Modeling method for outdoor multi-scale fruit partition classification marking quantitative detection

Abstract

The invention discloses a modeling method for outdoor multi-scale fruit partition classification marking quantitative detection, and mainly solves the problem of spatial heterogeneity of fruits of different scales during outdoor fruit detection. Because the sampling calibration and prediction inversion area angles are inconsistent, the precision of a quantitative detection model is greatly influenced. Abstracting and gridding the features of the orchard region to be detected, and performing partition classification; quantitatively describing the spatial heterogeneity of the fruit tree by using a semi-variation function and an anisotropy index, and simulating, calculating and marking a spatial spectrum invariant region corresponding to the canopy spatial structure of the fruit tree and spatial spectrum variation regions at different angles by using a spectral scale law; and carrying out a near-surface fruit tree partition classification marking experiment corresponding to the to-be-detected orchard region by using a preset experiment rule to obtain a fruit growth illumination feature model, and training the fruit growth illumination feature model for outdoor fruit quality detection.

Description

Technical Field

[0001] The present invention relates to the technical field of fruit detection, and particularly to a modeling method for outdoor multi-scale fruit zoning, classification, marking and quantitative detection. Background Art

[0002] There are certain differences in temperature difference and light conditions in different regions. Moreover, there are also significant differences in the quality of fruits at different growth parts of the same fruit tree. That is, the fruit quality is affected by environmental factors such as outdoor light and temperature, showing certain regional distribution characteristics, which is also a sign of the environmental adaptability of fruit quality. However, this also makes it impossible to quickly and massively determine the fruit quality.

[0003] Currently, when solving the problem of outdoor fruit detection, the calibration board and the fruit are usually not in the same orientation, making in-situ measurement impossible. Due to the inconsistent angles of the sampling calibration and the prediction inversion region, it has a great impact on the accuracy of the quantitative detection model. A drone multi-spectral camera is used to detect the fruit quality. Based on the conventional fruit detection spectral model, the spectral resolution is increased to improve the accuracy of fruit quality detection.

[0004] However, the above processing method has the disadvantages of large data volume and low efficiency due to improving the accuracy of fruit quality detection by increasing the spectral resolution. Summary of the Invention

[0005] In view of the above problems, the present invention provides a modeling method for outdoor multi-scale fruit zoning, classification, marking and quantitative detection. The main purpose is to realize the detection of fruits using light characteristics and temperature characteristics, and improve the accuracy and efficiency of detection. Solve the problem that when detecting outdoor fruits, the calibration board and the fruit are usually not in the same orientation, making in-situ measurement impossible. Due to the inconsistent angles of the sampling calibration and the prediction inversion region, it has a great impact on the accuracy of the quantitative detection model. Based on the theory of partially coherent spectral transmission, the present invention divides the space grid of environmental light and temperature factors, and converts the points with different spacings in the spectral image into angular relationships through progressive sampling and marking of light intensity, reducing environmental interference and improving the representativeness of samples. Secondly, on the basis of the original spectral detection, through the cross-validation of the multi-angle polarization reflectivity and spectral characteristics of the fruit, the angle classification and zoning calibration of the spectral transmission law is further carried out. Through the classification and marking of multi-angle polarization reflectivity characteristics, the accuracy of the multi-scale quantitative detection model is improved.

[0006] Perform progressive sampling and classification marking in the direction of increasing light intensity. Through the refined classification in different polarization directions, collect the spectra of fruit tissue sections in different directions by multi-angle polarized diffuse transmission, and obtain the quantitative relationship between the peak value of polarization degree and roughness in the light direction. Establish the corresponding relationship between the quality distribution of fruits in different spatial directions and the light intensity direction, and improve the accuracy of the multi-scale quantitative detection model.

[0007] To solve the above technical problems, the present invention proposes the following solutions:

[0008] The present invention provides a modeling method for outdoor multi-scale fruit zoning classification marking and quantitative detection, and the method includes:

[0009] Based on the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points, using a preset processing rule to process the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, and the grid temperature characteristics and the grid light characteristics are marked with zoning classification marks;

[0010] Using a preset rule to abstract and simplify the spatial structure of the fruit tree canopy in the orchard area to be measured, and using the spectral scaling law to simulate, calculate and mark the spatially spectral invariant area and the spatially spectral changing areas at different angles corresponding to the spatial structure of the fruit tree canopy;

[0011] Based on the spatially spectral invariant area and the spatially spectral changing areas at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured and the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, using a preset experimental rule to conduct a near-ground fruit tree zoning classification marking experiment corresponding to the orchard area to be measured to obtain a fruit growth light characteristic model.

[0012] According to a preset training strategy, training the fruit growth light characteristic model, extracting key parameters with obvious environmental light characteristics to improve the fruit growth light characteristic model, so as to determine the fruit quality of the orchard area to be measured through the improved fruit growth light characteristic model.

[0013] Further, the step of based on the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points, using a preset processing rule to process the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, and the grid temperature characteristics and the grid light characteristics are marked with zoning classification marks, includes:

[0014] Obtaining the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points;

[0015] Based on the near-ground remote sensing image data of the orchard area to be measured, using a preset grid processing rule to perform grid processing on the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured;

[0016] Based on the ground experimental control points corresponding to the orchard area to be measured, use the preset zoning and classification rules to perform zoning and classification marking on the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, and obtain the grid temperature characteristics and the grid light characteristics marked with zoning and classification marks.

[0017] Further, abstracting and simplifying the spatial structure of the fruit tree canopy in the orchard area to be measured by using a preset rule, and simulating, calculating and marking the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy by using the spectral scaling law, including:

[0018] Obtain multi-scale and multi-angle images of the fruit trees in the orchard area to be measured;

[0019] Perform grid feature extraction processing on the multi-scale and multi-angle images of the fruit trees in the orchard area to be measured to obtain the abstracted and simplified spatial structure of the fruit tree canopy;

[0020] Simulate, calculate and mark the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy by using the spectral scaling law.

