Sugar beet yield prediction method and system based on near-infrared remote sensing texture features

Abstract

The present application provides a kind of based on near infrared remote sensing texture feature's sugar beet yield prediction method and system, it is related to sugar beet yield prediction technical field, the present application obtains monitoring area remote sensing image and extracts near infrared band data, generates standard gray texture map by nonlinear stretching, effectively enhances crown layer texture details, subsequently traverses full map using spatial coexistence constraint condition, constructs omnidirectional gray coexistence probability matrix eliminating directionality difference, solves the problem of being limited by scanning direction and image size when extracting traditional texture, and then extracts leaf broken degree representing the degree of leaf edge overlap and local cohesion representing the tightness of crown layer growth, from spatial distribution dimension, it is sensitive to capture the micro feature of crop high biomass area, finally, generate crown layer shrinkage index using exponential decay mechanism, and substitute into nonlinear inversion equation to output sugar beet predicted yield.

Description

Technical Field

[0001] This invention relates to the field of sugar beet yield prediction technology, specifically to a method and system for predicting sugar beet yield based on near-infrared remote sensing texture features. Background Technology

[0002] With the development of precision agriculture, sugar beets, as an important sugar source, require yield forecasting for both raw material allocation by sugar mills and field management by farmers. Traditional manual sampling forecasting methods are not only inefficient but also fail to reflect the spatial heterogeneity of plots. In recent years, remote sensing technology, as a non-destructive, large-scale monitoring method, has gradually replaced traditional experience-based judgments, becoming a core element in realizing the transformation of agricultural production from extensive management to data-driven approaches.

[0003] Existing technologies primarily focus on using spectral reflectance characteristics to invert crop growth parameters. This typically involves extracting vegetation indices or calculating the photosynthetically active radiation absorption coefficient, constructing a transformation logic based on spectral intensity to photosynthetic efficiency and then to biomass. However, this type of technology has significant limitations when estimating yields for root crops: First, once the crop canopy closes, indices based on spectral intensity experience a saturation effect, leading to reduced differentiation of high-yield areas; second, existing schemes often focus on the color dimension, neglecting the crucial biological information contained in canopy texture, lacking a description of the nonlinear biological allometric growth mechanism between leaf spatial morphology and underground tuber enlargement, making it difficult to achieve high-precision and robust predictions in complex field environments.

[0004] In the prior art, CN117876870A discloses a method and system for crop yield estimation based on multi-source remote sensing data. The method includes constructing a mapping logic between spectral characteristics and yield using the photosynthetically active radiation absorption coefficient. However, this scheme relies solely on spectral intensity, which suffers from spectral saturation during the canopy-closing stage of sugar beets, leading to reduced discrimination of high biomass areas. Furthermore, it neglects key information contained in canopy texture and lacks a description of the nonlinear mechanism from canopy shrinkage to tuber enlargement. To address these shortcomings, this invention constructs a canopy shrinkage index by extracting leaf fragmentation and local cohesion, and establishes a nonlinear inversion equation based on the biological allometric growth law, significantly improving the robustness and biological interpretability of high-yield area predictions.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting sugar beet yield based on near-infrared remote sensing texture features, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting sugar beet yield based on near-infrared remote sensing texture features, comprising the following steps: Step 1: Acquire remote sensing image data of the sugar beet planting area to be monitored, extract near-infrared band data that characterizes vegetation reflectance as the original intensity input, perform nonlinear stretching enhancement and discretization processing on the original intensity input, and generate a standard grayscale texture map that enhances dark details. Step 2: Based on the obtained standard grayscale texture map, define several scanning displacement vectors in several directions. Using the spatial co-occurrence constraint, traverse all pixels in the image and their neighboring pixels. For all pairs of spatial pixels to be judged, which are composed of the current center pixel and the neighboring pixels pointed to by the displacement vector in all scanning directions, count, sum and normalize them to construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. Step 3: Based on the omnidirectional gray-level co-occurrence probability matrix, feature vectors describing the sugar beet canopy structure are extracted from the two dimensions of texture intensity and spatial distribution. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth. Step 4: Couple the blade fragmentation and local cohesion to generate a canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion. Step 5: Construct a nonlinear inversion equation, input the canopy wrinkling index into the nonlinear inversion equation, and after correction by the regional productivity coefficient, output the final beet yield prediction result.

[0008] Furthermore, the specific steps for constructing the standard grayscale texture map include: extracting near-infrared band data representing the internal structure of vegetation from the acquired remote sensing image data, and using it as the original digital quantization value; converting the original digital quantization value into physical reflectance using preset radiometric calibration parameters, while keeping the pixel spatial position unchanged; constructing a single-channel original physical reflectance matrix; traversing the matrix and calculating the global maximum physical reflectance value in the current scene. Using the global maximum physical reflectance value as a normalization benchmark, and combining it with a preset contrast enhancement coefficient and the total number of gray levels, the original physical reflectance matrix is ​​subjected to logarithmic nonlinear stretching and discrete quantization mapping to output a standard grayscale texture image. The specific calculation formula is as follows: In the formula, Represents the coordinates in the standard grayscale texture map. pixel grayscale levels, and These are the x and y indexes of the image pixels, respectively. This is a contrast enhancement coefficient used to adjust the curvature of the nonlinear mapping. In the original physical reflectivity matrix, the coordinates are... Physical reflectivity at that location This represents the global maximum physical reflectance value statistically analyzed from the current original physical reflectance matrix. This represents the total number of gray levels.

