ABSTRACT
22The aim of this study is to implement an artificial neural network (ANN) technique in order to differentiate between ripeness and unripeness stages of Citrus suhuiensis. Initially, these stages will be measured optically using nondestructive method via spectrometer of MSC600 Carl Zeiss. The spectrometer will transmit VIS (visible spectrum) photonic light radiation to the surface (skin) of the sample. The reflected light from the sample’s surface will be received and measured by the same spectrometer in terms of percentage that will form a line pattern representing each wavelength component in the VIS range. These patterns of wavelength components will be used as input parameters for designing two optimal ANN classifiers, where each one of them will be supervised with LevenbergMarquardt (LM) and radius basis function (RBF) algorithm, respectively. The performance of the optimized model is decided after observing the receiver operating curve (ROC) plot. The result outcomes have shown the optimized LM trained model has better performance in terms of sensitivity, specificity, and accuracy, and outclassed previous intelligent identification models when validated at a threshold of ±0.4. The overall accuracy for LM algorithm is 71.5% whilst the true positive rate (TPR) of each case is 54% for ripe and 89% for unripe.
