基于高光谱成像技术和MNF检测苹果的轻微损伤

苹果损伤是一种发生在水果采摘和产后处理阶段的不可避免的主要缺陷。为了快速有效地识别苹果的轻微损伤，以具有代表性的双色红富士苹果为研究对象，提出了一种以高光谱成像和最低噪声分离(MNF)变换的苹果轻微损伤识别检测方法。首先，使用高光谱成像系统获取苹果的可见-近红外波段(400~1 000 nm)的图像，对比发现全波段的最低噪声分离变换比主成分分析(PCA)变换可获得更好的识别效果;其次，利用I-RELIEF算法对正常表皮和损伤区域的光谱进行分析得出权值系数图，依据该系数曲线挑选出了5个特征波段(560，660，720，820和960 nm);最后，特征波段和最低噪声分离变换开发了损伤苹果的识别检测算法。利用该算法对80个正常苹果和含有不同时间阶段轻微损伤的苹果进行试验，损伤识别总体正确率为97.1%，试验结果表明，利用该方法和选取的特征波段可以快速有效地识别苹果的早期轻微损伤，为利用多光谱成像技术和最低噪声分离变换在线检测苹果轻微损伤奠定了基础。

Detection of Slight Bruises on Apples Based on Hyperspectral Imaging and MNF Transform

Bruising is one of the major defects occurring on apple surface inevitably during postharvest handling and processing stage. To detect slight bruises on apples fast and efficiently, a novel bruises detection algorithm based on hyperspectral imaging and minimum noise fraction transform is proposed. First, the hyperspectral images in the visible and near-infrared (400~1 000 nm) ranges are acquired, and MNF transform based on full ranges could obtain better detection performance compared to PCA transform; Second, five wavebands (560, 660, 720, 820 and 960 nm) are selected as the effective wavebands based on the coefficient curve of I-RELIEF method conducted on spectra extracted from intact and bruise surface; Third, the bruises detection algorithm is developed based on the effective wavebands and MNF transform method. For the investigated 40 sound samples and 40 different time stage bruise samples, the results with a 97.1% overall detection rate are got. The recognition results indicate that the proposed methods and the effective wavelengths selected in this paper are feasible and efficient. This research lays a foundation for the development of multispectral imaging system based on MNF transform for slight bruises detection on apples.