3.2.1. Selection of Sensitive Wavebands

Due to the noisy spectral reflectance data around 400 and 1000 nm, the wavelengths used in this study were in the range of 437.2–976.2 nm. By using an SPA algorithm, we obtained the conditional synergetic score (COSS) values of each wavelength, indicating its sensitivity to PM, as shown in Figure 4A. The higher the COSS value, the more the waveband is sensitive to disease. Then we ranked the spectral bands in descending order according to their COSS values and selected the top 20 bands further for construction of PLS-LDA models. We successively added these bands to the PLS-LDA model one by one according to their COSS values in descending order and recorded their calibration and validation accuracies for selection of the best combination of sensitive wavebands. These calibration and leave-one-out cross validation (LOOCV) accuracies are shown in Figure 4B. Cross validation accuracies gradually increased and achieved highest coinciding point at the sixth waveband, hence, top six wavebands with higher COSS values were selected as sensitive wavelengths, which were 533.5, 659.2, 528.5, 661.7, 968.2, and 523.6 nm by descending COSS values.

**Figure 4.** (**A**): COSS value computed by SPA for each waveband. (**B**): Calibration and cross validation classification accuracies at different number of variables based on PLS-LDA.

### 3.2.2. Selection of Optimal Vegetation Indices

The sensitive VIs were selected by correlation analysis between the VIs and DS of wheat PM in two growth stages. Figure 5 summarizes the sensitivity of 15 VIs to DS at each of the two individual growth stages and at both growth stages as a whole. Except for TVI, these selected VIs were significantly correlated to DS at both stages. The response of VIs to DS at different growth stages differed from that of both stages. Most of the VIs showed high correlation to DS at booting stage, while only a few showed good relationships with DS at the jointing stage. The response of the VIs in different growth stages is not the same. In total, six spectral vegetation indices (PMI, MSR, MCARI, ARI, SIPI, and PSRI) were significantly correlated with DS at each growth stage as well as for both stages as a whole. They were selected for further disease recognition regression and classification models.

**Figure 5.** Response of spectral vegetation indices to DS at both and single growth stages.