[0021] Further, based on the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, and the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, carry out the near-ground fruit tree zoning and classification marking experiment corresponding to the orchard area to be measured to obtain the fruit growth light characteristic model, including:

[0022] Based on the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, carry out the light intensity progressive zoning and classification marking sampling for the near-ground fruit trees in the orchard area to be measured, and use the quantitative descriptions of different shapes of ROIs to describe the mean distribution characteristics, max distribution characteristics and structural characteristics of the fruits in different regions;

[0023] Based on the mean distribution characteristics, max distribution characteristics and structural characteristics of the fruits in different regions, obtain the semi-variogram of ROIs at different angles, use the anisotropy index ANIF and BRVF to describe the statistical characteristics of BRDF in different directions, and use the fruit roughness and polarization peak power function relationship model to describe the spatial structure characteristics of the fruit trees in the orchard area to be measured;

[0024] Based on the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured and the spatial structure characteristics of the fruit trees in the orchard area to be measured, establish the statistical correlation and significance analysis of the four factors of temperature, light, angle and polarization to obtain the key parameters;

[0025] The key parameters are used to establish calibration and comparative verification data to obtain the fruit growth illumination characteristic model.

[0026] Furthermore, the fruit growth illumination characteristic model is trained according to a preset training strategy to extract key parameters with obvious environmental illumination characteristics for improving the fruit growth illumination characteristic model, including:

[0027] Establishing a preset training strategy based on the preset fruit sugar-acid ratio and the corresponding spectrum and the calibration and comparative verification data;

[0028] Based on the preset training strategy, the calibration and comparison verification data are used to train the fruit growth illumination characteristic model to change the weights of the relevant parameters of the fruit growth illumination characteristic model;

[0029] Based on the preset training strategy, the preset fruit sugar-acid ratio and the corresponding spectrum are used as constraints to train the fruit growth illumination feature model to extract key parameters with obvious environmental illumination characteristics.

[0030] Furthermore, the method further comprises:

[0031] The modeling method described is to first perform modeling indoors and then perform modeling outdoors.

[0032] Furthermore, the method further comprises:

[0033] When modeling outdoors, it is necessary to establish an angle spectral response function model, a multi-angle and spectral quantitative model transfer.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention combines satellite remote sensing images and ground interactive verification experiments to select typical orchard environments in the area to be tested, and uses spatial orientation grid illumination and temperature fields, and uses drone multi-spectral multi-angle polarization to quantitatively detect the reflectivity variation law of the same fruit in different polarization directions, and fits the reflectivity curve model to quantitatively detect the quality accuracy of the fruit. A multi-angle profiling calibration field is established through gridded orchard regional features, and spectral quantitative detection is combined with drone spectral image regional evaluation. The corresponding relationship between illumination and quality distribution is established through progressive light intensity sampling, thereby improving the accuracy of the multi-scale quantitative detection model.

[0036] Secondly, a polarizer is added to the detection end to change the polarization direction for polarization calibration of multi-polarization directions and polarization multi-angle reflectivity. The four polarization directions of the polarizer are (0p, 45p, 90p, 135p). Correspondingly, for the horizontal direction (0 - 360 degrees), it rotates 8 directions (0, 45, 90, 135, 180, 225, 270, 315) every 45 degrees; for the vertical direction, the elevation angle (0 - 90 degrees) is changed to obtain spectral images at multiple angles, and the multi-angle polarization reflectivity function model is fitted. The present invention realizes the calibration of a low-resolution spectral camera with a high-resolution spectrometer. The present invention makes a quality space distribution map for thematic data fusion, conducts the distribution of moisture, acidity, and sugar content of the fruit quality in the area, combines geographical information such as light, temperature, and GPS coordinates, marks and establishes a big data system for the environmental distribution of fruit quality, and provides data support for improving the quality and efficiency of the fruit industry and future layout and structural adjustment.

[0037] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention are specifically given below. Brief Description of the Drawings

[0038] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0039] Figure 1 Shows a flow chart of a modeling method for outdoor multi-scale fruit zoning, classification, marking, and quantitative detection provided by an embodiment of the present invention;

[0040] Figure 2 Shows the specific modeling method provided by an embodiment of the present invention;

[0041] Figure 3 Shows the schematic of the far-field spectral difference when the sampling calibration and the prediction inversion area are not in the same time;

[0042] Figure 4 Shows the schematic of the inconsistent angles of the sampling calibration and the prediction inversion area provided by an embodiment of the present invention;

[0043] Figure 5 Shows the schematic of multi-scale and multi-angle fruit classification, marking, and quantitative detection provided by an embodiment of the present invention. Detailed Description of the Preferred Embodiments

[0044] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0045] Term Explanation:

[0046] BRDF: (Full English name: Bidirectional Reflectance Distribution Function, Chinese full name: Bidirectional Reflectance Distribution Function) is used to define how the irradiance in a given incident direction affects the radiance in a given outgoing direction.

[0047] BRDF device: mainly used for measuring the hyperspectral bidirectional reflectance distribution function (BRDF) of indoor samples, characterizing the angular reflection characteristics of samples, for radiometric calibration, multi-angle remote sensing, the mean statistical characteristics of BRDF-characterized reflectance, and BRVF supplements its variance characteristics. The two jointly quantify the anisotropy of surface reflection. BRVF (Bidirectional Reflectance Variance Function) is a statistic based on BRDF and is used to characterize the spatial variation of BRDF in the direction. It describes the variance or degree of change of BRDF values at different observation angles. BRVF has important applications in the field of remote sensing, especially when analyzing the anisotropy of surface reflection characteristics.

[0048] ANIF (Anisotropy Index Factor), the anisotropy index, quantifies anisotropy by comparing the semivariogram parameters (such as range, sill value) in different directions, and the calculation formula is ANIF = max(x) / min(x).

[0049] Next, a specific embodiment will be combined to further introduce in detail a modeling method for outdoor multi-scale fruit zoning, classification, marking, and quantitative detection of the present invention.

[0050] The technical solution of the present invention solves the problem that the calibration plate and the fruit are usually not in the same orientation during outdoor fruit detection, making in-situ measurement impossible. Due to the inconsistent angles of the sampling calibration and the prediction inversion area, (such as Figure 4has a great impact on the accuracy of the quantitative detection model. Through the spatial grid division of environmental light and temperature factors, the present invention converts the points with different spacings in the spectral image into angular relationships, performs multi-scale sampling and classification marking in the illumination direction through the illumination direction, and performs progressive sampling and classification marking in the illumination intensity enhancement direction. Through the refined classification in different polarization directions, the spectral images of fruit tissue sections in different directions are collected through multi-angle polarized diffuse transmission, and the quantitative relationship between the peak value of the polarization degree in the illumination direction and the roughness is obtained. The corresponding relationship between the quality distribution in different spatial directions of the fruit and the illumination intensity direction is established.

