Solar-induced chlorophyll fluorescence (SIF) is tightly coupled with plant photosynthetic functioning and provides a sensitive indicator of vegetation stress. To address the issue of data redundancy that arises during the process of building models directly using raw full-band SIF data, this study first performed correlation analysis on the full-band SIF to identify wavelength regions that are most responsive to wheat stripe rust severity level (SL). Based on the selected sensitive bands, six SIF spectrum indices were subsequently developed through mathematical transformations, including the reciprocal SIF spectrum index (RSI_SIF), logarithmic SIF spectrum index (LSI_SIF), reciprocal logarithmic SIF spectrum index (RLSI_SIF), first order differential SIF spectrum index (FDSI_SIF), sum SIF spectrum index (SSI_SIF), and differential SIF spectrum index (DSI_SIF). Subsequently, the correlations between each index and SL were evaluated, and indices showing stronger associations with SL were selected to develop wheat stripe rust remote sensing monitoring models using random forest regression (RFR) and support vector regression (SVR), which were further validated with independent samples. The results indicated that, all SIF spectrum indices showed stronger correlations with SL than the raw full-band SIF data and the single-band FRSIF datas. Among them, LSI_SIF exhibited the highest sensitivity to SL, achieving correlation improvements of 91% relative to far-red SIF (FRSIF) and 72% relative to the original full-spectrum SIF. In terms of predictive performance, models driven by the SIF spectrum indices generally outperformed those based on FRSIF or untransformed full-spectrum SIF, while RFR delivered superior overall accuracy compared with SVR across experiments. In the controlled plot experiment, the RFR model using full-spectrum SIF as predictors increased R² by 42% and reduced RMSE by 22%, compared with the FRSIF-based model. Moreover, RFR models incorporating RSI_SIF, LSISIF, RLSI_SIF, and SSI_SIF further improved R² by 19%, 21%, 22%, and 21% over the full-spectrum SIF model, accompanied by RMSE reductions of 19%, 22%, 23%, and 22%, respectively. In the field experiment, the corresponding RFR models achieved additional gains in R² of 28%, 27%, 30%, and 23% relative to the full-spectrum SIF model, while decreasing RMSE by 21%, 20%, 22%, and 19%, respectively. Overall, these findings suggest that SIF spectrum indices constructed through mathematical transformations of full-spectrum SIF can effectively enhance the disease-related signal, leading to more accurate and robust estimates of wheat stripe rust severity. The consistent improvements observed across the controlled plot experiment and the field experiment further highlight the stability and potential transferability of the proposed indices, supporting their applicability for operational remote sensing-based crop disease monitoring.