适合度函数,设计基于遗传算法的特征选择策略,实现了对缺陷原始特征量的优化选择。利用BP神经网络分类器及选择的特征值对缺陷进行模式分类。研究结果表明,所提出的选择方法可以用于缺陷的识别与分类。

Defect feature selection of X-ray image on carbon product

Defect feature extraction and selection techniques were studied regarding the characteristics of X-ray detection images of carbon product.Mathematical morphology linking iteration threshold segmentation method was adopted to extract defect in order to eliminate the effect of noise.Based on this,nineteen features were extracted from defect samples.Mathematics model of feature combination classification was regarded as fitness function in order to develop pattern recognition robustness to noise,optimal selection of original flaw feature was realized with feature selection strategy based on genetic algorithm.Pattern classification of flaw was carried out with BP neural network and the feature selected.Experiment results show that the method of feature selection expounded in this paper is relatively effective and it can be used for the recognition and classification of defect.