基于OCT成像的淡水无核珍珠内部缺陷自动检测方法  

Automatic Detection of Internal Defects in Freshwater Nucle-free Pearls Based on OCT

作　　者：石龙杰 周扬[1] 岑岗[1] 刘铁兵 施秧 陈正伟 黄俊 汪凤林 岑跃峰 SHI Long-jie;ZHOU Yang;CEN Gang;LIU Tie-bing;SHI Yang;CHEN Zheng-wei;HUANG Jun;WANG Feng-lin;CEN Yue-feng(School of Information and Electronic Engineering,Zhejiang University of Science and Technology,Hangzhou,Zhejiang 310023,China)

机构地区：[1]浙江科技学院信息与电子工程学院,浙江杭州310012

出　　处：《计量学报》2020年第10期1226-1233,共8页Acta Metrologica Sinica

基　　金：2017年度教育部人文社会科学研究一般(规划基金)项目:“互联网+”背景下大学生创新创业环境构建研究(17YJA880004);浙江省科技计划(2017C31038,LGN19B050002)。

摘　　要：针对淡水无核珍珠质层内部缺陷检测的需求,提出一种通过光学相干层析图像实现淡水无核珍珠内部缺陷自动检测的方法。根据珠层灰度变化的特点,识别图像中缺陷区域的梯度特征和缺陷位置变化特征,并利用缺陷特征建立反向传播神经网络模型。实验中采集了内部无缺陷和内部有多种类型缺陷淡水无核珍珠的光学相干层析图像各20幅,对图像进行预处理并提取特征,利用K-means算法检测样本类型与所提取特征的匹配度,用特征与类型相匹配的样本特征训练反向传播神经网络模型,使用反向传播网络模型对淡水无核珍珠内部缺陷层进行分类识别。实验结果表明该方法提取特征的匹配度为92.5%,分类准确率达到100%,验证了该方法的可行性和有效性,提出的方法能够作为淡水无核珍珠内部缺陷识别和自动分类的有效手段。In order to expand the range of applications of optical coherence tomography technique,an automatic detection method for internal defects of freshwater nucleated pearls by optical coherence tomography is proposed. According to the grayscale change of defect layer,the proposed method extracted the gradient feature and defect location feature of the defect region in the image,and then established the back propagation neural network (BPNN) model for defection prediction based on the extracted feature. Twenty optical coherence tomography images of defect pearls and twenty optical coherence tomography of images of health pearls were collected for image preprocessing and feature extraction in experiments,and Kmeans algorithm test was used to test the feature compatibility,andthe compatible features were the input of the back propagation neural network model which finally classified the defect recognition. The experimental results show that the matching degree of feature extraction is 92. 5%,and the classification accuracy is up to 100%,which proved the feasibility and effectiveness of proposed method,and showed the proposed method can be used as an effective method for the identification and classification of internal defects of freshwater pearls.

关 键 词：计量学 光学相干层析成像 珍珠 缺陷 梯度特征 位置特征 反向传播神经网络 

分 类 号：TB96[机械工程—光学工程]