当前位置: X-MOL 学术Biosyst. Eng. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Non-destructive identification of slightly sprouted wheat kernels using hyperspectral data on both sides of wheat kernels
Biosystems Engineering ( IF 5.1 ) Pub Date : 2020-12-01 , DOI: 10.1016/j.biosystemseng.2020.10.004
Liu Zhang , Heng Sun , Zhenhong Rao , Haiyan Ji

Slightly sprouted wheat kernels are difficult to distinguish with the naked eye, and the mixing of sprouted kernels into sound wheat kernels will seriously reduce the quality of wheat products. This study explored the use of near-infrared hyperspectral imaging technology to identify sound wheat kernels and slightly sprouted wheat kernels, and obtained hyperspectral images on both sides of each wheat kernel. A variety of common spectral preprocessing methods and two characteristic extraction algorithms (competitive adaptive reweighted sampling (CARS), and successive projections algorithm (SPA)) were used to combine two traditional machine learning models (linear discriminant analysis (LDA), and support vector machine (SVM)) and a special deep learning model (deep forest (DF)) to establish classification models. After analysis, it was found that the modelling effect of using the reverse side spectral data was slightly better than the ventral side spectral data, and the Savitzky–Golay smoothing (SG)-CARS-DF model was the optimal model combination. Finally, combined with actual needs, the modelling effect of the characteristic wavelengths extracted from the reverse side spectral data in the mixed spectral data sets containing different ratios of reverse side spectral data was analysed, and the results were also satisfactory. The results showed that it was better to calibrate the model with the hyperspectral data on the reverse side of wheat kernels, as this would be more helpful for identifying sound and slightly sprouted wheat kernels.

中文翻译:

利用小麦籽粒两侧高光谱数据无损识别微发芽小麦籽粒

轻微发芽的麦粒用肉眼很难区分,发芽的麦粒混入完好的麦粒会严重降低小麦产品的质量。本研究探索了利用近红外高光谱成像技术对健全麦粒和微发芽麦粒进行识别,获得每粒麦粒两侧的高光谱图像。采用多种常见的光谱预处理方法和两种特征提取算法(竞争自适应重加权采样(CARS)和连续投影算法(SPA))结合两种传统机器学习模型(线性判别分析(LDA)和支持向量机) (SVM)) 和特殊的深度学习模型(深森林(DF))来建立分类模型。经过分析,发现使用反面光谱数据的建模效果略好于腹面光谱数据,Savitzky-Golay平滑(SG)-CARS-DF模型是最优的模型组合。最后,结合实际需要,对包含不同比例反面光谱数据的混合光谱数据集中从反面光谱数据中提取的特征波长进行建模效果分析,结果也令人满意。结果表明,最好用小麦籽粒背面的高光谱数据校准模型,因为这将更有利于识别健全和轻微发芽的小麦籽粒。Savitzky-Golay 平滑 (SG)-CARS-DF 模型是最佳模型组合。最后,结合实际需要,对包含不同比例反面光谱数据的混合光谱数据集中从反面光谱数据中提取的特征波长进行建模效果分析,结果也令人满意。结果表明,最好用小麦籽粒背面的高光谱数据校准模型,因为这将更有利于识别健全和轻微发芽的小麦籽粒。Savitzky-Golay 平滑 (SG)-CARS-DF 模型是最佳模型组合。最后,结合实际需要,对包含不同比例反面光谱数据的混合光谱数据集中从反面光谱数据中提取的特征波长进行建模效果分析,结果也令人满意。结果表明,最好用小麦籽粒背面的高光谱数据校准模型,因为这将更有利于识别健全和轻微发芽的小麦籽粒。分析了反面光谱数据中提取的特征波长在包含不同比例反面光谱数据的混合光谱数据集中的建模效果,结果也令人满意。结果表明,最好用小麦籽粒背面的高光谱数据校准模型,因为这将更有利于识别健全和轻微发芽的小麦籽粒。分析了反面光谱数据中提取的特征波长在包含不同比例反面光谱数据的混合光谱数据集中的建模效果,结果也令人满意。结果表明,最好用小麦籽粒背面的高光谱数据校准模型,因为这将更有利于识别健全和轻微发芽的小麦籽粒。
更新日期:2020-12-01
down
wechat
bug