Quality Identification of Kale Cultivated under Different Light Treatments by Hyperspectral Imaging Technology

Abstract

The development of a new method to accurately and non-destructively identify the quality of vegetables cultivated under different light treatments is urgent because traditional methods of quality identification are time-consuming, costly and destructively. A method based on hyperspectral imaging technology combined with machine learning was developed in this paper to rapidly identify the quality of kale cultivated under different light treatments in a plant factory. UV-A supplementation in different photoperiods was used to regulate the quality of kale by improving the contents of moisture, photosynthetic pigments and the phytochemicals accumulation, and the quality grades were first established based on these indicators obtained by the traditional method. Then, a non-destructive quality identification method was presented by constructing an identification model based on the hyperspectral images of kale leaves and machine learning. It was revealed that the accuracy of the identification model based on linear discriminant analysis reached a high value of 95% by employing different feature selection methods to optimize the model. The differences of model accuracy were also investigated when the reflectance spectra extracted from different regions of interest (ROIs) such as whole leaf, mesophyll and leaf veins were used for modeling. It was shown that the quality identification models had higher accuracy when the mesophyll was used as ROI than others ROIs, indicating that the selection of ROI-mesophyll was an efficient way to improve the accuracy of the model. These results demonstrate that the proposed method combining hyperspectral imaging technology and machine learning can be used to rapidly and accurately identify the quality of vegetables cultivated under different light treatments.

Plant phenotyping relevance

ケール葉の品質状態を非破壊推定するハイパースペクトル画像と機械学習の手法開発・精度評価が中心であり、植物フェノタイピング手法に該当する。

Code and data availability

The preprint describes kale hyperspectral images, quality trait measurements, and PCA/SPA-LDA/RF models, but provides no public deposit or URL for datasets, images, or code; data are available only from the corresponding author on request.

No evidence-backed public reproduction asset is currently recorded.