[0051] Further, the angle classification and zoning calibration of the spectral transmission law (as shown in the following formula) is equivalent to the far-field spectral difference when the sampling calibration and the prediction inversion area are not simultaneous (as Figure 3 shown). Figure 3a is a schematic diagram of the multi-factor detection of the light source, sample, detection orientation, and standard plate calibration. Figure 3b is a schematic diagram of the directional annotation of the fruit orientation difference of the fruit tree. Figure 3c is a schematic diagram of the prediction and calibration areas and different distances and multi-angles. The space to be measured is divided into a spectral invariant area and a spectral change area.

[0052]

[0053] By adjusting the distance and angle spatial orientation factors, the proportion of mixed pixels caused by spectral overlap and coherence factors is reduced. Through the multi-angle spectral detection database and the extraction of angle reflectivity characteristics, the corresponding relationship between multi-point spectra and a regional spectrum is established, and the method of iterative replacement and cross-validation is used to reduce the constraint condition of the root mean square error between the two spectra and improve the accuracy of the end-member characteristics of spectral unmixing separation.

[0054] Use the band math function of the envi software to perform far-field spectral replacement iteration to improve the accuracy of the quantitative detection model.

[0055] 1. Spatial partitioning and classification of the area to be measured: Calculate the boundary of the far-field spectral invariant area for the area to be measured for partitioning and classification, determine the spectral acquisition nodes and the spatial resolution angle, align the spectral invariant area of the field of view (FOV) of the spectrometer and the spectral camera, and ensure that the moving spectrometer can collect multiple points to reconstruct the spectral camera image. Through the multi-angle spectral detection database and the extraction of angle reflectivity characteristics, the corresponding relationship between multi-point spectra and a regional spectrum is established, and the method of iterative replacement and cross-validation is used to evaluate the accuracy of the quantitative inversion of the spatial area spectrum.

[0056] 2. Perform progressive sampling and classification marking in the illumination intensity enhancement direction. Through refined classification in different polarization directions. Through multi-angle polarized diffuse transmission acquisition, the spectral images of fruit tissue sections in different directions are obtained, and the quantitative relationship between the peak value of the polarization degree in the illumination direction and the roughness is obtained. The corresponding relationship between the quality distribution in different spatial directions of the fruit and the illumination intensity direction is established.

[0057] The specific steps are as follows: slice the fruit tissue slices in different directions (the polarization directions of the transverse slices and the longitudinal slices correspond to the multi-angle diffuse transmission tissue slice spectrum experiment: change the incident angle of 5 degrees, 10 degrees, and 15 degrees polarization direction (0, 45, 90, and 135 four polarization directions) to obtain the diffuse transmission spectrum of fruit tissue slices at different angles. Through the spectral morphological parameters at different angles, a multi-angle polarization spectral response function is established to quantitatively describe the fruit quality, and the illumination characteristics and the light response characteristic model of the fruit quality are obtained.

[0058] 3. Angle calibration of sampling calibration plate and sample: convert the distance between calibration plate and sample into angle by non-in-situ measurement, calculate the spectral angle difference between sampling calibration and sample prediction inversion area, calculate the spectral coherence u12 at different distances through 1-1, measure the spectral coherence u12' at different distances by spectrometer experiment, interactively verify the spectral coherence measurements at different distances, minimize the root mean square error of spectral coherence, substitute the fruit quality spectrum si(w) into formula 1-2 to obtain the far-field spatial spectral response function s(p,w).

[0059] 4. Azimuth calibration of sampling calibration area and prediction inversion area: The ground fruit quality spectrum is substituted into the LESS software for far-field spectrum simulation at different angles to obtain regional spectra of different fruit tree shapes. The scattering kernel function is reconstructed considering the spatial structure of the fruit tree, and the semi-empirical multi-angle BRDF model of Roujean is used for angle calibration to verify and invert the structure parameters of the fruit tree. The physical and chemical indicators of the ground fruit are measured, and the far-field spectrum at different angles is replaced and iterated using the band math function of the ENVI software. The error is reduced and the accuracy is improved through interactive verification of simulation and experimental results, and a quantitative model of multi-angle spatial spectrum response of fruit tree fruit quality is established.

[0060] 5. Prediction and inversion of fruit trees and fruits in unknown areas: Based on the ground point selection method of spectral partitioning, collect ground fruit spectra and substitute them into the fruit tree and fruit spatial spectral response function model to obtain the fruit area spatial spectrum s1. The multi-angle spectrum s1' of the fruit tree canopy from the drone is used to minimize the root mean square error between s1 and s1' through interactive verification. The fruit tree canopy spectrum is substituted into the fruit tree and fruit multi-angle quantitative detection model to obtain the fruit quality in the unknown area. The quality evaluation of the fruit in the orchard area is obtained through the spatial structure information of the orchard, such as the spacing between fruit trees and terrain data. Then, the multi-scale quantitative inversion of fruit is realized from indoor to outdoor, from point to surface, from drone to satellite.

[0061] 6. Figure 1 As shown, based on the UAV near-ground remote sensing image data corresponding to the orchard area to be tested and the ground experimental control points, the temperature characteristics and illumination characteristics of the orchard area to be tested are processed using preset processing rules to obtain the gridded temperature characteristics and gridded illumination characteristics corresponding to the orchard area to be tested, and the gridded temperature characteristics and the gridded illumination characteristics are marked with partition classification marks;

[0062] Abstract and simplify the spatial structure of the fruit tree canopy in the orchard area to be measured using preset rules, and simulate, calculate, and mark the spatially spectral invariant region and the spatially spectral variation regions at different angles corresponding to the spatial structure of the fruit tree canopy using the spectral scaling law;

[0063] Based on the spatially spectral invariant region and the spatially spectral variation regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, as well as the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured, conduct a near-ground fruit tree sub-region classification marking experiment corresponding to the orchard area to be measured using preset experimental rules to obtain a fruit growth illumination characteristic model;

[0064] Train the fruit growth illumination characteristic model according to a preset training strategy, extract key parameters with obvious environmental illumination characteristics, and use them to improve the fruit growth illumination characteristic model, so as to determine the fruit quality of the orchard area to be measured through the improved fruit growth illumination characteristic model.

