moldy peanut kernel identification using wavelet spectral features extracted from hyperspectral images

Moldy peanuts contain aflatoxin, a highly carcinogenic substance that threatens the health of humans and livestock. this study aimed to identify moldy peanuts using hyperspectral measurements and continuous wavelet transform (cwt). peanuts were allowed to develop mold in a simulation of natural process of fungal infection; detailed hyperspectral images of healthy and moldy peanuts were captured. based on these spectral data, cwt with separability analysis was conducted, generating a jeffries–matusita distance scalogram that summarized the separability of the wavelet power at different wavelengths and the decomposition scales between healthy and moldy peanuts. using thresholding, five wavelet features (wfs) were isolated to identify moldy peanuts. in addition, seven optimal bands obtained from a successive projection algorithm were compared with the wfs. partial least squares discrimination analysis (pls-da) and support vector machines (svm) were adopted as classifiers for evaluating the wfs and optimal bands. the results show that according to the wfs, both pls-da and svms achieved higher overall classification results (at least 96.19% for the test data) than those using optimal bands selected via the successive projection algorithm (spa). the cwt was found to be a promising method for analyzing the fungal infection of peanuts.