A method for distinguishing wheat seedling in wintering period based on remote sensing image data

Abstract

The present application relates to the field of wheat image analysis, and specifically discloses a kind of wintering period wheat seedling distinguishing method based on remote sensing image data, for solving the existing wintering period winter wheat seedling identification method more depends on artificial investigation or single color, vegetation index threshold, difficult to fuse multidimensional key parameters, and easily be disturbed by shadow, bare soil and frost residue, leading to strong and weak seedling recognition unstable, partition is not continuous, risk positioning is not accurate Problem, including sample plot layout and remote sensing data acquisition, remote sensing image radiation geometry correction and pixel level registration, vegetation mask extraction and single plant object segmentation, single plant multidimensional characterization extraction and strong seedling potential value construction and sample plot statistical classification and global extrapolation output;The present application constructs seedling potential value and classifies extrapolation by multi-source remote sensing single plant segmentation and C, V, H, L, B fusion, and stably outputs global strong seedling partition and vigorous seedling partition, weak seedling risk area.

Description

Technical Field

[0001] This invention relates to the field of wheat image technology, and in particular to a method for distinguishing robust wheat seedlings during the wintering period based on remote sensing image data. Background Technology

[0002] The robustness and evenness of winter wheat seedlings during the overwintering period are key factors affecting the safe overwintering of winter wheat and its final yield and quality. A robust and even population can fully utilize resources such as light, heat, water, and fertilizer, reducing ineffective competition among individual winter wheat plants. Controlling excessive growth and promoting weak seedlings can enhance the cold resistance and stress tolerance of winter wheat, enabling it to maintain continuous and stable growth during the overwintering stage, laying the foundation for subsequent tillering, ear formation, and grain filling. Therefore, accurately and objectively distinguishing the spatial distribution of vigorous, strong, and weak winter wheat seedlings during the overwintering period, and quantifying the differences in seedling quality, is an important aspect of modern wheat planting management and precision agronomic control. Currently, to improve detection and diagnostic efficiency, existing technologies use UAV RGB images or multi-spectral images to assess winter wheat seedling conditions, classifying them through single color features or single vegetation index thresholds, or using empirical regression models built from a small number of samples for inference. However, the seedling condition of winter wheat during the overwintering period is multidimensional, and relying on a single indicator can easily lead to problems. For example, seedlings with darker leaf color but insufficient aboveground biomass may be misjudged as robust seedlings; plots with taller plants but lower leaf area index and loose canopy structure may be misjudged as growing well; and plots with similar cover but significant differences in biomass and canopy thickness may be difficult to distinguish effectively. At the same time, during the overwintering period, factors such as soil exposure, changes in light intensity, shading, frost residue, varietal differences, and uneven distribution of water and fertilizer can cause the seedling condition classification boundary of a single characteristic or index method to drift, resulting in poor stability of identification results and weak cross-plot generalization ability. This can lead to inaccurate zoning for topdressing, inappropriate measures to promote weak growth and control excessive growth, and delayed identification of overwintering risks, thus affecting the accuracy of seedling condition assessment during the overwintering period and the effectiveness of subsequent precision management. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method for identifying strong and weak winter wheat seedlings during the overwintering period based on remote sensing image data. By using pixel-level registration of multi-source images from UAV remote sensing, single-plant segmentation under vegetation mask constraints, and multi-dimensional characterization of leaf color and index, seedling height, leaf area index, and biomass, a strong seedling potential value is constructed and extrapolated by quadrat classification. This method achieves objective and stable identification of strong and weak winter wheat seedlings during the overwintering period and continuous regional output across the entire area, thereby improving the accuracy of judging the seedling condition of winter wheat during the overwintering period.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying robust wheat seedlings during the wintering period based on remote sensing image data includes: In the target planting area, quadrats were identified and UAV remote sensing data covering the quadrats were obtained. The remote sensing image data included RGB visible light images, multispectral images and their imaging parameters. The multispectral images included red light band images and near-infrared band images. Perform radiometric correction, geometric correction, orthophoto stitching, and pixel-level registration on remote sensing image data; Based on the registered remote sensing image data, winter wheat vegetation mask was extracted and individual plant objects were obtained by segmentation within the quadrat. For each individual plant, leaf color characteristic C, vegetation index characteristic V, seedling height characteristic H, leaf area index characteristic L, and aboveground biological characteristic B are extracted from remote sensing image data. C, V, H, L, and B are standardized and fused to obtain the seedling vigor value P. The mean μ and standard deviation σ of the seedling vigor value P within the quadrat are calculated. Based on μ and σ, individual plants are classified into vigorous, strong, and weak seedlings, and a quadrat grade distribution map is generated. The results of each quadrat are extrapolated to the target planting area. Specifically, under the constraint of winter wheat vegetation masking, the quadrat grade results are used as seeds to perform geodesic dilatation morphological reconstruction extrapolation on the grade areas, and overlapping propagation areas are adjudicated according to the principle of minimum morphological distance. Finally, after purification by opening and closing operations, a spatial distribution map of the grade of the target area is generated, and the spatial distribution map of the winter wheat seedling condition grade and the risk area of ​​weak seedling aggregation are output.