[0065] Preferably, based on the near-ground remote sensing image data of the unmanned aerial vehicle corresponding to the orchard area to be measured and the ground experimental control points, use preset processing rules to process the temperature characteristics and illumination characteristics of the orchard area to be measured, obtain the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured, and the grid temperature characteristics and the grid illumination characteristics are marked with sub-region classification markings, including:

[0066] Obtain the near-ground remote sensing image data of the unmanned aerial vehicle corresponding to the orchard area to be measured and the ground experimental control points;

[0067] Based on the near-ground remote sensing image data of the unmanned aerial vehicle corresponding to the orchard area to be measured, use preset grid processing rules to perform grid processing on the temperature characteristics and illumination characteristics of the orchard area to be measured, and obtain the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured;

[0068] Based on the ground experimental control points corresponding to the orchard area to be measured, use preset sub-region classification rules to perform sub-region classification marking on the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured, and obtain the grid temperature characteristics and grid illumination characteristics marked with sub-region classification markings.

[0069] Preferably, the abstracting and simplifying the spatial structure of the fruit tree canopy in the orchard area to be measured using preset rules, and simulating, calculating, and marking the spatially spectral invariant region and the spatially spectral variation regions at different angles corresponding to the spatial structure of the fruit tree canopy using the spectral scaling law, includes:

[0070] Obtain multi-scale and multi-angle images of the fruit trees in the orchard area to be measured;

[0071] Perform grid feature extraction on the multi-scale and multi-angle images of the fruit trees in the orchard area to be measured, and obtain the abstract and simplified spatial structure of the fruit tree canopy;

[0072] Use the spectral scaling law to simulate, calculate and mark the spatially spectral invariant regions and spatially spectral varying regions at different angles corresponding to the spatial structure of the fruit tree canopy.

[0073] Preferably, based on the spatially spectral invariant regions and spatially spectral varying regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, as well as the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured, carry out the near-ground fruit tree zoning and classification marking experiment corresponding to the orchard area to be measured using a preset experimental rule to obtain a fruit growth illumination characteristic model, including:

[0074] Based on the spatially spectral invariant regions and spatially spectral varying regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, carry out the near-ground fruit trees corresponding to the orchard area to be measured. For the near-ground fruit trees corresponding to the orchard area to be measured, use quantitative descriptions of different shapes of ROIs to describe the mean distribution characteristics, max distribution characteristics and structural characteristics of fruits in different regions;

[0075] Based on the mean distribution characteristics, max distribution characteristics and structural characteristics of fruits in different regions, obtain the semi-variogram of ROIs at different angles, use the anisotropy index ANIF and BRVF to describe the statistical characteristics of BRDF in different directions, and use the fruit roughness and polarization peak power function relationship model to describe the spatial structure characteristics of the fruit trees in the orchard area to be measured;

[0076] Based on the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured and the spatial structure characteristics of the fruit trees in the orchard area to be measured, establish a statistical correlation and significance analysis of the four factors of temperature, illumination, angle and polarization to obtain key parameters;

[0077] Use the key parameters to establish calibration and comparison verification data, and obtain the fruit growth illumination characteristic model.

[0078] Preferably, train the fruit growth illumination characteristic model according to a preset training strategy, and refine the key parameters with obvious environmental illumination characteristics to improve the fruit growth illumination characteristic model, including:

[0079] Based on the preset fruit sugar-acid ratio and corresponding spectrum, as well as the calibration and comparison verification data, establish a preset training strategy;

[0080] Based on the preset training strategy, use the calibration and contrast verification data to train the fruit growth light characteristic model to change the weights of the relevant parameters of the fruit growth light characteristic model;

[0081] Based on the preset training strategy, use the preset fruit sugar-acid ratio and the corresponding spectrum as constraint conditions to train the fruit growth light characteristic model to extract key parameters with obvious environmental light characteristics.

[0082] Preferably, the method further includes:

[0083] For the modeling method described above, first perform modeling indoors and then outdoors.

[0084] Preferably, the method further includes:

[0085] When performing modeling outdoors, it is necessary to establish an angular spectral response function model and transfer multi-angle and spectral quantitative models.

[0086] Example 1.

[0087] The specific operation steps are as follows:

[0088] (The specific modeling method is as shown in 2a in Figure 2 , and for auxiliary reference, see 2b and 2c):

[0089] (1) Precise angular calibration of the indoor BRDF device:

[0090] Perform precise angular calibration on the indoor BRDF device, and use the BRDF device to detect fruits to characterize the angular reflection characteristics of the fruits; based on the angular reflection characteristics of the fruits obtained by the BRDF device, select the angular reflectance characteristic angles of the fruits to establish an angular database; based on the angular database, calibrate according to the standard plate angles in different detection spaces to establish a fruit angular reflectance model.

[0091] Based on the fruit angular reflectance model, the BRDF device collects the spectral reflectances of crispy-ripe jujubes at different angles; at the same time, the spectrometer collects the polarized spectral data of the crispy-ripe jujubes at different angles; through a large number of experiments, fit the spectral reflectances and polarized spectral data of the crispy-ripe jujubes at different angles to establish a fruit angular polarized spectral quantitative detection model for quantitatively detecting the fruit quality, where the fruit quality is the sugar content and moisture of the fruits.

[0092] Among them, the reason for selecting the red dates in the crisp and ripe period is that the surface of the red dates in the crisp and ripe period is smooth and the polarization characteristics are obvious, which is convenient for establishing the quantitative detection model of the angle polarization spectrum of the fruit; it should be noted that the red dates in the crisp and ripe period can be selected from a variety of colors, such as red, green and red and white. And during the data collection, the above-mentioned BRDF device and spectrometer use a ring light source to light up in sequence to simulate the lighting conditions of the sun at different periods, thus avoiding the problem of image registration; the BRDF device can collect the multi-angle reflectivity of the orchard area to be tested under the corresponding typical lighting conditions at different periods, forming a typical illumination multi-angle reflectivity database of the orchard area to be tested at different periods.

