Laser-preprocessed waxy corn chlorophyll inversion method based on unmanned aerial vehicle multispectral remote sensing

Abstract

The invention discloses a waxy corn chlorophyll inversion method after laser pretreatment based on unmanned aerial vehicle multispectral remote sensing. The method comprises the following steps: waxy corn seeds are divided into six groups and are subjected to pretreatment irradiation by adopting different laser parameters respectively; in the growth process of the waxy corn, acquiring a multispectral image of each processed land parcel by using an unmanned aerial vehicle carrying a six-waveband multispectral sensor; radiation correction and orthographic processing are carried out on the multispectral image, and texture features corresponding to six wavebands are extracted; synchronously adopting a chlorophyll meter to carry out SPAD measurement on the quadrat of each processed land parcel, and establishing a regression inversion model between the remote sensing texture features and a ground SPAD; in the prediction stage, the texture features obtained through calculation of the multispectral image of the land parcel to be detected are substituted into the model, and the inversion result of the waxy corn SPAD is obtained. Compared with the prior art, the method has the advantages that the inversion model is constructed by using the multispectral texture features, rapid and non-contact estimation of the chlorophyll value of the waxy corn can be realized under different laser pretreatment conditions, and the monitoring efficiency and consistency are improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing monitoring and crop physiological parameter inversion, and in particular to a method for inverting and estimating the chlorophyll content (characterized by SPAD value) of waxy corn after laser preprocessing by using six-band multispectral remote sensing images from UAVs and combining them with gray-level co-occurrence matrix texture features. Background Technology

[0002] Chlorophyll content is an important physiological indicator reflecting crop nitrogen nutrition status, photosynthetic capacity, and growth potential. The SPAD value, measured by a chlorophyll meter, is typically used as the characterization parameter. Existing chlorophyll monitoring methods mainly include manual sampling and instrument-based point measurement. However, these methods suffer from limitations such as limited measurement points, time and labor consumption, and insufficient spatial representativeness, making it difficult to meet the needs of large-scale, rapid, and continuous field monitoring.

[0003] With the development of UAV remote sensing technology, using UAVs equipped with multispectral sensors to acquire crop canopy reflectance information and inverting chlorophyll content through vegetation indices or regression models has become an important method. However, existing inversion methods based on single spectra or a few vegetation indices are easily affected by factors such as changes in light intensity, imaging conditions, soil background, and canopy structure. Especially in experimental scenarios where seeds are pretreated with different laser parameters, different treatments may cause differences in seedling emergence, growth vigor, and canopy spatial structure, making the stability of models that rely solely on spectral intensity or a single vegetation index insufficient, resulting in large fluctuations in inversion accuracy across plots and treatment conditions.

[0004] Therefore, a technical solution is needed that can take into account both multispectral information and canopy spatial structure characteristics, and is applicable to the rapid inversion of chlorophyll parameters in waxy maize under different laser pretreatment conditions, so as to improve the robustness and reliability of chlorophyll inversion. Summary of the Invention

[0005] The purpose of this invention is to provide a method for inverting chlorophyll in waxy corn after laser preprocessing based on UAV multispectral remote sensing. By introducing multispectral texture features into the modeling, a rapid and non-contact inversion estimation of the SPAD value of waxy corn can be achieved under different laser preprocessing conditions, thereby reducing the cost of manual measurement and improving the efficiency and stability of plot-scale monitoring.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Before sowing, waxy corn seeds were grouped and pretreated with different laser parameters (including a control group and combinations of red and blue light irradiation for different durations).

