基于近红外高光谱图像的冬枣损伤早期检测

为了对冬枣损伤进行早期检测,采用近红外高光谱图像技术对损伤区域成像。针对高光谱图像波长多的特点,分别采用连续投影算法、相关特征选择算法、一致性(Consistency)算法选择冬枣损伤的特征波长,对提取的特征波长分别应用k-邻近、朴素贝叶斯(naive bayes,NB)、支持向量机(support vector machine,SVM)3种分类方法进行损伤区域识别。结果表明:所有方法选择的一致特征波长在1 353 nm和1 691 nm附近。Consistency算法选择的特征波长在SVM分类器下分类识别正确率达到95.16%,一致特征波长在NB分类器下分类识别正确率达到84.26%,验证了一致波长的有效性,为多光谱成像技术实现在线检测冬枣损伤提供参考依据。

Early Detection of Mechanical Damage in Chinese Winter Jujube (Zizyphus jujuba Mill. cv. Dongzao) Using NIR Hyperspectral Images

Fruits of Chinese winter jujube (Zizyphus jujuba Mill. cv. Dongzao) are sensitive to mechanical stress and can easily develop brown spots after suffering mechanical stress during mechanical harvesting and postharvest handling. The damage cannot be detected easily by machine vision at very early stages of maturity. Thus, a near-infrared (NIR) hyperspectral imaging system was used to detect mechanical damage in Chinese winter jujubes. For reducing the dimensionality of hyperspectral data, three feature selection methods, successive projections algorithm, (SPA), correlation-based feature selection (CFS), and consistency, were used. In addition, three classifiers, i.e., k-nearest neighbor (k-NN), naive bayes (NB), and support vector machine (SVM), were evaluated to segment the pixels of the jujubes into two regions: damaged and nondamaged. Results revealed that two consistent wavebands, i.e., 1 353 nm and 1 691 nm, were established by all the feature selection methods. Besides, SVM offered the best performance with a correction recognition rate of 95.16% using the selected features by the consistency method. NB offered similar performance with a correction recognition rate of 84.26% in the selected wavebands. Hence, this work can pave the foundation for early on-line detecting Chinese winter jujube damage caused by mechanical stress.