基于激光图像次郎甜柿可溶性固形物含量检测

摘要

利用波长650 nm、功率13.25 mw的半导体激光照射贮藏期的次郎柿表面,并采集激光光斑特征响应区域图像.通过折半试探方法确定光斑区域的图像分割阈值区间后对目标图像进行分割.再分析计算目标图像分割区域(S1、S2)的像素面积参数(AS1、AS2、AS1-AS2、AS1/AS2),区域的灰度值信息熵(HS1、HS2)以及灰度值标准差(SDS1、SDS2).将以上参数作为体系的图像参数集,对次郎甜柿的可溶性固形物含量进行主成分分析(PCA).通过分析,得到对检测次郎甜柿可溶性固形物含量起主导作用的激光图像参数分量组合(AS1/AS2、HS2、SDS2).以该分量组合建立对次郎甜柿可溶性固形物含量检测的改进型支持向量机(SVM)回归模型.模型性能参数(相关系数R达到0.990 5,决定系数D达到0.870 9)和验证性试验均表明该模型具有较好的稳定性和准确性(检测SSC的准确率平均值达到94.1%,标准差为0.014).%A semiconductor laser generator with 650 nm wavelength and power of 13. 25 mW was used to irradiate the surface of "Jiro" persimmon during the storage and the characteristic laser refractive image was collected by a CCD camera. Through the midpoint subdivision method, the image region segmentation threshold was determined. Then, the image segmentation of the pixel size parameters, regional information entropy of the gray value as well as the standard deviation of gray value was calculated. The system parameters above were chosen as the parameters set. In order to get more compact model, the principal component analysis (PCA) was taken on the parameters set in the forecasting course of "Jiro" persimmon's soluble solids. Through the analysis, the most important laser image parameters were obtained for the contribution in forecasting the soluble solids content of "Jiro" persimmon. An improved SVM regression model was designed to forecast the "Jiro" persimmon's soluble solids content with the laser image parameters obtained by PCA. Both model performance parameters and verification experiments showed that the model had good stability and accuracy with the SVM related index R of 0. 990 5 and the average prediction accuracy was 94. 1 % .