(2) Local Extremum of Correlation Coefficient Method

The local extremum of the correlation coefficient method was proposed to improve the variable selection strategy of correlation analysis to solve multicollinearity between the adjacent wavelength variables: the correlation coefficient of spectral reflectance and chlorophyll content were calculated, and the correlation coefficient curve was drawn. Thereafter, the local extreme points of the correlation curve (the zero-crossing positions of the correlation first-derivative curve) were calculated as the chlorophyll characteristic wavelengths [46]. The maximum correlation wavelengths were also selected on the basis of the maximum correlation coefficient method for comparative analysis.

#### 2.5.2. Continuous Wavelet Analysis

From the perspective of signal processing, wavelet analysis can be used to perform data analysis in the frequency and time domains and extract available information from the signal [47]. The reflection spectrum analysis is highly similar to electronic signal analysis. Accordingly, CWT can be used to decompose the reflection spectrum curve at different frequency scales to generate a series of wavelet energy coefficients. The CWT process is shown in Equation (8).

$$\mathcal{W}\_f(a,b) = \frac{1}{\sqrt{a}} f(\lambda) \psi(\frac{\lambda - b}{a}) d\lambda,\tag{8}$$

where *a* is the frequency scale factor which is set to 2*<sup>n</sup>* (*n* = 1, 2, ... , 10) gradients, and translation factor *b* is the center wavelength of the mother wavelet function. The mother wavelet function ψ(λ) uses the second-order Gaussian function. *f*(*x*) is a 1D reflection spectrum, and the wavelet coefficient *Wf*(*a*,*b*) (denoted as *WFa,b*) is 2D data, including frequency scale (1, 2, ... , 10) and wavelength (325–1075 nm).

Correlation analysis of the wavelet energy coefficient and chlorophyll content was performed. The local extreme value of the correlation was calculated as the sensitive wavelet feature. The wavelet coefficient with the highest correlation was selected for comparative analysis.

The extreme point was calculated using the peak function in MATLAB software. The parameters of the peaks' function were set as follows: the correlation extreme value peak distance of the spectral reflectance was set to 10, and the minimum peak value was set to 0.1; the correlation extreme value peak distance of the wavelet coefficient was set to 50, and the minimum peak value was set to 0.3.
