Hyperspectral Image (HSI) Analysis of Wheat Kernels for Deoxynivalenol Quantification

Hyperspectral Image Analysis of Wheat Kernels for Deoxynivalenol Quantification uses machine-learning based methods to classify of peanut stem rot.

Introduction

In this study, we used Fusarium head blight (FHB) a fungal disease in grain kernels caused by Fusarium graminearum, as a model system to evaluate the ability of HSI for the disease classification and quantification of Deoxynivalenol (DON), a mycotoxin accumulated in kernels. Experiments were carried out to determine which machine learning methods had the best accuracy to classify different classes of kernels based on the DON content obtained from Gas Chromatography–Mass Spectrometry (GC-MS). Results from Mask-R-CNN, when combined with best classification methods, can be used to correlate the HSI data with the DON content in small grains with R2 =0.7497. This shows that we can apply HSI to quantify the DON content in wheat kernel.

Dependencies and requirements

Python 3.8.6 scipy: 1.6.0 numpy: 1.18.5 matplotlib: 3.3.3 pandas: 1.2.1 sklearn: 1.0.2 xgboost: 1.5.1 seaborn: 0.11.1

Instructions

Folder description

○ Input: This folder stores input files obtained from HSI. ○ Script: This folder stores scripts developed by the project team for the HSI analysis. ○ Output: This folder stores the output files generated by the scripts from the inputs.

Script description

○ Fig2*. ipynb: This script is to compare different machine learning algorithms three class classification viz Background, Healthy kernels and Infected kernels and further compare between healthy and infected kernels’ pixels by selecting 200 random pixels from each class.

○ Fig3*. ipynb: This script is to compare different machine learning algorithms two class classification viz Background and Kernels and further compare between healthy, mild and severe kernels’ pixels by selecting 200 random pixels from each class.