[0005] It should be noted that this invention obtains P by fusing multidimensional characterization of a single plant. The morphological reconstruction extrapolation and distance adjudication purification of the quadrat grading results under vegetation mask constraints enable continuous and stable propagation of the robust seedling grade from the quadrat to the entire area. This suppresses grading breaks and misclassifications caused by shadows, bare soil, and boundary noise, and accurately and usably outputs the overwintering vigorous seedling zones, robust seedling zones, and weak seedling cluster risk zones to support zoning management.

[0006] As a further aspect of the present invention, in the spatial distribution map of quadrat grades, the grading criteria for vigorous seedlings, strong seedlings, and weak seedlings are as follows: when P ≥ μ + σ, the seedlings are graded as vigorous seedlings; when μ - σ < P < μ + σ, the seedlings are graded as strong seedlings; and when P ≤ μ - σ, the seedlings are graded as weak seedlings.

[0007] It should be noted that the purpose of identifying vigorous, strong, and weak seedlings within the quadrat in this invention is to establish quantifiable seedling condition level labels and strength / weakness comparison benchmarks within the quadrat based on the statistical distribution of the individual seedling vigor value P. This, in turn, generates a quadrat level distribution map and assists in the subsequent calculation of S. R S I On the one hand, the diagnostic indicators for strong seedlings are used, and on the other hand, the results of this grade are used as seed inputs for subsequent morphological reconstruction extrapolation and the determination of risk areas for weak seedling aggregation, thereby transforming individual plant differences into spatial decision-making information that can be used for whole-area zonal management and precise disposal.

[0008] As a further embodiment of the present invention, the seedling potential value P of a single plant is calculated as follows: C, V, H, L, and B are normalized to obtain C′, V′, H′, L′, and B′, respectively, and then calculated according to P=W1·C′+W2·V′+W3·H′+W4·L′+W5·B′, where W1 to W5 are preset non-negative weights, and satisfy W1+W2+W3+W4+W5=1.

[0009] It should be noted that the five single-plant characterization quantities with different dimensions and inconsistent value ranges—leaf color, index, seedling height, leaf area index, and biomass—are first normalized to a unified scale. Then, an interpretable weighted fusion is performed using non-negative weights that sum to 1 to construct a single single-plant seedling potential value P. This value is used to achieve comparable ranking and grading threshold determination of growth potential among different single plants within a quadrat. At the same time, the contribution source of P is controllable and calibrated, and it is easy to maintain the consistency of evaluation criteria in different plots and under different imaging conditions.

[0010] As a further aspect of this invention, before outputting the grade distribution map, spatial consistency constraint correction is performed. Specifically, the crop row angle θ is estimated based on the wheat vegetation mask, and the center coordinates of individual plants are rotated to the row coordinate system. The row coordinate system is a two-dimensional orthogonal coordinate system (u,v) established with the crop row angle θ as the reference, where the u-axis is along the crop row direction and the v-axis is along the direction perpendicular to the crop row direction. An adjacency graph of individual plants is constructed in the row zone according to the row adjacency relationship. The same-zone nearest neighbor consistency penalty and isolated weak seedling and isolated strong seedling removal rules are applied to the grade results of individual plants. The grade of individual plants is corrected according to the cost minimization result in the adjacency graph. After the correction is completed, the number of weak seedlings is counted in the neighborhood of the weak seedling individual plant as the event point set, and the local density of weak seedlings D(x) is obtained. When D(x)≥D0 and the current position of the sample plot meets the judgment condition of the grade III strong seedling sample plot, the current position is marked as the weak seedling aggregation risk area, where r and D0 are preset parameters.

