论文部分内容阅读
针对红外无损检测中因特征信息缺失,致使识别与评估效果不佳这一问题,研究以铝板为对象,基于红外无损检测技术,结合主成分分析和概率神经网络对铝板正常区及三类孔洞缺陷区进行了识别与面积定量评估。研究首先采集铝板降温过程的红外时序热图,提取了正常区和各类孔洞缺陷区的时序灰度值作为初始特征。其次,采用主成分分析对初始特征进行提取,并结合概率神经网络,以像素点为单位实现孔洞缺陷的识别及面积定量评估,并采用了支持向量机进行了对比研究。实验结果表明,对于正常区和三类孔洞缺陷区测试样本的面积评估正确率分别为99.6%、97.0%、94.7%和93.0%,相比支持向量机的评估结果,所提出的研究方法具有更高的正确率。研究论证了采用主成分分析和概率神经网络,基于时序特征,以像素点为单位,实现孔洞缺陷识别和面积定量分析的有效性和准确性。
Aiming at the problem of lacking of feature information in infrared non-destructive testing and making the recognition and evaluation ineffective, this paper studied the aluminum plate as the object, based on the infrared non-destructive testing technology, combined with principal component analysis and probabilistic neural network, District has been identified and area quantitative assessment. First of all, the infrared thermogram of the aluminum plate cooling process was collected, and the gray value of sequential gray level in the normal area and various types of hole defects was extracted as the initial feature. Secondly, the principal component analysis was used to extract the initial features. Combined with the probabilistic neural network, the identification and area quantitative assessment of the hole defects were carried out in pixel units, and a comparative study was carried out by using SVM. The experimental results show that the accuracy of the area evaluation of the test samples in the normal zone and the three types of holes is 99.6%, 97.0%, 94.7% and 93.0% respectively. Compared with the SVM evaluation results, the proposed method has more High accuracy. The research proves that using principal component analysis and probabilistic neural network, the validity and accuracy of hole defect identification and area quantitative analysis are achieved based on the temporal features and pixel units.