[0009] Furthermore, the logic for converting the original digital quantization value into physical reflectance using preset radiometric calibration parameters is as follows: based on the spectral response characteristics of the sensor in the near-infrared band, the continuous spectral radiation energy within the coverage area of ​​this band is weighted and accumulated, that is, according to the sensor's sensitivity weights to different wavelengths, the continuously changing spectral energy is converted into the original digital quantization value characterizing the average energy intensity of this band. The original digital quantization value is corrected and calculated using linear operational logic to obtain the equivalent physical reflectivity at the center wavelength of this band. The specific calculation formula is as follows: In the formula, The coordinates obtained by the solution are Physical reflectivity at that location The preset radiation calibration gain coefficient is used to characterize the photoelectric conversion responsivity of the sensor. The preset radiation calibration bias coefficient is used to characterize the zero-point drift correction of the system. Indicates coordinates as The original digital quantization value at that location.

[0010] Furthermore, the specific steps for traversing all pixels and their neighboring pixels using spatial co-occurrence constraints include: based on the standard grayscale texture map, setting several scanning directions covering the two-dimensional plane and their corresponding displacement vectors; defining each pixel coordinate in the image as a reference center pixel; locking the corresponding target neighboring pixel according to the displacement vector in the current scanning direction; and forming a spatial pixel pair to be judged by combining the reference center pixel and the target neighboring pixel. The spatial co-occurrence constraint condition is used to detect whether the spatial pixel pair to be judged matches a preset specific gray level combination, and a judgment signal is output accordingly. The calculation formula of the judgment signal is as follows: In the formula, To distinguish the signal, it is indicated that in the standard grayscale texture image, at the th Coordinates in each scanning direction The pixel grayscale value at that location is equal to And its neighboring pixel grayscale value is equal to The matching status, Index for the scan direction. This refers to the specific combination of gray levels that the current omnidirectional gray-level co-occurrence probability matrix is ​​statistically analyzing. Represented as the first Spatial displacement vectors in the scanning direction.

[0011] Furthermore, the step of constructing an omnidirectional gray-level co-occurrence probability matrix that eliminates directional differences includes: summing all spatial coordinates and scanning directions of the standard gray-level texture image using the discrimination signal, and statistically analyzing specific gray-level combinations. The occurrence frequency is used as the normalization benchmark. The cumulative total number of all valid undiscriminated spatial pixel pairs constructed during the traversal process is used as the normalization benchmark. The occurrence frequency is divided by the cumulative total number to calculate the omnidirectional gray-level co-occurrence probability value. For gray-level combinations that are not detected during the traversal process, their corresponding omnidirectional gray-level co-occurrence probability values ​​are directly set to zero. This constructs an omnidirectional gray-level co-occurrence probability matrix that is numerically complete and eliminates the influence of image size and orientation. The formula is as follows: In the formula, Represents the 1st gray-level co-occurrence probability matrix in the omnidirectional gray-level co-occurrence probability matrix. Okay, number The omnidirectional gray-scale co-occurrence probability value of the column. This represents the cumulative total number of undiscriminated spatial pixel pairs across all scanning directions in the standard grayscale texture image. The number of scanning directions covering the two-dimensional plane.

[0012] Furthermore, the step of obtaining the blade breakage degree is as follows: based on the omnidirectional gray-level co-occurrence probability matrix, the gray-level difference value is nonlinearly amplified using cubic weighting to extract the blade breakage degree. The specific calculation is based on the following formula: In the formula, Leaf fragmentation is used to characterize the degree of overlap at the leaf edges. and Here, represents the row and column indices of the omnidirectional gray-level co-occurrence probability matrix, which physically represent the gray-level values ​​of the two pixels constituting a pixel pair. It is the absolute value of the difference between gray levels. The total number of gray levels; The steps for obtaining the local cohesion are as follows: statistically analyze the gray-level expectation value of the omnidirectional gray-level co-occurrence probability matrix, establish the average brightness benchmark of the image, use inverse variance weighted logic to iterate and sum the omnidirectional gray-level co-occurrence probability matrix, and quantify the density of texture distribution by giving higher weights to gray-level combinations that are close to the average brightness benchmark, thereby extracting the local cohesion.

[0013] Furthermore, the formula for calculating the expected grayscale value is as follows: In the formula, To determine the expected grayscale value of the brightness mean baseline; The formula for calculating the local cohesive force is: In the formula, Local cohesion force, which characterizes the compactness of canopy growth, This is a sensitivity adjustment coefficient used to adjust the magnitude of weight decay. and Representing reference pixels respectively The difference between the grayscale expected value and the neighboring pixels With grayscale expectation The difference.

[0014] Furthermore, the specific steps for generating the canopy wrinkling index are as follows: A basic texture energy term is constructed by coupling leaf fragmentation and local cohesion using the geometric mean method; a nonlinear adjustment coefficient is constructed using exponential decay; and the basic texture energy term is then weighted and corrected to generate the canopy wrinkling index. The calculation formula is as follows: In the formula, The canopy wrinkling index is used to comprehensively characterize the complexity of canopy texture. This is the basic texture energy term constructed based on the geometric mean method. The degree of breakage of the blade. For the local cohesive force, To balance the order-of-magnitude difference between blade fragmentation and local agglomeration force, the sensitivity adjustment coefficient of the suppression force is adjusted. To prevent numerical stability constants with a denominator of zero.

[0015] Furthermore, the sugar beet yield prediction step is as follows: a nonlinear inversion equation is constructed based on the biological allometric growth law, wherein the biological allometric growth law characterizes the nonlinear power function correlation between the predicted sugar beet yield and the canopy wrinkling index. By retrieving preset regional feature parameters and substituting the canopy wrinkling index into the nonlinear inversion equation, the predicted sugar beet yield is output. The calculation formula is as follows: In the formula, To obtain the final predicted output, This serves as the baseline production volume for correcting regional production capacity figures. Biomass conversion coefficient The canopy wrinkling index is the index of the canopy. The growth allotropy index is used to characterize the degree of nonlinear response of yield to changes in canopy texture.

