Monitoring Botrytis cinerea Infection in Kiwifruit Using Electronic Nose and Machine Learning Techniques

Gray mold, caused by the Botrytis cinerea fungus, is the most common and destroying disease in kiwifruit during storage. In this study, an experimental electronic nose system combined with machine learning (ML) approaches were utilized for early detection and monitoring of Botrytis cinerea in Hayward kiwifruit based on the odor-extracted information. Principal component analysis (PCA) and correlation-based feature selection (CFS) methods were used to extract the most significant electronic nose features. These characteristics were then introduced to the modeling algorithms for classification and prediction purposes. Results revealed that the radial basis function neural network (RBFNN), trained with CFS-selected features, outperformed the other ML algorithms in classifying non-affected and affected kiwifruit samples with accuracies of 99.9% and 100% on the training and test datasets, respectively. For the classification of kiwifruit samples based on the day after fungal inoculation, the combination of CFS and multilayer perceptron neural network (CFS-MLPNN) achieved the highest classification accuracies of 99.9% and 100% in the training and test phases, respectively. The CFS-MLP model was also the superior predictor of variations of fruit firmness (training R-2 = 0.9641, test R-2 = 0.9702), soluble solid content (SSC) (training R-2 = 0.9672, test R-2 = 0.9890), and titratable acidity (TA) (training R-2 = 0.9637, test R-2 = 0.9525) during fungal infection. It was concluded that the artificial olfactory system coupled with artificial intelligence could be efficiently applied for early detection and monitoring of the kiwifruit fungal infection during postharvest storage.