Aflatoxin contamination is a real concern for all classes of livestock. They are produced by certain mold fungi, Aspergillus flavus and Aspergillus parasiticus. Aflatoxin in food is hazardous for humans and animals. In this work, we propose a non-invasive system for detecting aflatoxin and classifying corn kernels based on the aflatoxin contamination levels. Fluorescence hyperspectral images of single corn kernels were used for experiments. Single and multi-classifier configurations of support vector machines are used to classify single corn kernels on a per-pixel basis. The performance of SVM classification with and without feature selection is assessed. Confusion matrices of different configurations are used for comparison, demonstrating that the multi-classifier system with non-uniform feature selection performs well, achieving an overall accuracy of 84%.
Unraveling Multimodality with Large Language Models.pdf
Support Vector Machines Classification of Fluorescence Hyperspectral Image for Detection of Aflatoxin in Corn Kernels
1. Support Vector Machine Classification of
Fluorescence Hyperspectral Image for
Detection of Aflatoxin in Corn Kernels
Sathishkumar Samiappan, Lori Mann Bruce, Haibo Yao, Zuzana Hruska,
Robert L. Brown*, Deepak Bhatnagar*, and Thomas E. Cleveland*
Mississippi State University, MS, USA
*USDA-ARS, Southern Regional Research Center, New Orleans, LA, USA
bruce@engr.msstate.edu
June 2013
3. Aflatoxin in Corn
• Aflatoxin – produced by Aspergillus
Flavus and Aspergillus Parasiticus
• Invades grain crops that are stressed by
heat and drought
• In US, FDA regulates alfatoxin levels up
to 20ppb in food and 100 ppb in feed.
• Contamination of food causes financial
loss to farmers due to rejection and
disposal of grain.
[farmprogress.com]
4. Introduction
•Grains with high concentrations of Aflatoxin are toxic to
humans and domestic animals when ingested in feed.
•Carcinogen associated with liver and lung cancer in humans
•Need a rapid, non-invasive screening method for Aflatoxin
in food and feed crops
[farmprogress.com]
5. Existing Approaches
• Blacklight presumptive & thin layer chromatography
• Mini column test
• Field and laboratory rapid test kits
• Enzyme linked immuno assay (ELISA) kits
• High performance liquid chromatography
• Mass spectrometry with HPLC
ELISA Plate Kit by Beacon Kits, Inc.
6. Drawbacks with Existing Approaches
• Lack of quantitative ability
• Time consuming
• Expensive
• Require intensive interpretation
• Invasive in nature
• Require destruction of samples Technician Cesar Ambrogio sets
up the kernel screening assay to
measure amount of Aflatoxin.
USDA-ARS [USDA.gov]
8. Experimental Test Site
• Field plots were infected with
toxigenic AF13
• Corn ears were hand harvested for
imaging and lab analysis
9. Hyperspectral Imaging
•Kernels adjacent to the inoculation site were extracted
• Only whole, undamaged kernels were extracted and imaged
• Several, control kernels from the same ear away from the
inoculation site were also extracted
365nm Spectral Band
10. Chemical Analysis
for Reference Information
•After imaging, each corn kernel was
crushed and weighed
•Each sample was extracted with
methanol/water (80/20%)
•Extracted samples were diluted and
passed through Aflatest affinity
columns
•Samples were eluted with pure
methanol and measured with a
fluorometer (VICAM)
11. Image Processing
•Kernels were imaged under a UV
light source using VNIR100E
hyperspectral sensor (ITD) – 365nm
•Kernels segmented from background
•Each kernel was assigned a unique
signature
13. NU-RFS Multi-Classifier
Data
D -dimensional
KernelDensity Fusion
NU-RFS NU-RFS NU-RFS
SVM
Classifier 1
Estimate
Density
Compute
Class Score
SVM
Classifier 2
Estimate
Density
Compute
Class Score
SVM
Classifier z
Estimate
Density
Compute
Class Score
….
Final Prediction