Integrated deep learning neural network and hyperspectral imaging for non-destructive prediction of biochemical properties of plant organs

Abstract

ABSTRACT The methods used to analyze the medicinal properties of plants are often destructive, time-consuming, expensive and labor-intensive, making them unsuitable for fast and online quality analysis. This context has motivated the development of non-destructive methods for plant analyses. The main objective of this study is to non-destructively determine the biochemical characteristics of plants based on the integration of hyperspectral imaging and nonlinear modeling. Two nonlinear modeling methods, deep learning neural network (DNN) and regression modeling (NRM), were used to predict the active substances of plant organs according to their spectral reflectance. Two datasets, important spectra determined using reconstruction independent components analysis (RICA) and all spectra, were used to train the modeling methods in which the biochemical characteristics of the plants were target. Higher R 2 value, lower RMSE value and higher p-value indicated that the accuracy of DNN model was much higher than that of the NRM. Therefore, the combined HSI and DNN model was proposed as a suitable method for predicting the biochemical and medicinal properties of plant organs. Using the DNN model, the highest accuracy was obtained for carotenoid content estimation, its R 2 = 0.9997 and RMSE= 0.234 for all spectra, and its R 2 = 0.9952 and RMSE= 0.239 for important spectra. Nonlinear regression analysis of the spectra showed that the spectral reflectance at 538 nm was maximized for medicinal plants with higher amounts of total flavonoids, antioxidant activity and total tannin contents. The HSI analysis of plant organs and measurement the biochemical properties showed that the greatest therapeutic effect of plants was obtained for flower and leaf organs. These results concluded that the integration of HSI, RICA and NRM was successful for predicting biochemical and medicinal properties of plant organs according to their spectral reflectance.

Plant phenotyping relevance

植物器官の生化学的特性を非破壊推定するためのHSI・DNN統合手法の開発と精度評価が研究の中心であり、単なる生物学的実験のルーチン測定ではない。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.