基于高光谱成像提取苹果糖度与硬度最佳波长

Optimal Wavelengths Extraction of Apple Brix and Firmness Based on Hyperspectral Imaging

利用高光谱成像技术提取可同时检测苹果糖度与硬度的最佳波长。首先双面采集苹果的高光谱图像,获取亮度相近感兴趣区域(ROIs)的反射波形,采用二阶导数结合标准正态变量(SD+SNV)的方法平滑波形,测试ROIs的糖度与硬度;之后采用连续投影算法(SPA)提取两项指标的特征波长,根据特征波长的分布提出二次连续投影算法,结合波形集特征与两次投影结果确定不同取样面的最佳波长;最后采用遗传算法开发神经网络(GA-BP)建立预测模型,双面取样波长(543 nm和674 nm)效果最优,糖度相关系数(R)为0.847 6,均方误差(MSE)为3.32;硬度R为0.793 8,MSE为9.6。结果表明,相同波长信息可以检测苹果糖度与硬度。

Hyperspectral imaging technology was brix and firmness test. Firstly, the hyperspectral used to extract the optimal wavelength for apple images of apples were acquired from double-sided sampling. The reflection waveforms of the regions of interest （RIOs） with similar brightness were ac- quired and smoothed by the second derivation and standard normal variate （ SD ＋ SNV ） method. The brix and firmness values of RIOs were also tested. Then, the characteristic wavelengths of two indicators were extracted by using the successive projections agorithm（SPA）. According to the dis- tribution of characteristics wavelengths, two times SPA was proposed. Combined the feature of wave- forms and the results of two projections, the optimal wavelengths of different sampling facets were determined. Finally, the genetic algorithm for back propagation（GA-BP） was used to build the pre- diction model. The best results were obtained from the double-sided sampling wavelengths （543 nm and 674 nm）. The correlation coefficient of brix （R） is 0. 847 6 and the mean square error （MSE） is 3.32, and for the firmness, R is 0. 793 8 and MSE is 9.6. The results show that the brix and firmness can be detected by the same wavelength information.