近赤外スペクトルを用いた自動的な果物の品質予測システムの開発 Development of on-line system for estimating fruits qualities using near-infrared spectroscopy
Access this Article
Consumers are more conscious of food quality because of recent terrible matters such as falsifying meat label or food poisoning by pesticide residue, while food factories are eager to check the quality of products quickly for good efficiency. However, in case of fruit, it is difficult to estimate its quality since it is so greatly influenced by inside conditions that it cannot be judged by appearance. In addition, many fruits are gathered into factories. In order to solve these problems, sorting machines with Near-Infrared spectroscopy become spread. But their predictive power is still low and it takes much time for measuring. We have chosen apple as a fruit, which is considered to be difficult for predicting inside quality. We constructed a regression model to predict sugar content by using Genetic Algorithm based Wavelength Selection (GAWLS) and classification models to diagnose water core and brown core by combining GAWLS with k-NN(k-nearest neighbor), based on Mahalanobis distance. Mahalanobis distance was expected to detect outlier samples and to increase predictive accuracy by reflecting similarity among spectra. Consequently, all proposed methods can extract wavelengths which are likely to be associated with each quality and achieve higher predictive accuracy.
- Proceedings of the Symposium on Chemoinformatics
Proceedings of the Symposium on Chemoinformatics 2010(0), J08-J08, 2010
Division of Chemical Information and Computer Sciences The Chemical Society of Japan