基于高光谱技术的果糖检测优化算法和可视化方法

Optimization and visualization for the prediction of apple sugar content based on hyperspectral imaging technology

为提升高光谱成像技术对果糖的无损检测精度,引进一种优化偏最小二乘(PLS)因子数的方法,以高光谱技术采集的苹果数据为基础,对比分析了全波段建模、遗传算法(GA)和连续投影算法(SPA)选择光谱特征波段建模在优化前后对果糖的预测性能。结果表明,优化方法不仅提高了模型的预测能力,同时降低了模型的复杂度,为PLS在化学计量学中的应用提供了改进方法。优化方法对全波段PLS模型的改善效果最优,预测误差均方根(RMSEP)和测试组相关系数(Rp)前分别为0.657、0.828,优化后改善至0.604、0.871。另外,本文从可视化角度对果糖含量的差异进行表征,并取得良好效果,为进一步提升高光谱检测水果内部品质的准确性提供了理论基础。

This paper presents a novel approach using hyperspectral imaging technology for fast and non- destructive detection of apple sugar contents. The optimization of partial least square （PLS） model focu- ses on a threshold in term of the PRESS derivative （the sum of squared error） in order to ignore the other inefficient factors. This novel approach is able to effectively improve the prediction performance and reduce computational complexity at the same time. A comparison of three models is conducted,including full spectra partial least square model （FS-PLS）,genetic algorithm associated with partial least square model （GA-PLS） and successive projections algorithm associated with partial least square model （SPA- PLS）. The numerical results show that the FS-PLS model is the best,which is able to achieve the root mean square error of prediction of 0. 614 and the correlation coefficient of prediction of 0. 841. The im- provement can meet the requirement in terms of accuracy of fruit inner quality using hyperspectral ima- ging technology. Meanwhile, the visualized difference of the sugar content is implemented by band mathe- matical tool with software ENVI 4. 7.