Nondestructive detection of cadmium content in oilseed rape leaves under different silicon environments using deep transfer learning and Vis-NIR hyperspectral imaging.

Abstract

In this paper, a transfer stack denoising autoencoder (T-SDAE) algorithm is proposed to implement the migration of cadmium (Cd) prediction depth characteristic model of oilseed rape leaves in different silicon environments. Stacked denoising autoencoder (SDAE) algorithm was used to reduce dimensionality, and the most effective SDAE deep learning network was transferred to create the T-SDAE model. The results showed that SVR model using SDAE to extract depth features had the best prediction effect on Cd content in silicon-free, low-silicon and higher-silicon environments. Moreover, the coefficient of determination of prediction set (R p 2 ) were 0.9127, 0.9829 and 0.9606, respectively. Specifically, the R p 2 value of the T-SDAE-SVR optimal prediction set under different silicon environments is 0.9273, RMSEP is 0.01465 mg/kg, and RPD is 3.237. By integrating hyperspectral imaging technology with a deep transfer learning algorithm, accurate detection of various Cd contents in oilseed rape leaves is feasible under different silicon environments.

Plant phenotyping relevance

油糧菜葉のCd含量という植物形質を、Vis-NIRハイパースペクトル画像と深層転移学習で非破壊推定する手法の開発・検証が中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.