EFEKTIVITAS METODE NONDESTRUKTIF NIR-JARINGAN SARAF TIRUAN DALAM MENENTUKAN KOMPOSISI KIMIA JAGUNG

Abstract

High maize production should be followed by good handling and preservation up to consumer. Near Infrared Reflectance (NIR) was nondestructive testing method, as well as high accuracy, free from pollution, and rapid method, therefore suggested as a testing method. The objective of this study was asses of NIR technology efectivity in determining four major compositions of maize. Fifty samples of maize (intact seeds) were scanned from 900-2000 nm NIR wavelength, interval 5 nm. Calibration model for NIR measurement using Artificial Neural Network (ANN) technique three layers. As input layer ANN are 5, 10, and 15 nodes principal component (PC), hidden layer 4,6, 8, 10, and 12 nodes and output layer are single chemical composition and simultaneously. Prediction of an external validation set showed low the SEP (standard error of prediction) and CV (coeficient of variability). As result, NIR technology is able to predict maize chemical composition accurately SEP ranged from 0.004-0.496, CV ranged from 0.047–0.518. ANN with 5 nodes input layer and single output layer were very strong recommended to generate NIR calibration model.