[0016] The present invention also provides a sugar beet yield prediction system based on near-infrared remote sensing texture features. This system is used to implement the aforementioned sugar beet yield prediction method based on near-infrared remote sensing texture features, and includes: The image preprocessing module is used to acquire remote sensing image data of the sugar beet planting area to be monitored, extract the near-infrared band data that characterizes the vegetation reflectance as the original intensity input, perform nonlinear stretching enhancement and discretization processing on the original intensity input, and generate a standard grayscale texture map that enhances the details in the dark areas. The matrix construction module is used to define several scanning displacement vectors based on the obtained standard grayscale texture image, traverse all pixels in the image and their neighboring pixels using spatial co-occurrence constraints, count, sum and normalize the pairs of spatial pixels to be judged in all scanning directions consisting of the current center pixel and the neighboring pixels pointed to by the displacement vector, and construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. The feature extraction module is used to extract feature vectors describing the sugar beet canopy structure from two dimensions: texture intensity and spatial distribution, based on the omnidirectional gray-level co-occurrence probability matrix. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth. The exponential calculation module is used to couple the blade fragmentation and local cohesion to generate the canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion. The yield inversion module is used to construct a nonlinear inversion equation. The canopy wrinkling index is input into the nonlinear inversion equation, and after correction by the regional productivity coefficient, the final sugar beet yield prediction result is output.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a standard grayscale texture map that enhances details in dark areas based on near-infrared data and generates an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. It overcomes the limitations of existing technologies that rely solely on spectral intensity, effectively solving the problem of yield estimation distortion caused by spectral saturation during the crop canopy-closing stage. By extracting leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth, this method can keenly capture subtle feature differences in high biomass areas from a spatial distribution perspective, significantly improving the resolution of yield prediction.

[0018] This invention also uses an exponential decay mechanism to couple the leaf fragmentation and local cohesion to generate a canopy wrinkling index, and constructs a nonlinear inversion equation based on the biological allometric growth law. Through this nonlinear regulation and inversion calculation, this scheme can output a sugar beet predicted yield with actual physical dimensions, ensuring the stability and interpretability of the yield estimation results in complex field environments. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a schematic diagram showing the relationship between blade fragmentation and local cohesion. Figure 3 This is a schematic diagram illustrating the relationship between the canopy wrinkling index and predicted yield. Figure 4 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] Example: Please see Figures 1-3 The present invention provides a technical solution: A method for predicting sugar beet yield based on near-infrared remote sensing texture features, comprising the following steps: Step 1: Acquire remote sensing image data of the sugar beet planting area to be monitored, and extract near-infrared band data that characterizes vegetation reflectance as the original intensity input. Perform nonlinear stretching enhancement and discretization processing on the original intensity input to generate a standard grayscale texture map that enhances dark details.

[0023] In this embodiment, a drone equipped with a multispectral camera is used for low-altitude remote sensing operations to acquire remote sensing image data of the sugar beet planting area to be monitored. Compared with traditional manual field excavation sampling, which is highly destructive and can only obtain sparse point data, remote sensing image data is non-destructive, high-throughput, and has full coverage. It can record the spectral response of the crop canopy at different wavelengths, thereby providing data support for analyzing the spatial heterogeneity of yield caused by differences in soil fertility and moisture in large areas of farmland.

[0024] After acquiring the image, since the camera's raw digital quantization value only represents the light intensity value sensed by the sensor, it is affected by the solar altitude angle, weather conditions and camera gain at the time of shooting and does not have physical consistency. Therefore, this solution does not directly process the camera's raw digital quantization value, but requires radiometric calibration.

[0025] The logic of converting the original digital quantization value into physical reflectance using preset radiometric calibration parameters is as follows: based on the spectral response characteristics of the sensor in the near-infrared band, the continuous spectral radiation energy within the coverage area of ​​this band is weighted and accumulated, that is, according to the sensor's sensitivity weights to different wavelengths, the continuously changing spectral energy is converted into the original digital quantization value that characterizes the average energy intensity of this band.

[0026] Subsequently, linear operational logic is used to correct and calculate the original digital quantization value to obtain the equivalent physical reflectivity at the center wavelength of this band. The specific calculation formula is as follows: In the formula, The coordinates obtained by the solution are Physical reflectivity at that location The preset radiation calibration gain coefficient is used to characterize the photoelectric conversion responsivity of the sensor. The preset radiation calibration bias coefficient is used to characterize the zero-point drift correction of the system. Indicates coordinates as The original digital quantization value at that location.

[0027] In this formula To calculate physical reflectance, which specifically characterizes the inherent spectral reflectance properties of sugar beet canopy in the near-infrared band, unlike the raw numerical quantization value that varies with illumination, physical reflectance is an intrinsic property of the material. This method eliminates the interference of ambient light variations, ensuring that canopy values ​​under the same growth condition remain physically consistent across different weather conditions.

[0028] Among them, the original digital quantization value This is a direct product of the sensor's photoelectric conversion and is significantly positively correlated with the output result. Radiation calibration gain coefficient. This determines the linear slope of the photoelectric conversion, used to correct for sensitivity differences between different sensors and accurately map digital counts to physical dimensions. Radiation calibration bias coefficient. This is additively correlated with the output and is mainly used to subtract zero-point drift caused by dark current and background noise. Through this step, the effects of sensor hardware differences and ambient light fluctuations are effectively eliminated, constructing a physically comparable single-channel original physical reflectivity matrix.

