猕猴桃挤压损伤高光谱快速检测研究

目的 探究猕猴桃挤压损伤较优的快速无损判别方法。方法 利用高光谱成像系统获得所有猕猴桃的高光谱图像，并提取猕猴桃损伤区域以及完好无损区域的光谱反射率;运用多元散射校正方法对原始反射光谱进行预处理，并运用主成分分析对光谱数据降维;比较并分析Fisher判别分析方法以及简化的K最近邻(Simplified K Nearest Neighbor，SKNN)模式识别方法对猕猴桃挤压损伤的判别效果。结果 在710~850 nm和960~1 030 nm这2个波段内，猕猴桃损伤区域的平均光谱反射率与完好无损区域的平均光谱反射率存在较明显差异;采用主成分分析从256个全波段中筛选了前5个主成分作为新变量，识别模型的检测效率得到了提升;构建的SKNN和Fisher模型对预测集中样本的正确识别率均为93.3%，从SKNN识别模型的混淆矩阵中得出，预测集中仅有2个样本出现误判，并且SKNN模型对校正集中样本的正确识别率高于Fisher模型。结论 在判别猕猴桃挤压损伤时，SKNN识别模型具有相对较好的判别效果。

Rapid Detection for Pressed Damage of Kiwifruit Based on HyperspectralImaging Technology

The work aims to explore a better rapid nondestructive method of detecting the pressed damage of kiwifruit. The hyperspectral imaging system was adopted to obtain hyperspectral images of kiwifruit and the spectra reflectance in damaged region and normal region was extracted. The multi-scatter calibration (MSC) was adopted to preprocess the primary reflectance spectra and principal component analysis was employed to conduct data mining. Then, the effects of Fisher discrimination analysis and simplified K nearest neighbor (SKNN) recognition method in distinguishing the pressed damage of kiwifruit were compared and analyzed. In the spectral range of 710-850 nm and 960-1 030 nm, the average spectra reflectance in damaged region of kiwifruit was obviously different from that in normal region. The first 5 principal components were selected as new variables by PCA from 256 full wavelengths and the detection efficiency of recognition model was improved. The accurate discrimination rates of SKNN and Fisher recognition models for prediction set both reached 93.3%. Only two samples in the prediction set were not distinguished accurately from the confusion matrix of SKNN model. The accurate discrimination rate of SKNN recognition model for calibration set was better than that of Fisher recognition model. SKNN recognition model has better effect in distinguishing pressed damage of kiwifruit.