基于偏振成像技术的油桃机械损伤检测

Mechanical bruise detection of nectarine based on polarization imaging technology

为了解决由于油桃表面颜色特征复杂所带来的早期机械损伤难以检测问题,提出了一种基于偏振成像技术的早期损伤检测分类模型。采用分焦平面偏振成像方法一次性获取油桃在4个偏振方向下的偏振图像,利用双线性插值和低照度增强(LIME)对偏振图像进行预处理,以提高运行实时性并降低水果曲率变化的影响;提取偏振图像中像素的颜色特征和灰度共生矩阵(GLCM)特征,分别用于训练两个最小二乘支持向量机(LSSVM)分类模型;通过理论分析和实验仿真,最后利用两个分类模型的串联(color-LSSVM→GLCM-LSSVM model)实现了油桃机械损伤的早期检测。结果表明,该分类器模型对油桃正常和损伤区域的检测精确率达到95.68%,召回率达到93.29%。分焦平面偏振成像技术在深色系水果的早期损伤无损检测领域具有良好的应用前景。

To solve the problem that the mechanical bruise of nectarine is difficult to be effectively detected due to the complex color features of nectarine skin,a polarization imaging technology was introduced into the mechanical bruise detection of nectarines.A pixel-level bruise classification model based on polarization imaging technology was proposed.In the experiment,the division of focal plane(DoFP)polarization camera was utilized to capture the degree of polarization images in the four polarization directions respectively.Firstly,bilinear interpolation was utilized to reduce the dimension of the polarization image cube to improve the operation speed of the whole algorithm,and low-light image enhancement(LIME)was utilized to compensate for the shape of nectarine fruit and to improve the light intensity of nectarine edge area,so as to reduce the influence of fruit curvature change.Secondly,the color features and gray-level co-occurrence matrix(GLCM)features of positive(bruised)and negative(non-bruised)pixels in the preprocessed image were extracted.Then,two least squares support vector machine(LSSVM)classifiers were trained independently based on the two features.Finally,two classifiers(color-LSSVM model and GLCM-LSSVM model)were connected in series to realize bruise detection.Results show that:Two independent classifiers with radial basis function(RBF)as kernel function were used in series(color-LSSVM→GLCM-LSSVM model)with the precision of 95.68%and the recall of 93.29%.This study proves that DoFP polarization imaging technology has a prosperous application prospect in the field of non-destructive detection of mechanical bruises of dark fruits.