[0029] In this embodiment, near-infrared band data is extracted directionally as the sole intensity input for constructing the texture map, instead of using the traditional visible light band. The technical basis for this mainly includes two aspects: Firstly, biological response characteristics: the enlargement of beet tubers is often accompanied by vigorous growth of the above-ground leaves. Near-infrared band (approximately...) It is extremely sensitive to the cellular cavity structure within healthy vegetation sponge tissue, and its reflectivity is much higher than that of the visible light band, enabling it to more sensitively reflect changes in biomass. Secondly, the ability to preserve texture details: This scheme aims to extract leaf fragmentation and local cohesion. During the canopy-closing stage of sugar beets, the mutual compression and shading of leaves create numerous shadow areas. In visible light images, these shadows often appear as indistinguishable black dead zones, while in the near-infrared band, these shadow areas still retain strong radiant energy and rich grayscale levels. Therefore, using the near-infrared band can preserve the geometric undulations and wrinkle details of the canopy surface to the greatest extent, preventing the loss of texture features in dark areas.

[0030] In this embodiment, the specific steps for constructing a standard grayscale texture image include: Near-infrared band data characterizing the internal structure of vegetation are extracted from the acquired remote sensing image data and used as the raw digital quantization value. The raw digital quantization value is converted into physical reflectance using preset radiometric calibration parameters, while keeping the spatial position of the pixels unchanged, and a single-channel raw physical reflectance matrix is ​​constructed.

[0031] After obtaining the original physical reflectance matrix, direct texture analysis often results in compression of dark details due to the extremely large dynamic range between high-reflectance leaf surfaces and low-reflectance shadows. Therefore, the matrix is ​​iterated through, and the global maximum physical reflectance value for the current scene is calculated. Using this global maximum physical reflectance value as a normalization benchmark, combined with a preset contrast enhancement coefficient and the total number of gray levels, the original physical reflectance matrix is ​​subjected to logarithmic nonlinear stretching and discrete quantization mapping to output a standard grayscale texture image. The specific calculation formula is as follows: In the formula, Represents the coordinates in the standard grayscale texture map. pixel grayscale levels, and These are the x and y indexes of the image pixels, respectively. This is a contrast enhancement coefficient used to adjust the curvature of the nonlinear mapping. In the original physical reflectivity matrix, the coordinates are... Physical reflectivity at that location This represents the global maximum physical reflectance value statistically analyzed from the current original physical reflectance matrix. This represents the total number of gray levels.

[0032] The total number of gray levels in this embodiment The value is set to 256 for quantization accuracy, ensuring that subtle texture differences in the beet canopy are captured while preventing statistical distortion caused by an excessively large dimension of the omnidirectional gray-level co-occurrence probability matrix.

[0033] In this formula Essentially, it is a relative height index of canopy texture after nonlinear enhancement. It no longer simply represents the intensity of light, but transforms the light and dark variations on the canopy surface caused by overlapping leaves into two-dimensional grayscale levels. This makes the originally blurry leaf gaps and depth shadows present clear layers, ensuring that the leaf fragmentation can truly reflect the geometric complexity of the canopy.

[0034] in, Although it is positively correlated with the output, but after After normalization and placement within the natural logarithm function, a non-linear diminishing effect is observed: that is, in the low reflectivity shadow region, a small increase in reflectivity leads to a significant increase in gray level, thereby directionally amplifying the details in the shadow area. The contrast enhancement factor determines the mapping curvature. The larger the value, the steeper the slope in the low-value area, and the stronger the stretching of shadow contrast. And... The quantization precision is determined by the value; a larger value results in richer texture levels. This formula converts continuous physical reflectance into discrete integer gray levels, effectively solving the problem of compressing dark details under high dynamic range conditions.

[0035] Step 2: Based on the obtained standard grayscale texture image, define several scanning displacement vectors in several directions. Using the spatial co-occurrence constraint, traverse all pixels in the image and their neighboring pixels. For all pairs of spatial pixels to be judged, which are composed of the current center pixel and the neighboring pixels pointed to by the displacement vector in all scanning directions, count, sum and normalize them to construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences.

[0036] In this embodiment, after acquiring the standard grayscale texture image, in order to capture the spatial correlation between pixels, the system first defines several scanning displacement vectors in several directions. Since the sugar beet canopy texture is a complex pattern composed of stacked leaves, vein orientation, and row arrangement, the grayscale value of a single pixel cannot reflect this structural relationship, and it is necessary to examine the joint distribution characteristics of the pixel and its neighboring pixels.

[0037] Therefore, this scheme uses spatial co-occurrence constraints to traverse all pixels in the image. Based on the standard grayscale texture map, it sets several scanning directions covering the two-dimensional plane and their corresponding displacement vectors. For each pixel coordinate in the image, it defines it as a reference center pixel and locks the corresponding target neighbor pixel according to the displacement vector in the current scanning direction. The reference center pixel and the target neighbor pixel together constitute a spatial pixel pair to be judged.

[0038] For any given scanning direction and gray level combination, this scheme utilizes spatial co-occurrence constraints to detect whether the spatial pixel pair to be judged matches a preset specific gray level combination, and outputs a discrimination signal accordingly. The calculation formula for the discrimination signal is as follows: In the formula, To distinguish the signal, it is indicated that in the standard grayscale texture image, at the th Coordinates in each scanning direction The pixel grayscale value at that location is equal to And its neighboring pixel grayscale value is equal to The matching status, Index for the scan direction. This refers to the specific combination of gray levels that the current omnidirectional gray-level co-occurrence probability matrix is ​​statistically analyzing. Represented as the first Spatial displacement vectors in the scanning direction.