[0093] BRVF calculation method: BRVF is a statistic based on BRDF, which is used to characterize the spatial variation of BRDF in direction. It describes the variance or degree of variation of BRDF values ​​at different observation angles. The calculation method is as follows: 1) Obtain BRDF data: Measure or calculate the BRDF value of the object at different incident angles and observation angles. 2) Calculate variance: Use statistical methods to calculate the variance of these BRDF values ​​to obtain BRVF.

[0094] Steps for anisotropy direction analysis based on semivariogram:

[0095] Directional semivariogram fitting: 1) Divide the space into multiple directions or polarization directions (such as 0 space 45 intervals 90 intervals 135 intervals), and set the angle tolerance (such as ±0.5°). Calculate the semivariogram for each direction and fit the model (such as spherical, exponential model) 2) Extract parameters: record the range (aθ) and sill value (Cθ) of each direction 3) Calculate geometric anisotropy: ANIF = max(aθ) / min(aθ), zonal anisotropy ANIF = max(cθ) / min(cθ) 4) Evaluate the interpolation accuracy of the anisotropic model through cross-validation. The semivariogram is a quantitative tool that describes the overall heterogeneity through the nugget value, sill value and range, and further reveals the directional heterogeneity through ANIF.

[0096] (2) Grid angle and azimuth calibration spectral model for indoor space to be measured:

[0097] ① Theoretical simulation calculation of different fruit scattering angle characteristic models:

[0098] Multi-scale scattering mechanism, through the characteristics of scattering angles in different directions, quantitatively describes the scattering characteristics with scattering coefficients, realizing cross-scale quantitative detection and evaluation. The equivalent scattering medium of fruits quantitatively describes different quality characteristics with particle size and refractive index. According to different ripening periods of fruits, tissue sections of individual fruits are collected to measure the pore size distribution. At the single-fruit scale, the mieplot scattering spectrum software is used to obtain the scattering cross-section, scattering coefficient, and scattering angle. Information such as the tree height and row spacing of the fruit tree shape (round head type, happy type) is collected. Calculate multi-scale scattering characteristics such as the scattering coefficient and absorption coefficient of the fruit tree canopy. At the fruit canopy scale, the three-dimensional radiative transfer less software can be used to simulate and calculate the scattering parameters of different fruits, such as the scattering characteristics at different scales, the scattering coefficient and absorption coefficient of leaves, and the reflectivity at different angles. Obtain the scattering coefficient and absorption coefficient of the fruit tree canopy. Select appropriate acquisition angles and grid patterns for spatial orientation calibration. The scattering coefficients and absorption coefficients of single fruits and canopies quantitatively characterize multi-scale scattering characteristics.

[0099] ② Method for gridifying the space to be measured according to the spectral coherence theory:

[0100] According to the spectral spatial coherence theory: The spectral coherence degree between two points for calibration and prediction is quantitatively described by the van Cittert of partially coherent light. The distance difference between adjacent two points is also the spectral coherence width, and the spatial correlation infers the correlation characteristics of different scale surface regions. Since the spatial coherence is uniform and stable only at a very small field of view angle or vertical observation, (as shown in Figure 3 Figure c) is also the theoretical basis for spatial classification and zoning marking.

[0101] The spatial solid angle △Ω 2 / 2 The spatial solid angle is related to the wavelength λ and the source size a, and is independent of the distance. Usually, the farther the distance, the larger the coherent area. [1] .

[0102] The coherence length L = 2 / △λ and A = 0.063R 2 λ 2 / = 0 2 The coherence area divides the boundary of the spatial region. For non-in-situ quantitative detection sampling of fruits, the corresponding angles of the calibration and prediction inversion regions are different. The coherent region is determined by calculating the solid angle through the corresponding characteristic wavelength. If it exceeds the range of the spatial solid angle, classification marking is performed. The solid angle is equivalent to the spectral correlation between any two points in space, so the classification marking of the coherent region is also an important constraint condition for multi-scale spectral imaging quantitative detection.

[0103] The spatial resolution of the camera is affected by the lens and the detection distance and orientation. The spatial area boundary is divided according to the diffraction angle of the camera lens θ=1.22λ / D=x / y and the field of view angle FOV=x / y of the front end of the camera lens, where D represents the diameter of the lens, x represents the distance between two points in space, and y represents the vertical distance from the central axis of the lens to the surface to be measured; the space to be measured is gridded by shooting height and field of view. Considering that the optical transfer function (MTF) describes the spatial frequency in different directions, as the distance increases, the field of view angle FOV becomes smaller, the spatial cutoff frequency decreases, and the high-frequency detail information cannot be obtained, resulting in a decrease in the spatial resolution of the camera; the solution of this embodiment is to obtain multi-angle acquisition data, and use the multi-angle acquisition data iterative algorithm to perform frequency splicing to compensate for the decrease in spatial resolution caused by the spatial cutoff frequency, thereby improving the spatial resolution of the camera; this embodiment improves the spatial resolution of the camera image by changing different angles, increasing large-angle high-frequency image details and spatial frequency registration.

[0104] Measurement of surface roughness of large-angle 3 fruits and the power function relationship between polarization peak and roughness: 3D surface roughness measuring instrument measurement, resolution is 0.005 roughness. Measurement parameters include arithmetic mean deviation (Ra), root mean square deviation (Rq) and ten-point average deviation (Rz). For each sample, 4 evenly distributed measurement areas are selected on the equatorial plane, and each area is scanned with a 5mm distribution. The number of scanning points is 1000 points / mm, and the average value of the 4 areas is calculated as the surface roughness characteristic value of the sample.

[0105] To ensure the accuracy and reliability of the measurement results, all measurements are carried out under standard environmental conditions (temperature 22°C, relative humidity 50°C). Before each measurement officially begins, a standard sample is used for calibration: the colorimeter is calibrated with a standard white plate; the polarization system is calibrated with a standard polarizer and reflector; the surface roughness meter is calibrated with a standard roughness sample block. All measurements are cross-validated for the measurement results.

