Integration of Hyperspectral Imaging System and Machine Learning to Predict Amylose Content in Rice

Abstract

Background and Objectives

This study evaluates the capability of a hyperspectral imaging (HSI) system combined with machine learning techniques as a rapid and non-destructive technology to predict the percentage of amylose content in rice. Ninety pure rice samples were procured from different geographical origins in Iran. The samples were scanned using the HSI system and then their amylose concentration was determined (based on ISO 6647-2) to create a reference database.

Findings

Spectral data were pre-processed using MSC and SG algorithms and then were fed to PCA for data reduction. Next, four machine learning methods, PLSR, SVR, MLP, and RBF, were applied to predict the percentage of the amylose content of the rice samples. Results showed that the amylose content was predicted using the PLSR with values of R²_val = 0.929, RMSE p = 0.006, and for MLP, RBF, and SVR with values of R²_val = 0.971, RMSE p = 0.43, R²_val = 0.976, and RMSEP p = 0.0038, and R²_val = 0.95, and RMSE p = 0.014, respectively.

Conclusions

The artificial intelligence algorithms, MLP and RBF, have predicted similar but better results than the SVR and PLSR methods. Therefore, the HSI system and artificial intelligence algorithms provided satisfactory results.

Significance and Novelty

The findings from this study will inform the rice supply chains that the HSI system could be used as a reliable, out-lab, and fast method for predicting the percentage of amylose content in rice samples.