Multilevel fusion of multispectral images to detect the artificially ripened banana

Abstract

Automatic detection of artificially ripened fruits based on a nondestructive approach has recently gained significant attention. This work explores the inherent properties of multispectral imaging to distinguish between natural and artificially ripened bananas. The proposed method combines the prediction scores computed from the support vector machine on the individual and fused spectral bands images to detect the artificially ripened banana. Extensive analyses are performed on 5760 banana images captured in eight different spectrum bands covering visible and near-infra-red ranges. Obtained results indicate the average detection accuracy of 97.1 plus or minus 3.6 percent, thereby illustrating our proposed work's applicability.