Detection of artificially ripened banana using spectral signature from multi-spectral imaging

Abstract

Ripening is a natural process of fruit maturation by which it attains desirable texture, aroma, colour and becomes more flavoursome. To meet the increasing needs of consumers, traders have resorted to artificial ripening of fruits. The ripening agent such as industrial-grade Calcium carbide (CaC2) degrades the overall quality of the fruit. In addition to CaC2 being a known carcinogen, it also contains traces of arsenic and phosphorus that can result in serious ramifications for human health. While detection may be possible by conventional methods such as chemical analysis or visual inspection, they may not be quick enough or convenient and thus not feasible. In this paper, we detect an artificially ripened banana non-invasively using multi-spectral imaging in eight narrow bands across the Visible (VIS) and Near-Infra-Red (NIR) spectrum, we construct multi-spectral images collected from artificially and naturally ripened samples of bananas. On a large scale data set consisting of 5760 samples, an experimental evaluation was conducted by performing 10 fold cross-validation. The average classification accuracy of 94.41 ±4.70% based on spectral signature shows the significance of using multi-spectral images for detecting artificially ripened banana.