Non-destructive recognition and classification of citrus fruit blemishes based on ant colony optimized spectral information

Abstract

Fast and accurate assessment of citrus fruit blemishes is critical to improve fruit quality and company profitability of citrus packinghouses and juice processing plants. This study aimed to identify spectral signatures of healthy fruit, and fruit exhibiting symptoms or damage from Huanglongbing (HLB), melanose, oleocellosis (oil spot), wind scar, leafminer and rust mites. Fruit samples were classified using identified spectral information. The current work proposes a characteristic waveband selection method based on the combination of the ant colony optimization (ACO) algorithm and variable selection principles. Six characteristic wavebands for each type of citrus blemishes were determined. Two different classification methods were established by the acquired characteristic wavebands, including simple layer support vector machine (SVM) classification models and tree-type SVM models. After using the tree-type SVM models, classification accuracies of healthy, HLB, melanose, oil spot, wind scar, leafminer and rust mite categories were 98.4%, 90.8%, 95.2%, 92.0%, 90.8%, 95.2% and 96.8%, respectively. The proposed characteristic wavebands selection methods were therefore very effective in extracting features of citrus fruit with these blemishes and the tree-type SVM classification models made it possible to correctly classify the fruit with high detection accuracies and universality.