*4.4. Chlorophyll Content Detection Model*

From the perspective of the method of selecting feature variables, the *Rv* <sup>2</sup> values of the detection models established by using as feature variables the CA bands and CA-WFs were 0.6319 and 0.6940, respectively. The *Rv* <sup>2</sup> values of the detection models established by using as feature variables the CA peak bands and CA peak WFs were 0.7082 and 0.7364, respectively. The distributions of CA bands and CA-WFs were relatively concentrated. A high degree of autocorrelation existed between variables. Meanwhile, CA peak bands and CA peak WFs were evenly distributed and comprehensively reflected information.

Comparing the CA peak bands and the CA peak WFs, under the same variable selection method, the *Rv* <sup>2</sup> of the detection model established using the CA peak WFs (0.7364) was larger than that using CA peak bands (0.7082). The CWT, a process of dimensionality-increasing operation, could dig out the spectral variable information of chlorophyll. The CA peak WFs provided more variable information related to chlorophyll content. Finally, the PLSR model established using CA peak WFs was preferred to detect the chlorophyll content of corn crops.

Comparing the detection models established using spectral reflectance (*Rc* <sup>2</sup> = 0.77) [14] and spectral index (*Rc* <sup>2</sup> = 0.70) [15], the wavelet features (*Rc* <sup>2</sup> = 0.7856) showed better detection capability, which further illustrated that the CWT can deeply mine the information in spectral data.

The model comparison demonstrated that the data after continuous wavelet decomposition can be used to effectively extract valuable information in the spectral reflectance through the dimensionality-increasing operation. In terms of the physical and chemical parameter inversion, the middle- and low-frequency wavelet features highlighted the characteristics of crop pigment and water absorption. In combination with the local extremum of correlation coefficient method, the interference of multicollinearity was eliminated, and the degree of information redundancy was reduced. The detection model established by combining these two methods showed advantages in accuracy and error elimination.

#### *4.5. Chlorophyll Distribution in the Field*

The distribution of chlorophyll in Figure 9 demonstrates that the chlorophyll concentration in the canopy of plants gradually increased with the advancement of growth period. The field observation during the field experiment also conformed to this conclusion. The green leaves slowly become dark with the gradual growth of the corn, as shown in Figure 1. The chlorophyll content gradient in the test field remained unchanged over the three growth periods. The figure shows that the chlorophyll contents in the four regions of No. 28–31, 39–48, 52–55, and 63–72 were lower than those in other regions. Some studies have shown a significant positive correlation between the chlorophyll and the nitrogen contents of plants, and chlorophyll can reflect the nitrogen demand of plants to a certain extent [65–68]. Therefore, the chlorophyll detection model based on spectral reflectance can play a guiding role in smart field management and differentiated fertilization.
