基于高光谱成像技术识别苹果轻微损伤的有效波段研究

为了筛选出适用于开发苹果轻微损伤自动分级仪器的有效波段，以200个烟台富士苹果为对象进行研究。首先获取400～1 000 nm波长范围内完好和轻微损伤后0、0.5、1 h的苹果高光谱图像，然后提取完好与损伤样本感兴趣区域的平均光谱反射率数据，再利用载荷系数法（x LW）、连续投影法（SPA）和二阶导数（second derivative）法提取特征波长，分别提取3、9和20个特征波长，并根据特征波长建立基于遗传算法优化的BP神经网络（GA BP）和支持向量机（SVM）损伤识别模型。结果显示，三种基于特征波长提取方法建立的SVM模型对测试集的识别率（分别为77.50%、91.88%、96.88%）均高于BP GA模型（分别为75.63%、90.63%、93.75%）,因此，SVM被确定为最佳苹果轻微损伤识别模型。最后，利用每一特征波长分别作为变量建立SVM模型。结果发现，波段811 nm识别率达到90.63%，优于其他波段，被确定为苹果轻微损伤识别的最优波段。

Effective Wavelengths Study on the Identification of Slight Bruises of Apples Based on Hyperspectral Imaging

In order to choose the effective band suitable for the development of automatic grading instruments for slight bruises on apples, the experiment took 200 Yantai Fuji apples as the research object. Firstly, hyperspectral images of intact and bruised samples after 0, 0.5, 1 h were obtained by hyperspectral imaging system across the wavelength rande of 400～1 000 nm, and reflectance of all pixels in region of interest (ROI) was extracted by ENVI 5.2 software. Then, the characteristic wavelengths, were extracted by different effective wavelengths selection methods including successive projections algorithm(SPA), x loading weights(x LW) and second derivative, and each method extracts 3, 9 and 20 characteristic wavelengths, respectively. The support vector machine (SVM) and BP neural network model based on genetic algorithm (GA BP) optimization were built respectively through the feature wavebands selected by three methods. The results showed that the accuracy rate of three SVM models for the test set were built through the feature wavebands were higher than the models based on GA BP. Finally, the SVM model was established respectively by using each of the characteristic wavelengths extracted by the three methods as variables. It was found that the recognition rate of the band 811 nm reached 90.625%, which was better than other bands, and was determined as the optimal band for the recognition of slight bruises on apples.