[0106] The surface roughness of the harvested red dates was significantly higher than that of the fully ripened red dates. a The average value was 6.74μ., which was 19.3% higher than that of 5.65μ. in the fully ripe stage. a The average value is 8.18μ, an increase of 18.2% from 6.92μ in the fully ripe stage. This change reflects the significant changes in the cell structure of red dates during the harvest period, mainly manifested in the thickening of cell walls, the tighter arrangement of flesh cells, and the shrinkage and deformation of epidermal cells.

[0107] Collect sample spectra at four polarization angles and calculate the polarization index P 0-90 and P 45-135 The peak value of polarization degree of red dates at harvest stage is significantly lower than that of red dates at full maturity stage. 0-90The mean value is 0.096, which is 14.3% lower than 0.112 at the fully ripe stage; the P of the whole fruit sample 0-90 The mean value is 0.078, which is 15.2% lower than 0.092 at the fully ripe stage. The peak of polarization degree mainly appears in the wavelength range of 1456 - 1460 nm, corresponding to the combined band of O stretching vibration and bending vibration of water molecules.

[0108] Analyze the relationship between surface roughness and the peak of polarization degree. The R of the sliced sample a and P 0-90 and P 45-135 The correlation coefficients are -0.904 and -0.886 respectively; the correlation coefficients of the whole fruit sample are -0.892 and -

[0109] 0.874, both showing a very strong negative correlation. Further establish the power function model of surface roughness and the peak of polarization degree:

[0110]

[0111] For the sliced jujube samples at the harvest stage, the P 0-90 The model parameters are

[0112] C = 0.582, and the product correlation. Further establish R 2 = 0.968; the parameters of the whole fruit sample are C =

[0113] 0.548, and the correlation. Further establish R 2 = 0.962. Compared with the fully ripe jujubes, the power function exponent γ of the jujubes at the harvest stage increases significantly, indicating that the influence of surface roughness on the peak of polarization degree is further enhanced.

[0114] The main mechanisms of surface roughness affecting polarization characteristics include the scattering enhancement effect and the multiple scattering effect. According to Beckmann scattering theory, the relationship between the scattered light intensity and surface roughness is:

[0115]

[0116] The calibration steps include: measuring the surface roughness parameter R of the sample a ; calculating the theoretical peak of polarization degree P according to the power function model theo ; measuring the actual peak of polarization degree P meas ; calculating the calibration coefficient K = P theo / P meas ; calibrating the spectral data: I corr = K·I orig。 . Apply this calibration method to the moisture content detection model of jujubes at the harvest stage.

[0117] For example:

[0118] Such asFigure 3 As shown in FIG. 3 , the spatial angle and distance of the camera shooting are changed in sequence. It can be seen that the farther the distance is, the smaller the camera field of view is, and the more corresponding missing frequency information is. Therefore, this embodiment compensates for the missing frequency information by shooting images at more angles. The specific method is as follows: Figure 3 As shown in c, the accuracy of gridded area radiation calibration is improved by iteratively taking images at long distances and multiple angles and interactively verifying the predicted image positions, thereby improving the accuracy of gridded space to be tested.

[0119] A correlation analysis was performed on the significant factors of reflectivity, angle, polarization degree, and soluble solids SSC; the characteristics of the correlation coefficient under the extremely significant condition of 0.01 showed that dolp, band, and ssc were significantly correlated.

[0120] ③Multi-angle ground profiling calibration method:

[0121] By gridding the space to be measured, the control points are selected to divide the inversion area, and the optical fiber spectrometer and multi-spectral camera are used for fixed-point multi-angle calibration and mutual verification. Specifically, the illumination characteristics of a single fruit are divided, and multiple fruits are combined according to the region of interest (ROI) of a single row of fruits to form a single row reflectance feature, and multiple fruits are combined according to the region of interest (ROI) of a vertical column of fruits to form the row and column characteristics of the overall image of the fruit. The reflectance characteristics of the fruit quality in the entire area are inverted by combining the regions of interest (ROI) of different rows and columns of fruits. A multi-angle profiling ground calibration field for fruit trees is established based on the structural parameters such as the growth tree shape, tree height, row spacing, and plant spacing of fruit trees in the orchard. A multi-angle profiling calibration field is established through the gridded orchard regional characteristics, and the gridded spatial spectrum of the multi-angle ground profiling calibration field is interactively verified with the image, and a multi-angle polarization spectral response function model for the fruit tree canopy is established. Combine spectral quantitative detection with regional evaluation of drone spectral images.

[0122] The multi-angle reflectance database is modified by inverting the reflectance of fruits at different positions through interactive verification from the reflectance of a single fruit, realizing the one-to-one correspondence between the spectral reflectance and the corresponding position of the spectral image, and by spatially segmenting fruits at different angles and distances and matching the angles and orientations in the model library. A multi-angle spectrum and image data model library is established to realize multi-angle quantitative detection.

[0123] (3) Azimuth calibration spectral image calibration model:

[0124] The standard plate is used to establish indoor calibration fields for calibration at different angles, and polarization calibration and spatial orientation calibration of the standard plate are carried out. The spectrum is calibrated at a certain point in the image, and the geometrically calibrated image is calibrated at different orientations through the spectral coherence law. The spectrum and image are calibrated and verified with each other. The spectrum is synthesized into an image, and the spectrum and angle reflectivity information are extracted from the image.

[0125] The experiment of calibration plates with different orientations is used to establish a multi-angle calibration database. The multi-angle reflectance experiment of fruits is carried out, and the angle and azimuth model is cross-validated and optimized to establish a multi-angle spatial azimuth spectral model of fruits. A spectral angle reflectance model is established to achieve refined quantitative description of spatial spectral images in different orientations. Specifically:

[0126] ① Establish a calibration model for the angle of the standard white board

[0127] Since there are differences in the reflectance of the standard white board at different angles during the actual experiment, which has a certain impact on the detection accuracy, it is necessary to calibrate the reflectance of the standard white board at different angles. The impact brought by the angle of the standard white board is greater than that of the fruit quality. By establishing a functional relationship for the optimal angle of the standard white board, the correlation coefficients at different angles of the standard white board are calculated to predict the placement angles of the standard white board under different detection conditions.

[0128] ② Establish a calibration model for the distance of the standard white board

[0129] During the transmission of the spectrum, it will be affected by the spectral coherence. There are certain differences in the sample spectra at different distances. The spectra are corrected using standard white boards at different distances from the sample, and a calibration model for the standard white board at different distances is established to calibrate the sample spectra at different distances.

