Development of Deep Learning Models for Amylose Estimation in Cereal Grains with Near-Infrared Spectroscopy

NIRS is a popular secondary analytical method that is being used for non-destructive quantification of compounds and mixtures in the agriculture and agri-food sector. The study aims to estimate the starch content (amylose and amylopectin) in rice samples with NIRS. A dataset is being established by obtaining NIRS spectra (400 to 2500 nm, 0.5 nm resolution) on over 400 milled and ground rice samples. Iodine-binding and spectrophotometric techniques will be used for acquiring the ground-truth. Upon analysis, this study would report the methodologies and evaluation metrics comparing the conventional (PLS and PCA) algorithms with deep learning (ANN and CNN) algorithms. Moreover, If the deep learning models outperforms conventional models, a Python-based data analysis pipeline will be developed for the end-users

Faculty Supervisor:

Young Chang

Student:

Prabahar Ravichandran

Partner:

Cerasoidus Analytica Inc

Discipline:

Engineering - mechanical

Sector:

Professional, scientific and technical services

University:

Dalhousie University