[0008] Aerial surveys were conducted using unmanned aerial vehicles (UAVs) during the target growth period of waxy corn. This target growth period is a pre-defined stage for chlorophyll inversion, selectable from one or more of the following stages: jointing stage, large trumpet stage, tasseling and flowering stage, or grain-filling stage. Preferably, the aerial survey was conducted during the tasseling and flowering stage to utilize the more complete canopy cover and the greater sensitivity to chlorophyll characterization, thereby improving the consistency and application value of the inversion results. During the aerial survey, a UAV equipped with a six-channel multispectral camera acquired multispectral images of the test plot. The center wavelengths of the six channels of the camera were 450 nm, 555 nm, 660 nm, 720 nm, 750 nm, and 840 nm, respectively. Prior to the aerial survey, radiometric calibration was performed using a white reflectance calibration board (white board) to reduce the impact of light variations on the consistency of image reflectance.

[0009] The acquired multispectral images are stitched together, radiometrically corrected, and orthorectified to obtain reflectance images for calculation.

[0010] Feature extraction is performed within the plot area using random sampling points. A window is constructed with the sampling points as the center to calculate the gray-level co-occurrence matrix texture features. Texture calculation uses 0°, 45°, 90°, and 135° directions and takes the average value. The window is, for example, 11×11 pixels. At the same time, the window pixels are stretched and quantized to calculate the GLCM index.

[0011] Furthermore, to improve the stability of the inversion model and avoid overfitting caused by the introduction of redundant texture indices, this invention screens and optimizes candidate texture features before constructing the regression model: First, a set of candidate features is constructed by calculating various GLCM texture indices on the six-band reflectance image, and the texture features of random sampling points within the same plot are statistically summarized to obtain plot-level candidate features; then, correlation analysis is performed based on the plot-level candidate features and the ground SPAD reference value, and combined with collinearity test and regression fitting evaluation, a subset of texture features with high correlation to SPAD and low redundancy is selected as the model input variables; after screening, this invention preferentially selects the third band contrast feature b3_contrast, the fifth band contrast feature b5_contrast, and the sixth band correlation feature b6_correlation for constructing the SPAD regression inversion model, thereby improving the model's generalization ability and applicability across processing conditions.

[0012] Simultaneously, a chlorophyll meter was used to obtain ground-level SPAD reference values. In this invention, 12 plants were selected from each plot, and one leaf was selected from each plant. Each leaf was measured three times and the average was taken to obtain the plot-level SPAD.

[0013] Finally, a regression inversion model between remote sensing texture features and ground SPAD was established, and in the prediction stage, the texture features calculated from the image of the land parcel to be measured were substituted into the model to output the SPAD inversion results.

[0014] Compared with existing technologies, the present invention has at least the following advantages: First, it improves the consistency of image data under different aerial survey conditions through whiteboard calibration and reflectivity processing; second, it introduces gray-level co-occurrence matrix texture features on the basis of multispectral information, enabling the model to characterize the spatial structure and uniformity differences of the canopy, thereby improving the inversion stability under different laser preprocessing conditions; third, through random sampling points within the plot and plot-level statistical summarization, it can obtain more representative plot-scale SPAD estimation results while maintaining computational efficiency. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the process for retrieving chlorophyll in waxy maize after laser preprocessing based on UAV multispectral remote sensing according to the present invention.

[0016] Figure 2 A schematic diagram illustrating the determination of the sampling area for a land parcel, the generation of random sampling points, and feature extraction;

[0017] Figure 3 This is a schematic diagram comparing the measured and inverted values ​​of SPAD on the ground. Detailed Implementation

[0018] The specific embodiments of the present invention will be described below with reference to the accompanying drawings. Figure 1 As shown, this invention provides a method for inverting chlorophyll (SPAD) in waxy maize after laser preprocessing based on UAV multispectral remote sensing; Figure 2 The diagram illustrates the determination of the land parcel sampling area, the generation of random sampling points, and feature extraction. Figure 3 The comparison results between the measured and inverted values ​​of ground-based SPAD are shown. It should be understood that the following embodiments are used to illustrate the technical solution of the present invention, and any equivalent substitutions made by those skilled in the art to the parameters, order of steps, and implementation methods without departing from the concept of the present invention are all within the protection scope of the present invention.

