**1. Introduction**

Corn, one of the major food crops in the world, provides an important guarantee of food security and economic development. Proper nitrogen application is one of the keys to a good harvest of corn [1,2]. A number of studies have shown that the leaf chlorophyll content (LCC) can be used to predict the nitrogen requirement of crops [3,4]. Chlorophyll content is an important indicator of crop photosynthesis ability and nutrition level. Variable fertilization can be achieved using nitrogen fertilizers according to different chlorophyll contents through accurate monitoring of the chlorophyll content of corn leaves [5,6]. Appropriate fertilization can ensure that crops receive adequate nitrogen and avoid soil and water pollution caused by excessive fertilization [7]. This mechanism is the key to

improving the photosynthetic performance of crops, thereby regulating the growth and development of corn and increasing the input–output ratio of corn fertilizers [8]. The topic of detecting chlorophyll content of corn is one of the active areas in field management research today. Thus, this study aimed to detect the chlorophyll content in field crops to evaluate the growth status and providing guidance for fertilization.

The traditional chlorophyll detection method is an analytical chemical method, which has high precision. However, the process is complex, time-consuming, and may damage crops. This method cannot meet the requirements of rapid and nondestructive testing on site. Spectral analytical technology has been widely used in qualitative and quantitative analysis of the physicochemical parameters of farmland crops because of its fast, nondestructive, and nonpolluting characteristics. Kapp-Junior et al. [9] developed a novel regression model able to produce a prescription of the required nitrogen (N) for maize by combining spectral reflectance data and agronomic efficiency. Lu et al. [10] used hyperspectral techniques to analyze the vertical distribution of nitrogen in corn. Zhang et al. [11] used leaf characteristic spectra to forecast apple sugar content. These studies highlight the feasibility and efficiency of evaluating crop nutrients via spectroscopy. Therefore, rapid detection of the chlorophyll content was conducted in the present research by using hyperspectral technology during the growth stages.

Most current studies on the detection of chlorophyll content via spectral analysis have focused on the exploration of spectral characteristics to quantitate the intensity and position of molecular absorption or reflection [12,13]. The two types of methods to quantify the spectral absorption and reflectance of specific matters include multivariate statistical analysis and region positioning calculation. First, multivariate statistical methods are used to select and enhance the parameters of spectral reflectance, derivative spectrum, and vegetation index using maximum correlation coefficient analysis. Liang et al. [14] compared fifty hyperspectral vegetation indices, such as the photochemical reflectance index and canopy chlorophyll index, to identify the most appropriate vegetation indices for crop LCC and canopy chlorophyll content (CCC) inversion. Xu et al. [15] used simulated datasets from the PROSAIL model to establish a 2D-matrix-based relationship between leaf chlorophyll and red-edge relative indices (RERI(705nm) and RERI(783nm)). The leaf chlorophyll content can be retrieved using the two vegetation indices from observations on the basis of the matrix. Neto et al. [16] created a sunflower leaf chlorophyll model with the spectral reflectance in the band of 500–1039 nm by using partial least-squares regression (PLSR). Rei et al. [17] used two methods, namely machine-learning algorithms and the inversion of a radiative transfer model, to detect the LCC of tea. Second, the characteristic spectral positions show changes with the local correlation extreme values, which generally include the red edge and green peak. Li et al. [18] used spectral reflectance to construct red-edge spectral parameters and newly developed red-edge region parameters to detect the chlorophyll content in rapeseed leaves. Sun et al. [19] indicated that the blue edge, red valley, and eight other spectral parameters could be used to reflect the chlorophyll content of potato crops. Zheng et al. [20] developed a model of chlorophyll content in potato leaves on the basis of the red-edge location. The mentioned characteristic parameters selected by maximum or local extreme correlation have been widely used as sensitive spectral variables for detecting the chlorophyll content.

However, the canopy reflectance spectrum and sensitive bands of crops are easily affected by external interferences of dynamically changing soil background, vegetation canopy geometry, and atmospheric conditions during the growth periods [21–23]. Numerous studies have attempted to improve the detection models of chlorophyll content by eliminating the irrelevant and noise information of the spectral data [24,25]. With regard to the ideas and principles of radiation transmission, the combination of the PROSPECT leaf optical property model and SAIL (Scattering by arbitrarily inclined leaves) model, also referred to as PROSAIL, has been used to develop methods for retrieval of vegetation biophysical properties. Mridha et al. [26] used the broadband canopy radiation transfer model PROSAIL to invert the leaf area index (LAI), LCC, CCC, and leaf equivalent water thickness of the biophysical variables in soybeans. Botha et al. [27] evaluated the ability of the PROSAIL

canopy-level reflectance model to detect LCC of spring wheat (*Triticum aestivum* L.) during the growth stages between pretillering (Zadoks growth stage (ZGS15)) and booting (ZGS50). Lunagaria et al. discussed the spectral sensitivity of crop canopy parameters using a theoretical model. The results indicated that the reflectance in the visible range was important for chlorophyll retrieval. Reflectance in the near-infrared range has importance for retrieval of leaf inclination angle, dry matter, and LAI. Accordingly, the influencing factors are difficult to reduce, and the modeling results are challenging to improve because of the external and internal interferences affecting the field canopy spectrum [28]. The primary concern of this research area is to overcome such challenges to detect the chlorophyll content by hyperspectral technology during the growth stages.

Although scholars have tried to use preprocessing methods (such as continuum removal, first-order differentiation, and high-pass filtering) to eliminate the noise and enhance the characteristic signals caused by certain factors (such as sample background and stray light), challenges and problems in effectively removing the interference signals, especially random and low-frequency signals, exist during dynamic growth periods [29–31]. We addressed the primary concern using continuous wavelet transform (CWT) to overcome such problems. CWT, with its rich wavelet base function, multiresolution analysis, time-frequency localization, and other advantages, has received increasing attention in image and signal analysis, decomposition, compression, and denoising [32–37]. This method can effectively separate low-frequency signals from high-frequency signals and extract the weak information from the spectral signal. Chen et al. [38] studied the CWT, taking 265 leaves of 47 plants as the sample spectrum and effectively inverting the water content in the sample, with a high precision of up to 75%. Li et al. proposed a new technique (WREP) to extract red-edge positions (REPs) on the basis of the application of CWT to the reflectance spectra. The results demonstrated that WREP obtained the best detection accuracy for LCC and CCC compared with the traditional techniques. High scales of wavelet decomposition were favorable for the detection of CCC and low scales for the detection of LCC [39]. These studies highlight that the CWT can be used to improve the modeling results of LCC detections. However, great uncertainty still exists regarding the effects of this method to help improve LCC detection during growth periods in which the spectral characteristics are dynamically changed and influenced by soil background, vegetation canopy geometry, and atmospheric conditions. Similar to spectral wavelengths, whether to use the local correlation extremum method or the maximum correlation coefficient method to select sensitive wavelet features is worth discussing. Thus, this study aimed to clarify and create a model to monitor the CCC of corn on the basis of CWT during the growth stages.

This study focused on the relationship between LCC and corn canopy spectral reflectance to propose an efficient method to evaluate the chlorophyll content of corn. The main aims of this study were as follows: (1) to comparative analyze the advantages of maximum correlation value and local correlation extreme value in selecting feature variables; (2) to use CWT to decompose the original spectral data and extract the weak information in the spectrum to detect the chlorophyll content.
