Skip to main content
ARS Home » Southeast Area » Athens, Georgia » U.S. National Poultry Research Center » Quality & Safety Assessment Research » Research » Publications at this Location » Publication #318954

Research Project: Optical Detection of Food Safety and Food Defense Hazards

Location: Quality & Safety Assessment Research

Title: Near-infrared hyperspectral imaging for detecting Aflatoxin B1 of maize kernels

Author
item Wei, Wang - China Agriculture University
item Lawrence, Kurt
item Ni, Xinzhi
item Yoon, Seung-chul
item Heitschmidt, Gerald - Jerry
item Feldner, Peggy

Submitted to: Food Control
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/29/2014
Publication Date: 12/10/2014
Citation: Wei, W., Lawrence, K.C., Ni, X., Yoon, S.C., Heitschmidt, G.W., Feldner, P.W. 2014. Near-infrared hyperspectral imaging for detecting Aflatoxin B1 of maize kernels. Food Control. Volume 51, Pages 347-355, May 2015.

Interpretive Summary: Maize kernels can be contaminated by aflatoxins, a type of mycotoxin produced by fungi species, Aspergillus flavus and Aspergillus parasiticus. The most toxic aflatoxin causing a naturally occurring microbial carcinogen is known to be aflatoxin B1 (AFB1). A study was conducted to develop a non-destructive and rapid near-infrared hyperspectral imaging technique in the wavelength range between 1,000 and 2,500 nm, to detect maize kernels contaminated with AFB1. Maize kernels were inoculated with Aspergillus flavus to cause AFB1. The results of the study showed that the spectral angle mapper classifier combined with principal component analysis of near-infrared hyperspectral images could detect the presence of AFB1 in maize kernels with 92 percent accuracy.

Technical Abstract: The feasibility of detecting the Aflatoxin B1 in maize kernels inoculated with Aspergillus flavus conidia in the field was assessed using near-infrared hyperspectral imaging technique. After pixel-level calibration, wavelength dependent offset, the masking method was adopted to reduce the noise and extract region of interest (ROI's) of spectral image, then an explanatory principal component analysis (PCA) followed by inverse PCA and secondary PCA was conducted to enhance the signal to noise ratio (SNR), reduce the dimensionality, and extract valuable information of spectral data. By interactive analysis between score image, score plot and load line plot, the first two PCs were found to indicate the spectral characteristics of healthy and infected maize kernels respectively. And the wavelengths of 1729 and 2344 nm were also identified to indicate AFB1 exclusively. The n-dimensional visualization method based on PC3 to PC7 was adapted to select the two classes of end members as the input data of the spectral angle mapper (SAM) classifier to separate the aflatoxin infection and clean kernels. The result was compared with chemical analysis of Aflatest®. And the verification accuracy of pixel level reached 100 percent except the tip parts of some healthy kernels were falsely identified as aflatoxin contamination. Furthermore, another 26 maize kernels were selected as an independent data set to verify the reproducibility of the method proposed, and the detection accuracy attained to 92.3 percent, which demonstrated that hyperspectral imaging technique can be used to detect aflatoxin in artificially inoculated maize kernels in the field.