[0130] (4) Establish an angular spectral response function model:

[0131] Multi-spectral camera multi-angle polarization outdoor angle calibration, spatial position calibration, geometric calibration and radiometric calibration. Select the spectral response function and angular response function by choosing the spectral background of the characteristic azimuth space. The multi-angle experiment equivalently improves the spectral resolution and detection accuracy. And when there are large differences in the reflectance at different angles, the multi-angle differences in the spectrum and light intensity, and the spectral camera combining the spectrum and imaging give the spatial distribution characteristics of the fruit quality. The correlation analysis of the angle, wavelength, polarization, and reflectance correlation coefficients is fitted into an angular function model. Place the standard plate during imaging, perform radiometric calibration, geometric calibration and ambient light field calibration. Refined quantitative description of the ambient light and angle, and construct a spatial angular azimuth spectral response function.

[0132] (5) Multi-scale multi-angle and spectral quantitative detection model transfer:

[0133] On the basis of consistent fruit quality, select the spectral reflectance, angle, polarization, and reflectance collected under the same conditions to establish a statistical regression relationship model. Take the spatial azimuth distance, angle, and polarization direction as key parameters, and fit the experimental data to the angle, wavelength, reflectance, and polarization characteristic data to establish a correlation coefficient model. Through the transfer of the multi-angle reflectance quantitative detection database and the spectral database quantitative detection model, multi-scale multi-angle quantitative detection is achieved.

[0134] (6) Spatial grid division of the outdoor area to be measured:

[0135] Select the regional orchard area image through satellite images, and obtain the orchard with obvious color differences in the image through differential processing as the research area. Conduct a calibration experiment on the comparison of reflectance images from multiple angles near the ground by drones and from multiple angles on the ground. Use a ground illuminometer and a ground thermometer to obtain the ground temperature and conduct an experiment on the temperature field distribution by a thermal imager. Use the thermal imager and spectrometer for spatial division and partition calibration, and reconstruct the calibration models of typical environmental temperature, illumination, and orientation in the orchard.

[0136] ① Grid model of the spatial temperature and illumination field in the area to be measured:

[0137] Based on the temperature and illumination reflectance differences at different test points, and according to the actual orchard situation, taking the difference between each point being less than a certain specific value as the constraint condition, select typical environmental control points to conduct grid processing on the geographical orientation, temperature, and illumination conditions of the area to be measured. Extract the typical environmental orientation, temperature, and illumination characteristics of the area through outdoor orchard spatial grid processing. Establish a multi-angle profiling calibration field through the characteristics of the grid orchard area, and combine spectral quantitative detection with the evaluation of drone spectral image areas.

[0138] ② Grid model of the temperature and illumination field at different time periods in the area to be measured:

[0139] Since the solar altitude angle and azimuth angle change at different time periods, the above spatial grid model will change. The changed area and angle can be determined by simulating and calculating the solar altitude angle and azimuth angle at different time periods to correct the above grid space area. For example, it is about 1 degree in half an hour, and at the same time, the illuminance needs to be corrected.

[0140] ③ Reconstruct the typical environmental orientation model

[0141] Through the simulation calculation of the geographical information coordinates (GPS) of the orchard fruit trees that determine the geographical orientation, and the solar altitude angle and azimuth angle at different time periods of morning, noon, and evening, and through the reconstruction and reproduction of the typical environmental illumination of the fruit quality spectrum, accurately and quantitatively describe the fruit area such as geographical coordinates and growth orientation characteristic spectra. Establish a calibration spectrum model of the typical environment and fruit orientation of actual fruit trees.

[0142] (7) Multi-angle polarization reflectance experiment of drone fruit near the ground:

[0143] ① Multi-angle polarization azimuth experiment of drones at different heights:

[0144] The multi - spectral camera of the drone collects at different heights of 3 meters, 5 meters, and 8 meters. At each height, the drone hovers, and a polarizer is horizontally added to the spectral camera in four polarization directions (0p, 45p, 90p, 135p). It rotates 8 directions (0, 45, 90, 135, 180, 225, 270, 315) every 45 degrees to obtain spectral images from multiple angles. The longitude and latitude in the header file of each picture are modified, with the longitude and latitude coordinates of the jujube orchard, and the longitude and latitude coordinates of the area to be measured are input. For example, for the spectral image of E81, the longitude and latitude in the header file of each picture are modified.

[0145] In this embodiment, it is specifically input and arranged in the order of the following 4 polarization directions: 0p(0, 45, 90, 135, 180, 225, 270, 315), 45p(0, 45, 90, 135, 180, 225, 270, 315), 90p(0, 45, 90, 135, 180, 225, 270, 315), 135p(0, 45, 90, 135, 180, 225, 270, 315). There are 4 * 8 * 6 = 192 reflected - band images in one scene. Different angles and polarization azimuths are used as the abscissa of the multi - angle polarization reflectance. The Envi software synthesizes 192 reflectance files to establish a multi - angle reflectance database of fruits of different orientations with classification marks.

[0146] ② Collection of physical and chemical indexes of artificially marked fruits and classification - marked spectral determination experiment:

[0147] Use a spectrometer to collect the spectra of in - situ artificially marked fruits, record the corresponding fruit - shape characteristics such as color and texture, and measure the physical and chemical indexes such as fruit moisture and sugar content. Through the spectral database of fruits of different orientations, establish a quantitative detection model for the spectra of fruits of different orientations.

[0148] ③ Classification - marked fruit spectral reflectance and multi - angle reflectance cross - validation model

[0149] For the ROI characteristic region of the multi - spectral multi - angle image of the drone, select red and green fruits for classification marking, such as the reflectance of red dates in green and red according to color. Further orientation marking and single - quality marking detection are carried out to achieve the quantitative characterization of fruits, providing a reference for quality classification and pricing.

