*3.6. Chlorophyll Distribution*

The spectral reflectance of the corn canopy in three growing stages was introduced into the chlorophyll content detection model. The distribution of chlorophyll in the field during the three growing periods was obtained (Figure 9).

**Figure 9.** Field chlorophyll distribution in three growth stages.

#### **4. Discussion**

#### *4.1. Sensitive Spectral Wavelengths*

This study demonstrated that spectral measurements can be used for corn CCC detection. In the range of visible light, the chlorophyll content in the blue and red light regions was positively correlated with spectral reflectance, and reflection peaks were present; in the green light region, the chlorophyll content was negatively correlated with spectral reflectance, and an absorption valley was present [16,51,52]. When the maximum correlation coefficient method was used to filter the sensitive wavelengths, the 15 selected wavelengths were all in the absorption valley of green light, and a serious multicollinearity existed between them. The sensitive wavelengths screened by the local extremum of correlation coefficient method were relatively dispersed and distributed in the visible and near-infrared regions. In the visible light region, the chlorophyll was is characterized by the reflection peak in the red and blue light regions and the absorption valley in the green light region. The extremum characteristic wavelengths in the near-infrared region reflected the composition of other substances [53–55]. The reflectance of the 896 nm band indicates CH3 methyl groups, that in the 976 nm band reflects the moisture content, and that in the 1051 nm band reflects CH2 methylene groups [56–58]. These substances are relative to the canopy structure of the corn crop. Therefore, the characteristic wavelengths in the NIR region will improve the robustness of the chlorophyll detection model. The CA peak bands were more evenly distributed and had less redundant information compared with CA bands.

#### *4.2. Continuous Wavelet Analysis*

Each wavelet feature contains the information of scale and wavelength position, which corresponds to the state of the generating wavelet function in the process of CWT, namely the scaling factor and the position of shift. The physical meaning of the wavelet feature can be explained by plotting the generating wavelet function corresponding to the wavelet feature that is sensitive to the biochemical parameters. In this study, the Gaussian second derivative was chosen as the generating function of CWT. Each wavelet feature reflects the similarity between the generating wavelet function and the reflectivity spectrum at a specific wavelength position and scale. The absorption characteristics of biochemical parameters at different positions and intensities in the reflectivity spectrum were detected. This situation can be seen as the result of smoothing the spectrum at a particular wavelength and finding the second derivative. The spectral bending degree caused by the different absorption intensities of the biochemical parameters in various bands can be characterized. These two understandings of CWT can be combined to explain the physical meaning of wavelet transform.

The absorption of chlorophyll in the green light band was weaker than that in red and blue light regions. In the reflectivity spectrum, a reflection peak was formed in the green light band. The wavelet features sensitive to chlorophyll were located near the reflection peak of the green light band. Different chlorophyll contents can affect the shape and size of the reflection peak. These changes are easily captured by wavelet features in low scales. A small wavelet feature of medium and high scale near the green light band covered the whole visible band, thus providing the amplitude information of reflectance. The wavelet features at red edge and near-infrared band were stable at the leaf level, thereby indicating that this region is important for chlorophyll monitoring.

Wang et al. used a Mexican hat as the generating wavelet function to obtain the correlation between the wavelet coefficient and the SPAD value of wheat leaves, which can reflect the amount of chlorophyll content. The results showed that the wavelet features sensitive to chlorophyll were located in the red-edge band of 720–740 nm, which was consistent with the research results in References [59–62]. The multiple scattering of light inside the leaves leads to high reflectivity in the near-infrared region, resulting in the rapid increase of spectral reflectance of green vegetation at 680–750 nm, the "red edge" of the area. In the vegetation reflectance spectrum curve, the red edge is one of the most obvious spectral features, and it is an important indicator band used to describe the chlorophyll state of vegetation. The red-edge position (REP) is the wavelength position at which the reflectance of vegetation increases fastest in this interval, and it is also the inflection point of the first derivative of spectral reflectance in this interval. The position of the red edge is an important spectral parameter for detecting the chlorophyll content of vegetation. Liao et al. extracted the wavelet characteristics sensitive to the chlorophyll content of maize leaves in different layers on the basis of the canopy spectrum. The wavelet features sensitive to the chlorophyll content in the upper leaves were distributed in green light and red-edge bands. Meanwhile, the wavelet features sensitive to chlorophyll content in the middle and lower leaves were all located in the red-edge bands, and those in the green light bands disappeared [63]. This phenomenon could have been due to the strong absorption of chlorophyll in the visible range that made it difficult for the green light to penetrate to the middle and lower layers; the visible band of the canopy spectrum mainly contained information regarding chlorophyll in the upper leaves [64]. The comparison of existing literature indicated that the wavelet features at the green light band at the blade level perform better among different datasets. The wavelet features at the red edge perform better among the different datasets at the canopy level.

#### *4.3. Sensitive Wavelet Features*

The frequency-scale analysis indicated that CA-WFs were mainly distributed on the low frequency scale (1–4), and the other two types were distributed on the mid-frequency scale (5). The wavelength position analysis indicated that CA-WF1 and CA-WF10 were distributed in the blue light region, and CA-WF 3, CA-WF 5, CA-WF 6, CA-WF 8, CA-WF9, and CA-WF11 were distributed in the red light area. These regions are the strong absorption bands of leaf chlorophyll. CA-WF2, CA-WF4, and CA-WF7 were distributed in the green light regions. These regions are strong reflection bands of leaf chlorophyll. All wavelet features were distributed in the visible light region, which represented the pigment information of the leaves. CA-WFs selected using the maximum correlation coefficient method also had redundant information. For example, the 12 characteristic wavelet coefficients of CA-WF10 were distributed in 472–483 nm, and had serious variable autocorrelation. These autocorrelation problems also existed in CA-WF7, CA-WF8, and CA-WF9.

The frequency-scale analysis indicated that CA peaks WF1–WF4 were distributed in the low-frequency scale (1–4). Meanwhile, CA peaks WF5–WF7 were distributed in the medium-frequency scale (5–7), and CA peak WF8 was located in the high frequency scale (8). Most wavelet features were in the middle- and low-frequency scales, which was consistent with the result of Yao et al [47]. The wavelet features in middle and low frequencies could effectively detect the water content of wheat leaves.

The analysis from the distribution of characteristic wavelet features indicated that all the variables in CA-WFs were distributed in the visible light area and could only reflect the pigment information in the leaves. CA peak WFs were more evenly distributed compared with CA-WFs. The sensitive variable WF4—839 nm in the near-infrared region—reflected the molecular structure of RNHR. WF5—915 nm reflected the CH2 methylene group, and WF5—973 nm showed the moisture information. WF6—985 nm reflected the starch material, WF7—894 nm showed the CH3 methyl group, and WF8—909 nm provided protein information. CA peak WFs comprehensively reflected the material structural information of corn leaves, thereby improving the stability of the chlorophyll detection model.