[0039] Among them, the scan direction index Index variables used to discretize the geometric orientation of the texture (values ​​range from 1 to...) This corresponds to a specific angle from which the system observes the canopy texture. In this embodiment... The value is 4, and the four angles are respectively , , , By traversing this index, the continuity or breakage characteristics of the blade texture in the horizontal, vertical, and diagonal directions are scanned from all angles, preventing information omissions caused by a single viewpoint. Displacement vector Then the first is defined The relative offsets in each direction based on the Cartesian coordinate system, the component ratios of which determine the spatial azimuth angle of the neighboring pixels relative to the center pixel, and their modulus (Euclidean distance) determine the spatial scale of texture observation.

[0040] in, As a binary indicator function, it plays a role in precise filtering when When, it indicates the coordinates along the direction Accurately captures the grayscale level Jump to grayscale Texture structure (such as from the high-gloss leaves) Transition to dark shadows (edge); when When this condition is met, it means that such a structure does not exist at that location. This mechanism can filter out background noise and irrelevant flat area textures, ensuring the purity and signal-to-noise ratio of subsequent feature extraction.

[0041] In this embodiment, sugar beet cultivation typically exhibits a distinct ridge structure, resulting in strong directionality (anisotropy) in texture features. If only pixel pairs in a single direction are counted, the extracted features will be unstable due to the uncertainty of the angle between the UAV's flight path and the ridge direction. Therefore, this solution eliminates the observation bias caused by a single direction by summing the matching results across all scanning directions.

[0042] In this embodiment, the discrimination signal is used to sum all spatial coordinates and scanning directions of the standard grayscale texture image to statistically analyze specific grayscale level combinations. The occurrence frequency is used, and the cumulative total number of all valid undiscriminated spatial pixel pairs constructed during the traversal process is used as the normalization benchmark. The occurrence frequency is divided by the cumulative total number to calculate the omnidirectional gray-level co-occurrence probability value. For gray-level combinations not detected during the traversal process, their corresponding omnidirectional gray-level co-occurrence probability values ​​are directly set to zero, thereby constructing a numerically complete omnidirectional gray-level co-occurrence probability matrix that eliminates the influence of image size and orientation number. The formula is as follows: In the formula, Represents the 1st gray-level co-occurrence probability matrix in the omnidirectional gray-level co-occurrence probability matrix. Okay, number The omnidirectional gray-scale co-occurrence probability value of the column. This represents the cumulative total number of undiscriminated spatial pixel pairs across all scanning directions in the standard grayscale texture image. The number of scanning directions covering the two-dimensional plane.

[0043] in, This reflects a specific combination of light and dark textures (grayscale) on the surface of the beet canopy in any direction. and The joint probability density of the eigenvalues ​​is calculated and normalized so that the eigenvalues ​​are independent of the size of the monitored area.

[0044] Step 3: Based on the omnidirectional gray-level co-occurrence probability matrix, feature vectors describing the sugar beet canopy structure are extracted from two dimensions: texture intensity and spatial distribution. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth.

[0045] In this embodiment, after obtaining the omnidirectional gray-level co-occurrence probability matrix that can objectively describe the probability of texture distribution, this scheme does not use general texture parameters (such as entropy or correlation), but instead constructs leaf fragmentation and local cohesion separately for the growth characteristics of sugar beet during the canopy closure period. During the critical period of sugar beet yield formation, the complexity of the canopy structure is directly related to the expansion rate of underground tubers: on the one hand, the more intense the leaf interlacing (high fragmentation), the larger the photosynthetic area and the more vigorous the growth; on the other hand, the more compact the leaf cluster (strong cohesion), the healthier the canopy structure and the stronger the resistance to lodging and disease.

[0046] In this embodiment, the step of obtaining the blade breakage degree is as follows: based on the omnidirectional gray-level co-occurrence probability matrix, the gray-level difference value is nonlinearly amplified using cubic weighting to extract the blade breakage degree. The specific calculation is based on the following formula: In the formula, Leaf fragmentation is used to characterize the degree of overlap at the leaf edges. and Here, represents the row and column indices of the omnidirectional gray-level co-occurrence probability matrix, which physically represent the gray-level values ​​of the two pixels constituting a pixel pair. It is the absolute value of the difference between gray levels. The total number of gray levels.

[0047] Among them, blade fragmentation This reflects the degree of overlap at the leaf edges. Healthy leaves have high reflectivity (large grayscale value), while the shadows formed by overlapping leaves have low reflectivity (small grayscale value). and The greater the difference, the more obvious the leaf edge or pores are in the field of view. This formula adopts a high-order power design, which has a significant signal amplification effect: it assigns extremely high weights to pixel pairs with large gray level differences (i.e., strong light-dark boundaries), while almost ignoring flat areas with small gray level differences (such as simple soil or continuous leaf surfaces).

[0048] In this embodiment, the steps for obtaining local cohesion are as follows: statistically analyze the gray-level expected value of the omnidirectional gray-level co-occurrence probability matrix, establish the average brightness benchmark of the image, use inverse variance weighted logic to iterate and sum the omnidirectional gray-level co-occurrence probability matrix, and quantify the density of texture distribution by giving higher weights to gray-level combinations that are close to the average brightness benchmark, thereby extracting local cohesion.

[0049] The formula for calculating the expected gray level is: In the formula, To determine the expected gray level of the brightness mean baseline, this variable reflects the weighted average gray level of the texture image within the current monitoring area.

[0050] The formula for calculating the local cohesive force is: In the formula, Local cohesion force, which characterizes the compactness of canopy growth, This is a sensitivity adjustment coefficient used to adjust the magnitude of weight decay. and Representing reference pixels respectively The difference between the grayscale expected value and the neighboring pixels With grayscale expectation The difference.

[0051] Local cohesion This reflects the compactness of canopy growth. A higher value indicates that the beet canopy is compact, evenly distributed, and free of obvious gaps or lesions, reflecting the robustness of the population structure; while a lower value indicates a more robust structure. The value indicates that the canopy may be sparse, lodged, or unevenly growing.

