*3.4. Correlation Analysis of Chlorophyll Content and Wavelet Energy Coe*ffi*cient*

CWT was performed with 10 frequency scales on the spectral reflection curve. We calculated the correlation coefficients of the wavelet energy coefficient and chlorophyll content at each frequency scale. Thereafter, we took the absolute value of the correlation coefficient results and drew the distribution map of the correlation coefficients at different scales. The result is shown in Figure 6.

The wavelet energy coefficient bands with high correlation with chlorophyll content (|*r*| > 0.5) are shown in Table 2. Most of high-correlation wavelengths were concentrated in the visible light area of 325–700 nm. Selected numbers of wavelet energy bands were reduced with the scale increase. In the low-frequency scales from 1 to 4, the high-correlation bands had narrow wavelength ranges, more numbers, and clear division of intervals. In the mid- and high-frequency scales from 5 to 8, selected bands had wider wavelength ranges, less numbers, and ambiguous division of intervals. At the low-frequency scale, the high correlation bands of scale 1 and 2 were consistent, and the high correlation bands of scale 3 and 4 were consistent. With the increase of frequency scale, the correlation between the wavelet energy coefficient and chlorophyll decreased gradually from scale 5, and the boundary between the high-correlation band and the low-correlation band became ambiguous. It can also be seen from Figure 6 that as the frequency scale increased, the correlation between the wavelet energy coefficient and the chlorophyll content gradually decreased. No high-correlation wavelet energy coefficient exists on the high-frequency scale of 9–10. Therefore, the optimal features of wavelet energy should be selected on a scale of 1 to 5.

**Figure 6.** Distribution of the absolute values of the correlation coefficients between the wavelet energy coefficient and the chlorophyll content.


**Table 2.** Wavelet energy coefficient bands with high correlation with chlorophyll content (|*r*| > 0.5).

3.4.1. Sensitive Wavelet Feature Selection Based on the Maximum Correlation Coefficient

Fifty wavelet energy coefficients with high correlations were selected as the chlorophyll-sensitive wavelet features (denoted as CA-WFs). These result are shown in Table 3. The absolute values of the correlation coefficients of CA-WFs and chlorophyll content were all higher than 0.8.

**Table 3.** Location and frequency scale parameters of the chlorophyll-sensitive wavelet features (CA-WFs) with |*r*| higher than 0.8.


3.4.2. Sensitive Wavelet Feature Selection Based on the Local Extrema of the Correlation Coefficient

Fifty-five chlorophyll-sensitive wavelet energy coefficients were selected as sensitive wavelet features (denoted as CA peak WFs) on the basis of the extreme values of the correlation coefficient. The absolute values of the correlation coefficients of CA peak WFs and chlorophyll content were all higher than 0.5. The distribution of CA peak WFs at different frequency scales is shown in Figure 7.

**Figure 7.** Result of the sensitive wavelet feature selection based on the local correlation coefficients.

The wavelength position analysis indicated that all sensitive wavelet variables of the CA Peak WF1, CA Peak WF2, and CA Peak WF3 were distributed in the visible light region, reflecting the leaf pigment information. CA Peak WF4 contained eight sensitive variables, seven of which were located in the visible light region, and another variable was located at 839 nm. CA Peak WF5 contained eight sensitive variables, six of which were located in the visible light region, and the other two variables were located at 915 and 973 nm. The 915 nm wavelength can reflect CH2 methylene groups, and 973 nm can present the leaf moisture information. CA Peak WF6 contained eight sensitive variables, five of which were located in the visible light region, and the other three variable positions were 821, 896, and 985 nm. CA Peak WF7 and CA Peak WF8 contained four and eight variables, respectively; each of them had one sensitive variable located in the near-infrared region at 894 and 909 nm. The remaining sensitive variables were located in the visible region.

#### *3.5. Establishment of Chlorophyll Content Detection Model with PLSR*

The PLSR algorithm was used to establish a chlorophyll content detection model on the basis of the spectral characteristic variables of the CA bands, CA peak bands, CA-WFs, and CA peak WFs. Both models used LOOCV for internal cross-validation to eliminate the influence of spectral information redundancy and multicollinearity on the model accuracy. The modeling results are shown in Table 4 and the verification results are shown in Figure 8. The comparison of the four detection models showed that the PLSR chlorophyll content detection model based on CA peak WFs had the optimal performance. The decision coefficient (*Rc* 2) of the modeling set was 0.7856, the *RMSEC* of the modeling set was 3.0408, the decision coefficient (*Rv* 2) of the verification set was 0.7364, and the *RMSEV* of the verification set was 3.3032.

**Figure 8.** Results of the PLSR detection model for the chlorophyll content. (**a**) Results of the maximum correlation coefficient method; (**b**) Results of the local extremum of correlation coefficient method; (**c**) Results of the maximum correlation coefficient method with wavelet energy coefficient; (**d**) Results of the local extremum of correlation coefficient method with wavelet energy coefficient.


**Table 4.** Result statistics of PLSR detection model for the chlorophyll content.