[0011] It should be noted that, in response to the technical problems of isolated misjudgment points within rows, broken partition boundaries, and inaccurate location of weak seedling risk in remote sensing single-plant grading during the overwintering period due to interference from noise, shadows, and bare soil, the spatial continuity and stability of single-plant grading results and the identification of weak seedling cluster risk areas are achieved by minimizing the adjacency consistency cost of row direction constraints and combining it with the determination of local density thresholds for weak seedlings.

[0012] As a further aspect of the present invention, while generating the graded spatial distribution map, the proportion of robust seedlings S is calculated. R With the seedling index S I First, determine the seedlings as vigorous, strong, and weak in different zones. Then, classify the quadrats within the strong seedling zones and output treatment labels. Simultaneously, classify the weak seedling zones and output treatment labels: where S R S is the ratio of the number of robust seedlings in the quadrat to the total number of seedlings. I Press SI =(P z -P r ) / μ calculation, P z P represents the average seedling vigor value P of a single strong seedling. r The average seedling vigor value P of weak seedlings; set the threshold S for judging strong seedlings. R0 S I0 When S R ≥S R0 And S I ≥S I0 At that time, the quadrat is assigned to the robust seedling zone and a grading threshold S is set within the current robust seedling zone. R1 S R2 S I1 S I2 And satisfy S R2 >S R1 ≥S R0 S I2 >S I1 ≥S I0 When S R ≥S R2 And S I ≥S I2 At that time, the quadrat was identified as a Grade I strong seedling quadrat and marked as no nitrogen fertilizer to be applied before winter. When S R ≥S R2 And S I1 ≤S I <S I2 When the sample plot is classified as a Grade II robust seedling sample plot, the treatment label is output as avoiding topdressing with nitrogen fertilizer and implementing compaction or deep cultivation to control excessive growth and conserve moisture. R1 ≤S R <S R2 And S I1 ≤S I <S I2 When the sample plot is classified as a Level III robust seedling sample plot, it is marked as not requiring nitrogen fertilizer and is designated as a key field monitoring sample plot; when S R <S R0 or S I <S I0 The quadrats were then assigned to weak seedling zones, and the mean leaf color characteristic value within each quadrat was used as the basis for classification. Mean value of aboveground biomass Mean value of leaf area index Mean of seedling height Weak seedlings were categorized, and a categorization threshold C was set. W B W L W H W With M D0 C W B WL W H W With M D0 These are the threshold values ​​for weak seedling yellowing, weak seedling low biomass, weak seedling low leaf area index, weak seedling short and tall, and missing and sparse seedlings. M D The ratio of the total number of individual plants in a quadrat to the area of ​​the quadrat is given by the following formula: ≥C W and ≤B W At that time, the seedlings were identified as weak and chlorotic due to nutrient deficiency and were marked for treatment. During the peak tillering period before winter, urea (8-10 kg / mu) was applied in conjunction with irrigation, along with diammonium phosphate (5-8 kg / mu). When M... D ≤M D0 When seedlings are identified as sparse or weak, the corresponding action is to check and replant them. The remaining weak seedling plots are identified as slow-growing weak seedlings, and the corresponding action is to cultivate the soil promptly after winter irrigation or rain.