[0052] Table 1 shows the feature vectors extracted from the two dimensions of texture intensity and spatial distribution to describe the structure of the beet canopy.

[0053] Table 1: Feature Vector Table of Beet Canopy Structure The data in Table 1 show that the bare soil area, due to its flat and uniform surface, exhibits background noise characteristics of low leaf fragmentation and local cohesion. Although the weed-affected area has high leaf fragmentation, it lacks a row-ridge structure due to disordered growth, exhibiting discrete characteristics of high leaf fragmentation and low local cohesion. This effectively explains the technical defect of traditional single indicators that easily misclassify weeds as crops. The high-yield sugar beet area exhibits a unique dual-high characteristic, namely, the coexistence of high leaf fragmentation and high local cohesion, proving that only by conducting specific combination analysis of leaf fragmentation and local cohesion can the effective crop biomass be accurately identified.

[0054] Step 4: Couple the blade fragmentation and local cohesion to generate a canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion.

[0055] In this embodiment, single-dimensional texture features often have ambiguity in physical indices. In order to overcome the limitations of single features, this solution constructs a canopy wrinkling index, which aims to accurately locate the effective biomass texture of the beet canopy through the physical coupling of multi-dimensional features.

[0056] The specific steps for generating the canopy wrinkling index are as follows: A basic texture energy term is constructed by coupling leaf fragmentation and local cohesion using the geometric mean method. A nonlinear adjustment coefficient is then constructed using an exponential decay method. This basic texture energy term is weighted and corrected to generate the canopy wrinkling index. The calculation formula is as follows: In the formula, The canopy wrinkling index is used to comprehensively characterize the complexity of canopy texture. This is the basic texture energy term constructed based on the geometric mean method. The degree of breakage of the blade. For the local cohesive force, To balance the order-of-magnitude difference between blade fragmentation and local agglomeration force, the sensitivity adjustment coefficient of the suppression force is adjusted. To prevent numerical stability constants with a denominator of zero, where .

[0057] Among them, the canopy wrinkling index, through coupled calculation, achieved a significant background noise suppression effect. This means that the base value of this term will only increase when a region simultaneously possesses high fragmentation (vigorous growth) and high cohesion (compact structure), thus filtering out invalid regions with extremely high values ​​for one characteristic but extremely low values ​​for another. In low biomass regions (such as bare soil), the index value is forcibly lowered to near zero; while in high biomass regions, the index value exhibits a highly sensitive linear response, ensuring accurate differentiation of differences in sugar beet yield.

[0058] Step 5: Construct a nonlinear inversion equation, input the canopy wrinkling index into the nonlinear inversion equation, and after correction by the regional productivity coefficient, output the final beet yield prediction result.

[0059] In this embodiment, the sugar beet yield prediction step is as follows: A nonlinear inversion equation is constructed based on the biological allometric growth law, which characterizes the nonlinear power function correlation between the predicted sugar beet yield and the canopy wrinkling index. Preset regional feature parameters are retrieved, and the canopy wrinkling index is substituted into the nonlinear inversion equation to output the predicted sugar beet yield. The calculation formula is as follows: In the formula, To obtain the final predicted output, This serves as the baseline production volume for correcting regional production capacity figures. The biomass conversion coefficient, in areas with better conditions ( (With larger values), the same canopy structure often corresponds to a higher underground block conversion rate. The canopy wrinkling index is the index of the canopy. The growth allotropy index is used to characterize the degree of nonlinear response of yield to changes in canopy texture.

[0060] Table 2 shows the nonlinear inversion prediction table of sugar beet yield, which maps the canopy wrinkling index to specific physical yield using a nonlinear inversion equation.

[0061] Table 2: Nonlinear Inversion Prediction Table for Sugar Beet Yield As shown in Table 2, the canopy wrinkling index increased significantly from 4.34 to 9.31 in high-yield areas, driving the predicted yield up from 25.37 to 55.91. This result is highly consistent with the biological allometric growth pattern, confirming that the compactness of the aboveground canopy is positively correlated with the accumulation rate of underground tubers.

[0062] Among them, by collecting a large amount of synchronous remote sensing image data and actual ground harvest yield data in specific medium and high yield sugar beet planting areas, a paired sample set was constructed. The nonlinear least squares method was used to conduct iterative regression analysis with the goal of minimizing the root mean square error between theoretical yield and measured yield. The optimal solutions for biomass conversion coefficient, growth allometric index and basic yield benchmark were determined, so that they can accurately characterize the local water and fertilizer conversion efficiency, nonlinear growth law during the canopy closure period and the field background biomass, respectively.

[0063] In this embodiment, the biomass conversion coefficient is set to 3.5 to construct an amplification mapping mechanism for environmental productivity. That is, by quantifying local water, fertilizer and climate potential, the dimensionless texture index is accurately mapped to physical yield that conforms to the characteristics of medium- and high-yield areas. At the same time, the growth allometric index is set to 1.2 to construct a power function curve with increasing marginal benefits, thereby accurately capturing the accelerated yield increase characteristics caused by maximizing light energy utilization during the canopy closure period, and significantly widening the yield prediction gradient between superior and inferior canopies. In addition, the baseline yield is set to 5, which not only conforms to the objective fact that there is stubble or root biomass in the field, but also effectively prevents numerical cliffs or negative value oscillations caused by extreme texture loss.