[0150] Example 2. Comparative verification of multi - scale fruit quality simulation theory:

[0151] 1) Combine the mieplot simulation of scattering spectrum software and the less simulation of radiative transfer software, combine the spectral invariance theory and the P - model of the vegetation multiple - scattering spectral invariance theory, compare and verify the multi - scale fruit scattering characteristics with fruits equivalent to random scattering media, and establish a multi - angle scattering characteristic model

[0152] 2) The fruit is equivalent to a random scattering medium. The refractive index distribution quantitatively characterizes the differences in quality components. The high-order optical parameter g-function describes the structure, and the scattering potential function describes the quality of fruits at different maturity stages. Driven by the experimental data of multi-angle diffuse transmission and combined with the parameter-driven model of key fruit parameters, the parameters are optimized for the experiments on the differences in the near-field and far-field spectral density distributions of fruits, and a multi-scale scattering kernel-driven function model for fruit quality is constructed.

[0153] 3) The spectral coherence degree and spectral polarization degree of the spectral image quantitatively describe the spectral modulation, describe the shape of the light source from the angles of the illumination direction and intensity, the relationship between the spatial coherence of the light source and the propagation distance, the cross-correlation function describes the correlation and coherence relationship, and the spectral and cross-spectral densities represent the temporal coherence and spatial coherence respectively.

[0154] 4) A multi-angle polarization reflectance database is established for detection in different directions using single wavelength and multi-wavelength respectively. By studying the variation laws of the scattering spectra of different fruit qualities, multi-angle polarization characteristic parameters are extracted to establish a scattering polarization model.

[0155] The flow chart is as Figure 4 、 Figure 5 shown.

[0156] As described above, it is only the preferred embodiment of the embodiments of the present invention, and does not impose any form of limitation on the embodiments of the present invention. Any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the embodiments of the present invention still fall within the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A modeling method for quantitative detection of outdoor multi-scale fruit zoning and classification marking, characterized in that, The method includes: Based on the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points, using preset processing rules to process the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, and the grid temperature characteristics and the grid light characteristics are marked with zoning and classification marks; Using preset rules to abstract and simplify the spatial structure of the fruit tree canopy in the orchard area to be measured, and using the spectral scaling law to simulate, calculate and mark the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy; Based on the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy in the orchard area to be measured, and the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, using preset experimental rules to conduct a near-ground fruit tree zoning and classification marking experiment corresponding to the orchard area to be measured to obtain a fruit growth light characteristic model; Training the fruit growth light characteristic model according to a preset training strategy, extracting key parameters with obvious environmental light characteristics to improve the fruit growth light characteristic model, so as to determine the fruit quality of the orchard area to be measured through the improved fruit growth light characteristic model.

2. The method according to claim 1, wherein The step of, based on the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points, using preset processing rules to process the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, and the grid temperature characteristics and the grid light characteristics are marked with zoning and classification marks, includes: Obtaining the near-ground remote sensing image data of the orchard area to be measured and the ground experimental control points; Based on the near-ground remote sensing image data of the orchard area to be measured, using preset grid processing rules to perform grid processing on the temperature characteristics and light characteristics of the orchard area to be measured, obtaining the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured; Based on the ground experimental control points corresponding to the orchard area to be measured, using preset zoning and classification rules to perform zoning and classification marking on the grid temperature characteristics and grid light characteristics corresponding to the orchard area to be measured, obtaining the grid temperature characteristics and the grid light characteristics marked with zoning and classification marks.

3. The method according to claim 1, characterized in that, The step of using preset rules to abstract and simplify the spatial structure of the fruit tree canopy in the orchard area to be measured, and using the spectral scaling law to simulate, calculate and mark the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy, includes: Obtaining multi-scale and multi-angle images of the fruit trees in the orchard area to be measured; Performing grid feature extraction processing on the multi-scale and multi-angle images of the fruit trees in the orchard area to be measured to obtain an abstracted and simplified spatial structure of the fruit tree canopy; Using the spectral scaling law to simulate, calculate and mark the spatially spectral invariant region and the spatially spectral change regions at different angles corresponding to the spatial structure of the fruit tree canopy.

4. The method according to claim 1, characterized in that, Based on the spatially spectral invariant regions and spatially spectral varying regions at different angles corresponding to the fruit tree canopy spatial structure of the orchard area to be measured, as well as the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured, carry out the near-ground fruit tree sub-region classification marking experiment corresponding to the orchard area to be measured using a preset experimental rule to obtain the fruit growth illumination characteristic model, including: Based on the spatially spectral invariant regions and spatially spectral varying regions at different angles corresponding to the fruit tree canopy spatial structure of the orchard area to be measured, carry out light intensity progressive sub-region classification marking sampling for the near-ground fruit trees corresponding to the orchard area to be measured, and use quantitative descriptions of different shapes of ROIs to describe the mean distribution characteristics, max distribution characteristics, and structural characteristics of fruits in different regions; Based on the mean distribution characteristics, max distribution characteristics, and structural characteristics of fruits in different regions, obtain the semi-variogram of ROIs at different angles, use the anisotropy index ANF and BRVF to describe the statistical characteristics of BRDF in different directions, and use the fruit roughness and polarization peak power function relationship model to describe the fruit tree spatial structure characteristics of the orchard area to be measured; Based on the grid temperature characteristics and grid illumination characteristics corresponding to the orchard area to be measured and the fruit tree spatial structure characteristics of the orchard area to be measured, establish a statistical correlation and significance analysis of the four factors of temperature, illumination, angle, and polarization to obtain key parameters; Use the key parameters to establish calibration and comparison verification data, and obtain the fruit growth illumination characteristic model.

5. The method according to claim 4, wherein Training the fruit growth illumination characteristic model according to the preset training strategy, and refining key parameters with obvious environmental illumination characteristics to improve the fruit growth illumination characteristic model, including: Based on the preset fruit sugar-acid ratio and corresponding spectrum, as well as the calibration and comparison verification data, establish a preset training strategy; Based on the preset training strategy, use the calibration and comparison verification data to train the fruit growth illumination characteristic model to change the weights of the relevant parameters of the fruit growth illumination characteristic model; Based on the preset training strategy, use the preset fruit sugar-acid ratio and corresponding spectrum as constraint conditions to train the fruit growth illumination characteristic model to refine key parameters with obvious environmental illumination characteristics.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: For the modeling method, first carry out modeling indoors and then outdoors.

7. The method according to claim 6, wherein The method further includes: When carrying out modeling outdoors, it is necessary to establish an angle spectral response function model and multi-angle and spectral quantitative model transfer.

Images