[0013] It should be noted that, from an image analysis perspective, leaf color, index, canopy height, and texture in overwintering remote sensing images are significantly affected by light and shadow, soil exposure, frost residue, and varietal differences. Distinguishing between strong, medium, and weak seedlings at the individual plant level is prone to local drift and inconsistencies across quadrat diameters. Therefore, this invention first uses S at the quadrat scale... R With S I The criteria for population structure and the significance of differences in strength are quantified into stable statistical standards. First, a rough classification of strong and weak seedling zones is completed to avoid misclassifying areas with overall weakness or insufficient samples into the fine classification of strong seedling zones. Then, within the strong seedling zones, a more stringent S... R1 S R2 S I1 S I2 The classification is carried out, with the classification boundaries constrained by both population proportion and the magnitude of difference. This spatially separates areas of excessive growth from those of normal, robust seedlings, and directly links them to treatment indicators such as controlling excessive growth, not applying fertilizer, and conducting key field inspections. Simultaneously, the mean value of quadrats is further introduced into the weak seedling zones. , , , With M D The classification decouples remotely observable yellowing (chroma), biomass or canopy amount, seedling height (structure), and seedling absence (spatial sparsity) into interpretable categories of weak seedling causes. This reduces misclassification caused by relying on a single index and ensures that each risk category has clear triggering conditions in the image. Under complex imaging interference, it improves the stability and cross-plot consistency of strong and weak seedling zoning and grading, reduces the mismatch between topdressing and growth control strategies caused by misclassification, and elevates weak seedlings from a result label to an executable output of cause classification and treatment suggestions, enabling the precise implementation of management measures such as zoning topdressing, rolling, deep cultivation, and replanting.

[0014] As a further aspect of the present invention, the seedling height characterization quantity H is calculated by the canopy height model, which is a difference grid between the digital surface model and the digital ground model, and the quantile value of the canopy height grid within the coverage area of ​​a single plant object is used as the seedling height characterization quantity H of the single plant object.

[0015] It should be noted that, in order to address the problem that the plant height sampling or pixel extreme values ​​in the overwintering remote sensing images are easily affected by topographic undulations, soil exposure, shadow noise, and local anomalies, resulting in unstable seedling height estimation, the canopy height is obtained by using the difference between the digital surface model (DSM) and the digital terrain model (DTM), and the height quantile value within a single plant area is taken to form H. This achieves robust suppression of outliers and noise, thereby more stably and comparablely representing the height of a single seedling, and improving the accuracy of seedling grading and zoning judgment.

[0016] As a further aspect of the present invention, the leaf area index (L) is obtained by jointly mapping the vegetation index (V) and the canopy coverage (F). The canopy coverage (F) is the percentage of winter wheat vegetation pixels within the winter wheat vegetation mask. The vegetation index (V) includes NDVI, where NDVI = (NIR - Red) / (NIR + Red). NIR is the surface reflectance in the near-infrared band, which is collected by a multispectral camera in the near-infrared channel. Red is the surface reflectance in the red band, which is collected by a multispectral camera in the red channel. The values ​​are calculated as L = a1·NDVI + a2·F + a3, where a1, a2, and a3 are preset calibration coefficients.

[0017] It should be noted that, in order to address the problem that using only NDVI during the overwintering period is easily affected by the proportion of bare soil, the mixed pixels of shadow and low cover, which leads to the distortion of LAI estimation and the difficulty in distinguishing between canopy structures with similar indices, the method of jointly mapping NDVI and canopy cover F and calculating L with calibration coefficients is used to achieve compensation and correction for low cover and mixed background. This allows for a more accurate characterization of canopy leaf volume and growth, and improves the stability of strong seedling grading and weak seedling identification.

[0018] As a further aspect of the present invention, the aboveground biomass characterization quantity B is obtained by jointly mapping the vegetation index characterization quantity V, the canopy coverage quantity F and the seedling height characterization quantity H, and is calculated according to B=b1·NDVI+b2·F+b3·H+b4, where b1, b2, b3 and b4 are preset calibration coefficients.

[0019] It should be noted that, in response to the technical problem that relying solely on NDVI or canopy coverage during the overwintering period is insufficient to reflect differences in canopy thickness and structure, leading to insensitivity of biomass estimation to dense planting, shading, and mixed pixels, and easy misjudgment of strong and weak seedlings, a biomass characterization quantity B is calculated by jointly mapping NDVI, canopy coverage F, and seedling height characterization quantity H. This enables the differentiation of situations with similar indices but different heights or structures, thereby improving the accuracy and robustness of wheat biomass characterization and enhancing the reliability of strong seedling grading, weak seedling zoning, and treatment decisions.

