基于高光谱成像和模式识别的无损检测苹果表面损伤

为了实现苹果表面损伤的快速无损检测,基于高光谱成像技术结合模式识别算法建立了苹果表面损伤检测模型。首先,利用高光谱图像采集系统采集完好无损和表面损伤苹果样本的高光谱图像,提取正常区域和损伤区域的平均光谱反射率曲线;然后,采用标准正态变换(SNV)和多元散射校正(MSC)分别对原始光谱数据进行预处理;最后,利用偏最小二乘判别分析(PLS-DA)方法,建立了苹果表面损伤SNV+PLS-DA和MSC+PLS-DA检测模型。结果表明:采用SNV和MSC光谱预处理方法可有效地消除高光谱图像中的噪声;利用SNV+PLS-DA检测模型对校正集和检验集样本的正确识别率分别为70.8%和77.5%,而采用MSC+PLS-DA检测模型对校正集和检验集样本的正确识别率分别为71.7%和77.5%。因此,基于高光谱成像技术结合模式识别方法,可实现苹果表面损伤的无损检测。

Nondestructive detection of bruises on apples using hyperspectral imaging technology combined with pattern recognition

In order to realize the nondestructive detection of bruises on app les,the detection model is established based on hyperspectral imaging techno logy combined with pattern recognition. Firstly,the hyperspectral imaging system is used to collect the hyperspectral i mages of the normal apples and the bruise apples.And the average spectral reflectance curves of the normal areas a nd the bruise areas are acquired. Then,the standard normal variation (SNV) and the multi-scatter calibration (MS C) are applied to reduce the impact of noise in spectra of apples,respectively.Finally,the partial least -square discriminant analysis (PLS-DA) is applied to build detection model of SNV+PLS-DA and MSC+PLS-DA.The results show that the noise of the hyperspectral image of apples can be effectively removed by SNV and MSC preproce ssing.The correct identification rates predicted by SNV+PLS-DA detection model for the normal app les and the bruise apples in calibration set and prediction set reach 70.8% and 77.5%,respectively.Moreo ver,the correct identification rates predicted by MSC+PLS-DA detection model for the normal apples and the bruise ap ples in calibration set and prediction set reach 71.7% and 77.5%,respectively.This study indicates that the hyperspectral imaging technology combined with pattern recognition is effective for identifying the bruises on apples.