Purpose The aim of this study was to fi nd the optimal detection method for cucumber powdery mildew and improve the identifi cation effi ciency. Methods Image segmentation technology was used to extract spot images and grade classifi cation of powdery mildew. The relationship between powdery mildew spot and spectral refl ectance and intensity was studied. The powdery mildew detection model was established by using qualitative analysis and quantitative prediction methods combined with greenness ( a* ) indices of cucumber leaves. Results There were strong positive correlations between greenness and spectrum in some characteristic bands. Through the extraction of disease spot images and disease classifi cation, it was found that the higher the disease grade of leaves was, the higher the spectral refl ectivity and fl uorescence intensity. In the quantitative prediction model, the R 2 of the NIR spectrum was improved (0.8742) after MSC and SPA, and the eff ect of the fl uorescence spectrum model was not ideal. In the qualitative discriminant model, KNN and ensemble subspace discriminant were obtained for two kinds of spectra, and the identifi cation accuracy of the qualitative model was 97.5% after verifi cation. Conclusion An NIR spectral model can be used for the quantitative prediction of cucumber powdery mildew. The qualitative discriminant model had high accuracy for cucumber powdery mildew.