Enhancing the robustness of the 1-D CNN model through NIRS data augmentation based on sparse autoencoder and CARS feature selection for mango DMC determination

Abstract

Accurate, rapid, and online determination of mango dry matter content (DMC) holds great significance for the mango industry. The integration of near-infrared spectroscopy and deep learning theory offers an opportunity to enhance determination accuracy. In this paper, we propose a spectral data augmentation method based on the sparse autoencoder and establish a one-dimensional convolutional model to predict mango DMC. The test results indicate that the model performs optimally when trained on a training set comprising 80 % of the augmented data. The root mean square error (RMSE) of the test set was 0.4073, and the coefficient of determination (R²) was 0.9782. The prediction accuracy of our model surpasses that of models such as Gaussian Process Regression, Support Vector Machines, and Partial Least Squares Regression. This study can assist in fruit quality inspection, processing optimization, variety selection, and breeding, as well as storage and preservation, and has a wide range of application potential and value. It also provides novel insights into data augmentation techniques for near-infrared spectral regression modeling.

Plant phenotyping relevance

マンゴー果実の乾物含量という植物器官形質をNIRSと1-D CNNで推定する手法を開発・評価しており、形質取得・抽出法が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.