[0019] This embodiment uses waxy corn as the subject and conducts a field experiment under seed laser pretreatment conditions. Before sowing, waxy corn seeds were divided into 6 treatment groups, and each treatment group was pretreated with different laser parameters, including at least the type of light source (red light / blue light) and the duration of irradiation. A control group was also set up. After pretreatment, the seeds were sown in the corresponding experimental plots according to the treatment groups, and the field management measures were kept consistent to reduce interference from non-laser factors.

[0020] Unmanned aerial surveys were conducted during the target growth period of waxy corn. In this embodiment, image acquisition was preferably carried out during the tasseling and flowering stage to leverage the relatively complete canopy cover and the greater sensitivity of chlorophyll characterization at this stage, thereby improving the consistency and application value of the inversion. The UAV was equipped with a six-channel multispectral camera to acquire images of the test plot. The center wavelengths of the six channels were 450 nm, 555 nm, 660 nm, 720 nm, 750 nm, and 840 nm, respectively. To reduce the impact of light variations, radiometric calibration was performed using a white reflectance calibration board (whiteboard) before the aerial survey: images of the whiteboard were acquired under the same lighting conditions, and the digital quantization values ​​of the images were converted into reflectance benchmarks based on the nominal reflectance of the whiteboard.

[0021] The acquired multispectral images were preprocessed, including radiometric correction, geometric correction, and orthorectification stitching, to obtain a six-band reflectance orthorectified image in the same coordinate system. Radiometric correction used whiteboard calibration results to reflectance-enhanced each band of the image, thereby improving the consistency of data from different sorties or under different lighting conditions.

[0022] Subsequently, the sampling area for the land parcel was determined and its features were extracted (see...). Figure 2 First, the land parcel boundary vector is imported. Considering the potential for mixed pixels at the parcel boundaries, this embodiment buffers the parcel boundaries inward to form an internal sampling region. A preset number of random sampling points are generated within the sampling region, and features are extracted from the reflectance image centered on these random sampling points. An 11×11 pixel sampling window is constructed for each sampling point. After stretching and quantizing the pixels in the window, the Gray-Level Co-occurrence Matrix (GLCM) texture index is calculated. The GLCM is constructed in four directions: 0°, 45°, 90°, and 135°. The arithmetic mean of the directional results is used to obtain the texture feature value of the sampling point in the corresponding band. The texture features of all random sampling points within the same parcel are statistically summarized (e.g., the mean is taken) to obtain the parcel-level texture features, which characterize the spatial structure and uniformity differences of the parcel canopy.

[0023] Simultaneously, ground-based SPAD reference values ​​were acquired. During the aerial survey, 12 waxy corn plants were selected from each plot, and one representative leaf was selected from each plant. The SPAD value of each leaf was measured three times using a chlorophyll meter and averaged to obtain the value of a single plant. The SPAD values ​​of the 12 plants were then statistically summarized at the plot level to obtain the SPAD reference value for that plot, which was used for model training and validation.

[0024] Before constructing the regression model, to avoid overfitting due to redundancy of candidate texture indicators and to improve generalization ability, this embodiment screens and optimizes candidate texture features: based on plot-level candidate texture features and plot-level SPAD reference values, correlation analysis is conducted, and combined with collinearity tests and regression fitting evaluation, a subset of texture features with high correlation to SPAD and low redundancy is selected. After screening, this embodiment preferentially selects the third band contrast feature b3_contrast, the fifth band contrast feature b5_contrast, and the sixth band correlation feature b6_correlation as model input variables.

[0025] A SPAD inversion model was established using multiple linear regression, with plot-level texture features as the independent variable and plot-level SPAD reference values ​​as the dependent variable. The model form is as follows:

[0026]

[0027] in, To retrieve the SPAD values, b3_contrast, b5_contrast, and b6_correlation represent the contrast texture features of the third band, the fifth band, and the correlation texture features of the sixth band, respectively, while a, b, c, and d are regression coefficients. As a preferred implementation, the values ​​are obtained by fitting training samples.

[0028]

[0029] In the prediction phase, the texture features extracted from the plot to be tested are substituted into the model to output the SPAD inversion results.

