Black spot is one of the fungal diseases of Korla pear. It is of great significance to realize early diagnosis of black spot disease before the symptoms are evident, as it can prevent the spread of the disease and reduce the economic loss. Hyperspectral imaging technology was combined with Stacking ensemble learning algorithm to construct early and rapid diagnosis model of Korla pear black spot. Hyperspectral images of healthy, incubation period, mildly diseased and severely diseased Korla pear were obtained, and the average spectra in the region of interest were extracted. After pretreated by standard normal variable transformation, the first derivative, second derivative and their combinations, principal component analysis was implemented to reduce the data dimension. Then, the Stacking ensemble learning prediction model for black spot disease was constructed with K-nearest neighbor method (KNN), least squares-support vector machine (LS-SVM) and random forest (RF) algorithm as the base learner and LS-SVM as the meta-learner. The results showed that with the deepening of the disease degree, the reflectance spectra showed a downward trend, significant difference was observed, which provided a theoretical basis for the establishment of classification models. The total classification accuracy of healthy and different disease degrees of Korla pear was 98.28%, and the classification accuracy for incubation period pear was 100%. Compared with the results using single classifier, the classification accuracy for all pear and incubation period pear was increased by 5.18 and 23.08 percentage points, respectively. The results showed that Stacking ensemble learning had strong feature learning ability, and its combination with hyperspectral imaging technology can realize the recognition of incubation period of black spot in Korla pear. The results can provide a method for the early diagnosis and real-time monitoring of black spot of Korla pear.