利用高光谱成像技术对库尔勒梨早期损伤进行快速识别检测。 以60个库尔勒梨为研究对象, 采集380～1 030 nm波段范围内完好样本和损伤后1～7天样本的480幅高光谱图像。 提取图像中感兴趣区域（ROI）的平均光谱信息, 利用小波变换（WT）对光谱数据进行去噪平滑, 将去噪后的全部样本按2∶1的比例分成建模集（320个）和预测集（160个）。 利用二阶导数从全谱信息中提取出19个特征波长, 分别基于全谱和提取出的特征波长对建模集和预测集进行支持向量机（SVM）建模分析。 结果表明, 基于全谱和特征波长的判别分析模型中, 两者预测集的识别率都达到93.75%, 表明提取的特征波长包含了光谱数据中的关键信息。 然后, 基于特征波长运用波段比运算挑选最佳波段比, 根据波段比F值的分布确定光谱图像分割的最佳波长684和798 nm。 对最佳波段比（684/798 nm）下的图像, 利用选择性搜索（SS）对高光谱图像中样本的完好和损伤区域进行分割, 从分割结果来看, 1～7天损伤样本的受损区域能够被准确检测出来。 研究结果表明: 基于高光谱成像技术对库尔勒梨进行损伤鉴别是可行的, 该研究所获得的特征波长和波段比为研发在线实时的库尔勒梨损伤检测系统提供支撑。

In this paper, hyperspectral imaging combined with chemometrics was applied for the detection of internal defects of Korla pear. The hyperspectral images covering the spectral range of 380～1 030 nm were acquired for 60 Korla pears before, and seven consecutive days after internal damages were induced by being dropped from a distance of 30 cm. The mean spectrum were computed from region of interests (ROI) of pear in each image, and was preprocessed with wavelet transform for eliminating system noise and external disturbances, and optimizing the spectral identification region (470～963 nm). Based on the preprocessed samples, the support vector machine models were built respectively through the full and feature wavebands selected by the second derivative. The results on testing set demonstrate that both of the two approaches achieved the discrimination accuracy of 93.75%. Furthermore, F-value based method was applied for image analysis to find out the optimal waveband ratio for the visual discrimination of bruises against normal surface. Based on the optimal waveband ratio images, the selective search algorithm was utilized for segmenting bruises from the pear surface, and shows the accurate identification results. Our research revealed that the hyperspectral imaging technique for detecting bruised features in pears is feasible, which could provide a theoretical reference and basis for designing classification system of fruits in further work.