[0030] The model's performance can be evaluated using fitting accuracy and cross-validation error, such as... Figure 3 The diagram shows that effective inversion of ground-based SPADs can be achieved.

Claims

1. A method for chlorophyll inversion in waxy maize after laser preprocessing based on UAV multispectral remote sensing, characterized in that, The process includes the following steps: S1: Divide waxy corn seeds into 6 treatment groups, pretreat the seeds of each treatment group with different laser parameters, and complete sowing and field management; S2: During the target growth period of waxy corn, use a UAV equipped with a six-band multispectral sensor to conduct aerial surveys of each treatment plot to acquire multispectral images; S3: Perform radiometric correction, geometric correction, and orthorectification on the multispectral images to obtain reflectance images for inversion calculations; S4: Based on the reflectance image, calculate the gray-level co-occurrence matrix texture features corresponding to the six bands in the corresponding area of ​​each treated plot, including at least: the third band contrast feature b3_contrast, the fifth band contrast feature b5_contrast, and the sixth band correlation feature b6_correlation; S5: Use a chlorophyll meter to measure the chlorophyll value of the quadrats of each treated plot to obtain the ground chlorophyll value SPAD; S6: Construct a regression inversion model between the remote sensing texture features and the ground SPAD, and in the prediction stage, substitute the texture features obtained in step S4 into the model to output the SPAD inversion result of waxy corn.

2. The method according to claim 1, characterized in that, The texture features mentioned in step S4 are calculated based on the gray-level co-occurrence matrix. The gray-level co-occurrence matrix is ​​constructed in four directions: 0°, 45°, 90°, and 135°. The texture features are obtained by taking the arithmetic mean of the texture feature values ​​in the four directions.

3. The method according to claim 1, characterized in that, The texture features described in step S4 are calculated using a sliding window, the size of which is... ,in It is an odd number between 9 and 31, preferably 11.

4. The method according to claim 1, characterized in that, The regression inversion model described in step S6 is a linear regression model, which satisfies: Where a, b, c, and d are model coefficients obtained by fitting the training samples.

5. The method according to claim 4, characterized in that, The model coefficients satisfy: a=323.058, b=-0.05937, c=0.00940, d=-225.189, thus obtaining:

6. The method according to claim 1, characterized in that, Before constructing the regression inversion model described in step S6, the method further includes a step of screening and selecting candidate texture features. The screening and selection involves: calculating a set of candidate texture features containing different bands and different texture indices based on the reflectance image; performing initial screening of candidate texture features based on the correlation between candidate texture features and ground SPAD; further performing multicollinearity test to eliminate redundant features; and then determining a subset of key texture features for modeling based on regression fitting evaluation index and / or cross-validation error. The subset of key texture features includes at least b3_contrast, b5_contrast, and b6_correlation.

7. The method according to claim 1, characterized in that, In step S4, the plot area is determined by the plot boundary vector, and the plot boundary vector is buffered inward to form a sampling area. A preset number of random sampling points are generated within the sampling area, and the spectral features and / or texture features of the corresponding pixels or windows are extracted from the reflectance image with the random sampling points as the center. The texture features include at least b3_contrast, b5_contrast, and b6_correlation used for inversion in claim 1. The features of multiple random sampling points within the same plot are statistically summarized to obtain plot-level features, so as to avoid the influence of mixed pixels at the plot boundary on feature calculation.

8. The method according to claim 1, characterized in that, In step S3, radiometric correction is performed using a white reflectance calibration board (white board) to convert the digital quantization values ​​of the multispectral image into reflectance.

9. A system for retrieving chlorophyll values ​​from waxy corn for implementing the method according to any one of claims 1 to 7, characterized in that, include: The system includes an unmanned aerial vehicle (UAV) platform, a six-band multispectral sensor, a chlorophyll meter, and a data processing terminal; the data processing terminal is used to perform image correction, texture feature calculation, regression model training, and SPAD inversion output.

10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images