Abstract:Banana is easily infected by spoilage fungus during transportation and post-harvesting storage. Identifying process of fungus infecting banana is conducive for timely detection of potentially infected fruit and taking scientific measures for prevention and control. In this paper, the near-infrared spectra (930-1 650 nm) of banana fruit infected by Fusarium solani at different infective stages were collected. Based on the full range data, principal component analysis-support vector machine (PCA-SVM) discriminant model and partial least squares discriminant (PLSDA) model based on the original spectrum were established respectively after comparing effects of different preprocessing methods on the models. Both models achieved good results with discriminant accuracies of 83.33% and 76.67% for validation sets, respectively. Furthermore, ten characteristic wavelengths (1 117.5,1 140.7,1 146.4,1 255.5,1 284.0,1 312.5,1 403.2,1 493.2,1 498.8,1 621.5 nm) were screened out using competitive adaptive reweighted sampling (CARS) algorithm, and SVM and PLSDA models were established based on these characteristic wavelengths, respectively. The performance of CARS-SVM model was better than that of CARS-PLSDA model, with identification accuracies of 84.78% and 78.57% for training and validation sets, respectively. Results indicated that NIR spectra could be used to identity process and degree of Fusarium solani infecting banana fruit.