Aflatoxin detection in whole corn kernels using hyperspectral methods

Hyperspectral (HS) data for the inspection of whole corn kernels for aflatoxin is considered. The high-dimensionality of HS data requires feature extraction or selection for good classifier generalization. For fast and inexpensive data collection, only several features (lambda responses) can be used. These are obtained by feature selection from the full HS response. A new high dimensionality branch and bound (HDBB) feature selection algorithm is used, it is found to be optimum, fast and very efficient. Initial results indicate that HS data is very promising for aflatoxin detection in whole kernel corn.