[0064] Please see Figure 4 The present invention also provides a sugar beet yield prediction system based on near-infrared remote sensing texture features. This system is used to implement the aforementioned sugar beet yield prediction method based on near-infrared remote sensing texture features, and includes: The image preprocessing module is used to acquire remote sensing image data of the sugar beet planting area to be monitored, extract the near-infrared band data that characterizes the vegetation reflectance as the original intensity input, perform nonlinear stretching enhancement and discretization processing on the original intensity input, and generate a standard grayscale texture map that enhances the details in the dark areas. The matrix construction module is used to define several scanning displacement vectors based on the obtained standard grayscale texture image, traverse all pixels in the image and their neighboring pixels using spatial co-occurrence constraints, count, sum and normalize the pairs of spatial pixels to be judged in all scanning directions consisting of the current center pixel and the neighboring pixels pointed to by the displacement vector, and construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. The feature extraction module is used to extract feature vectors describing the sugar beet canopy structure from two dimensions: texture intensity and spatial distribution, based on the omnidirectional gray-level co-occurrence probability matrix. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth. The exponential calculation module is used to couple the blade fragmentation and local cohesion to generate the canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion. The yield inversion module is used to construct a nonlinear inversion equation. The canopy wrinkling index is input into the nonlinear inversion equation, and after correction by the regional productivity coefficient, the final sugar beet yield prediction result is output.

[0065] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0066] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for predicting sugar beet yield based on near-infrared remote sensing texture features, characterized in that the steps include... include: Step 1: Acquire remote sensing image data of the sugar beet planting area to be monitored, extract near-infrared band data that characterizes vegetation reflectance as the original intensity input, perform nonlinear stretching enhancement and discretization processing on the original intensity input, and generate a standard grayscale texture map that enhances dark details. Step 2: Based on the obtained standard grayscale texture map, define several scanning displacement vectors in several directions. Using the spatial co-occurrence constraint, traverse all pixels in the image and their neighboring pixels. For all pairs of spatial pixels to be judged, which are composed of the current center pixel and the neighboring pixels pointed to by the displacement vector in all scanning directions, count, sum and normalize them to construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. Step 3: Based on the omnidirectional gray-level co-occurrence probability matrix, feature vectors describing the sugar beet canopy structure are extracted from the two dimensions of texture intensity and spatial distribution. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth. Step 4: Couple the blade fragmentation and local cohesion to generate a canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion. Step 5: Construct a nonlinear inversion equation, input the canopy wrinkling index into the nonlinear inversion equation, and after correction by the regional productivity coefficient, output the final beet yield prediction result.

2. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 1, characterized in that: The specific steps for constructing a standard grayscale texture image include: Near-infrared band data characterizing the internal structure of vegetation are extracted from the acquired remote sensing image data and used as the raw digital quantization value. The raw digital quantization value is converted into physical reflectance using preset radiometric calibration parameters while keeping the pixel spatial position unchanged. A single-channel raw physical reflectance matrix is ​​constructed, and the matrix is ​​traversed to calculate the global maximum physical reflectance value in the current scene. Using the global maximum physical reflectance value as a normalization benchmark, and combining it with a preset contrast enhancement coefficient and the total number of gray levels, the original physical reflectance matrix is ​​subjected to logarithmic nonlinear stretching and discrete quantization mapping to output a standard grayscale texture image. The specific calculation formula is as follows: In the formula, Represents the coordinates in the standard grayscale texture map. pixel grayscale levels, and These are the x and y indexes of the image pixels, respectively. This is a contrast enhancement coefficient used to adjust the curvature of the nonlinear mapping. In the original physical reflectivity matrix, the coordinates are... Physical reflectivity at that location This represents the global maximum physical reflectance value statistically analyzed from the current original physical reflectance matrix. This represents the total number of gray levels.

3. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 2, characterized in that: The logic for converting the raw digital quantization value into physical reflectance using preset radiometric calibration parameters is as follows: Based on the spectral response characteristics of the sensor in the near-infrared band, the continuous spectral radiation energy within the coverage area of ​​this band is weighted and accumulated. That is, according to the sensor's sensitivity weights to different wavelengths, the continuously changing spectral energy is converted into a raw digital quantization value that characterizes the average energy intensity of this band. The original digital quantization value is corrected and calculated using linear operational logic to obtain the equivalent physical reflectivity at the center wavelength of this band. The specific calculation formula is as follows: In the formula, The coordinates obtained by the solution are Physical reflectivity at that location The preset radiation calibration gain coefficient is used to characterize the photoelectric conversion responsivity of the sensor. The preset radiation calibration bias coefficient is used to characterize the zero-point drift correction of the system. Indicates coordinates as The original digital quantization value at that location.

4. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 2, characterized in that: The specific steps for traversing all pixels and their neighboring pixels in the image using spatial co-occurrence constraints include: Based on the standard grayscale texture image, several scanning directions covering the two-dimensional plane and their corresponding displacement vectors are set. For each pixel coordinate in the image, it is defined as a reference center pixel, and the corresponding target neighbor pixel is locked according to the displacement vector in the current scanning direction. The reference center pixel and the target neighbor pixel together constitute a spatial pixel pair to be judged. The spatial co-occurrence constraint condition is used to detect whether the spatial pixel pair to be judged matches a preset specific gray level combination, and a judgment signal is output accordingly. The calculation formula of the judgment signal is as follows: In the formula, To distinguish the signal, it is indicated that in the standard grayscale texture image, at the th Coordinates in each scanning direction The pixel grayscale value at that location is equal to And its neighboring pixel grayscale value is equal to The matching status, Index for the scan direction. This refers to the specific combination of gray levels that the current omnidirectional gray-level co-occurrence probability matrix is ​​statistically analyzing. Represented as the first Spatial displacement vectors in the scanning direction.

5. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 4, characterized in that: The steps for constructing an omnidirectional gray-level co-occurrence probability matrix that eliminates directional differences include: The discrimination signal is used to sum all spatial coordinates and scanning directions of the standard grayscale texture image to statistically analyze specific grayscale level combinations. The occurrence frequency is used as the normalization benchmark. The cumulative total number of all valid undiscriminated spatial pixel pairs constructed during the traversal process is used as the normalization benchmark. The occurrence frequency is divided by the cumulative total number to calculate the omnidirectional gray-level co-occurrence probability value. For gray-level combinations that are not detected during the traversal process, their corresponding omnidirectional gray-level co-occurrence probability values ​​are directly set to zero. This constructs an omnidirectional gray-level co-occurrence probability matrix that is numerically complete and eliminates the influence of image size and orientation. The formula is as follows: In the formula, Represents the 1st gray-level co-occurrence probability matrix in the omnidirectional gray-level co-occurrence probability matrix. Okay, number The omnidirectional gray-scale co-occurrence probability value of the column. This represents the cumulative total number of undiscriminated spatial pixel pairs across all scanning directions in the standard grayscale texture image. The number of scanning directions covering the two-dimensional plane.

6. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 5, characterized in that: The steps for obtaining the blade breakage degree are as follows: Based on the aforementioned omnidirectional gray-level co-occurrence probability matrix, the gray-level difference value is nonlinearly amplified using cubic weighting to extract the blade breakage degree. The specific calculation is based on the following formula: In the formula, Leaf fragmentation is used to characterize the degree of overlap at the leaf edges. and Here, represents the row and column indices of the omnidirectional gray-level co-occurrence probability matrix, which physically represent the gray-level values ​​of the two pixels constituting a pixel pair. It is the absolute value of the difference between gray levels. The total number of gray levels; The steps for obtaining the local cohesion are as follows: statistically analyze the gray-level expectation value of the omnidirectional gray-level co-occurrence probability matrix, establish the average brightness benchmark of the image, use inverse variance weighted logic to iterate and sum the omnidirectional gray-level co-occurrence probability matrix, and quantify the density of texture distribution by giving higher weights to gray-level combinations that are close to the average brightness benchmark, thereby extracting the local cohesion.

7. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 6, characterized in that: The formula for calculating the expected gray level is: In the formula, To determine the expected grayscale value of the brightness mean baseline; The formula for calculating the local cohesive force is: In the formula, Local cohesion force, which characterizes the compactness of canopy growth, This is a sensitivity adjustment coefficient used to adjust the magnitude of weight decay. and Representing reference pixels respectively The difference between the grayscale expected value and the neighboring pixels With grayscale expectation The difference.

8. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 7, characterized in that: The specific steps for generating the canopy wrinkle index are as follows: A basic texture energy term is constructed by coupling leaf fragmentation and local cohesion using the geometric mean method. A nonlinear adjustment coefficient is then constructed using an exponential decay method. This basic texture energy term is weighted and corrected to generate the canopy wrinkling index. The calculation formula is as follows: In the formula, The canopy wrinkling index is used to comprehensively characterize the complexity of canopy texture. This is the basic texture energy term constructed based on the geometric mean method. The degree of breakage of the blade. For the local cohesive force, To balance the order-of-magnitude difference between blade fragmentation and local agglomeration force, the sensitivity adjustment coefficient of the suppression force is adjusted. To prevent numerical stability constants with a denominator of zero.

9. The method for predicting sugar beet yield based on near-infrared remote sensing texture features according to claim 8, characterized in that: The steps for predicting beet yield are as follows: A nonlinear inversion equation is constructed based on the biological allometric growth law, which characterizes the nonlinear power function correlation between the predicted yield of sugar beets and the canopy wrinkling index. By retrieving preset regional feature parameters and substituting the canopy wrinkling index into the nonlinear inversion equation, the predicted sugar beet yield is output. The calculation formula is as follows: In the formula, To obtain the final predicted output, This serves as the baseline production volume for correcting regional production capacity figures. Biomass conversion coefficient The canopy wrinkling index is the index of the canopy. The growth allotropy index is used to characterize the degree of nonlinear response of yield to changes in canopy texture.

10. A sugar beet yield prediction system based on near-infrared remote sensing texture features, characterized in that: The sugar beet yield prediction system based on near-infrared remote sensing texture features is used to implement the sugar beet yield prediction method based on near-infrared remote sensing texture features as described in any one of claims 1-9, comprising: The image preprocessing module is used to acquire remote sensing image data of the sugar beet planting area to be monitored, extract the near-infrared band data that characterizes the vegetation reflectance as the original intensity input, perform nonlinear stretching enhancement and discretization processing on the original intensity input, and generate a standard grayscale texture map that enhances the details in the dark areas. The matrix construction module is used to define several scanning displacement vectors based on the obtained standard grayscale texture image, traverse all pixels in the image and their neighboring pixels using spatial co-occurrence constraints, count, sum and normalize the pairs of spatial pixels to be judged in all scanning directions consisting of the current center pixel and the neighboring pixels pointed to by the displacement vector, and construct an omnidirectional grayscale co-occurrence probability matrix that eliminates directional differences. The feature extraction module is used to extract feature vectors describing the sugar beet canopy structure from two dimensions: texture intensity and spatial distribution, based on the omnidirectional gray-level co-occurrence probability matrix. The feature vectors include: leaf fragmentation, which characterizes the intensity of leaf edge overlap, and local cohesion, which characterizes the compactness of canopy growth. The exponential calculation module is used to couple the blade fragmentation and local cohesion to generate the canopy wrinkling index. The coupled calculation uses an exponential decay mechanism to nonlinearly control the geometric characteristics of the blade fragmentation and local cohesion. The yield inversion module is used to construct a nonlinear inversion equation. The canopy wrinkling index is input into the nonlinear inversion equation, and after correction by the regional productivity coefficient, the final sugar beet yield prediction result is output.

Images