Chlorophyll is a critical evaluation content of sugarcane growth monitoring, especially when diseases infected sugarcane. Accurate estimation of chlorophyll content is beneficial for the early detection and control of diseases, which is of great importance in practical production. In order to determine the best prediction model of chlorophyll content in sugarcane leaves, this study infected sugarcane leaves with mosaic from July to November, 2021 through artificial inoculation of pathogenic bacteria. Among them were 35 infected plants and 35 healthy plants, and two leaves were collected for each pot. Repeat the measurement of the leaf hyperspectral data using the spectrometer. The chlorophyll content of chemical leaves was measured to establish a hyperspectral data set of sugarcane leaves. In this study, five pre-processing methods, SG, MSC, SNV, 1st D and 2nd D, were used to establish the PLSR detection model and determine the best preprocessing method. Based on the optimized pretreatment results, correlation coefficient, SPA and RF were used to select the characteristic bands of chlorophyll content in sugarcane leaves, and the selected bands were combined with BPNN, SVR and KNN to establish chlorophyll prediction models. The results showed that the PLSR model based on SG treatment has the highest accuracy R2p=0.995 2, and the RMSEp=0.235 3 mg·cm-2. The model combined with the BPNN algorithm using RF screening was the optimal prediction model of chlorophyll content in sugarcane leaves under mosaic disease stress, the decision coefficient of the SG-RF-BPNN model was R2p=0.996 4, and the RMSEp=0.205 8 mg·cm-2, which had high accuracy and predictive power. It will provide a theoretical basis for the accurate and injury-free disease stress detection of large-scale sugarcane.