Plasmopara viticola (P. viticola)infection poses a serious threat to grape production. Early prevention and treatment is essential to the control of P. viticola infection. In order to detect this disease early, the relative biomass of P. viticola detected by PCR as the basis of P. viticola infection, the chlorophyll fluorescence images of 80 grape leaves inoculated with P. viticola and 80 healthy control leaves were collected for 6 consecutive days from the three continuous changes of photosynthetic physiological state, namely dark adaptation, light adaptation and dark relaxation, using the relative biomass of downy fungus as the basis of P. viticola infection. The sensitivity of chlorophyll fluorescence parameters to downy mildew infection was evaluated by one-way analysis of variance (ANOVA). The optimal feature subset of chlorophyll fluorescence parameters extracted by feature selection strategies was input to machine learning classifiers to establish the early detection model of P. viticola infection. The results showed that with the increase of DPI, the degree of downy mildew infection was deepened, and the chlorophyll fluorescence dynamics curves and parameters of healthy and inoculated leaves were significantly different from 2DPI (p<0.01). Due to the infection, the photochemical quenching rate of inoculated leaves decreased (Rfd decreased), and the photosynthetic efficiency decreased (F_v/F_m decreased). Leaf vitality and photoprotection ability continued to decline (NPQ and qN decreased), and the light energy absorbed by leaves was more released in the form of fluorescence (F_t and F_m increased). BP neural network model using the feature subset (qN-L3, RFD-L2, NPQ-L1 and F_v/F_m-D1) optimized by the SFFS algorithm had the best detection accuracy, and the detection accuracy of healthy, and inoculated leaves at 3DPI was 83.75%. The average accuracy of the whole experiment period for 6 consecutive days reached 85.94%. These results provide a fast and accurate method for photosynthetic phenotype analysis and early detection of grape downy mildew.