[0020] As a further aspect of the present invention, the leaf color characterization quantity C is obtained as follows: the RGB image is converted to the Lab color space in the registration coordinate system, and color a is extracted. * Channels and Colors b * Channel, based on the brightness statistics of non-vegetated areas within the sample plot, for brightness L * After performing illumination normalization correction on the channel, the color mean and color dispersion within the individual plant area are calculated and fused to obtain the leaf color characterization quantity C, where L * For the dual-anchor point brightness normalization correction channel based on the wheat vegetation reference set and background reference set within the sample plot, a * b is the red-green axis chromaticity channel characterizing the degree of leaf greening after background bias elimination. * To characterize the degree of leaf yellowing after background bias elimination, the yellow-blue axis chromaticity channel was extracted in combination with chromaticity invariants.

[0021] It should be noted that, regarding the issue of unstable leaf color discrimination in overwintering remote sensing RGB images due to the susceptibility of leaf color characteristics to changes in light intensity, shadow occlusion, soil background, and residual frost reflections, a solution was found: the images were converted to Lab space and L-shaped anchor points were used for vegetation and background. * Perform brightness normalization, and for a * With b * By performing background bias elimination and introducing chromaticity invariant extraction, the leaf color characterization quantity C maintains a consistent caliber under different light and background conditions, thereby stably distinguishing between yellowing weak seedlings and greening strong seedlings, and improving the accuracy and robustness of strong seedling grading and weak seedling classification.

[0022] As a further aspect of this invention, when extracting C, V, H, L, and B, interference suppression and credibility fusion are performed. Specifically, a shadow mask, a bare soil mask, and a frost residue mask are constructed within the vegetation mask. A credibility weight ρ is calculated for each pixel. The credibility weight ρ is jointly determined by the pixel's spectral consistency, neighborhood texture stability, and exponential anomaly amplitude. Within the vegetation mask, the spectral consistency score is obtained by the deviation of the pixel's multispectral vector from the neighborhood robust center. The texture stability score is obtained by the fluctuation of the gradient texture within the neighborhood. The exponential anomaly score is obtained by the normalized deviation of NDVI from the neighborhood median. Combined with the preset basic suppression coefficients given by the shadow, bare soil, and frost masks, the spectral consistency score, texture stability score, and exponential anomaly score are weighted according to preset weights and normalized to 0-1 to jointly determine the pixel credibility weight ρ. When calculating C, V, H, L, and B for a single plant object, the pixel contribution is weighted by ρ, and adaptive redistribution is performed on W1-W5 based on the mean of ρ within the single plant object.

[0023] It should be noted that, in response to the problems caused by interference from shadows, bare soil, and frost residue in overwintering remote sensing images, which leads to abnormal pixel spectra and textures, NDVI mutations, and contamination of C, V, H, L, and B extraction, resulting in instability of single-plant seedling strength values ​​P and grading results, as well as cross-plot aperture drift, an interference mask is constructed within the vegetation mask. A pixel confidence weight ρ is generated by jointly using spectral consistency, texture stability, and exponential anomaly amplitude to weight the pixel contribution. At the same time, W1 to W5 are adaptively redistributed based on the mean ρ of a single plant to reduce the weight of contaminated features. This achieves the suppression of interfering pixels and robustness of multi-feature fusion, thereby improving the stability of single-plant characterization and P calculation, the consistency of strong and weak seedling zoning and grading, and the reliability of weak seedling risk area location.

[0024] The technical advantages of this invention's method for identifying robust wheat seedlings during the wintering period based on remote sensing image data are as follows: This invention utilizes UAV RGB and multispectral remote sensing images under vegetation mask constraints to segment individual plants. It extracts multidimensional characterizations from the image data, including leaf color (C), index (V), seedling height (H), leaf area index (L), and biomass (B), and standardizes and fuses these to obtain a single plant vigor value (P). Then, it uses statistical distribution within quadrats to classify vigorous, strong, and weak seedlings. Finally, it extrapolates the quadrat grades as seeds using geodesic dilatation morphological reconstruction within the mask, suppressing breaks and misclassifications caused by overwintering shadows, bare soil, and boundary noise. This results in a continuous and stable spatial distribution map of vigorous seedling grades and output of risk areas for weak seedling aggregation from quadrats to the entire region, providing a reliable basis for precise management of fertilization based on seedling conditions, moisture control, and replanting. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for identifying robust wheat seedlings during the wintering period based on remote sensing image data according to the present invention. Figure 2 This is a visible light image of Embodiment 1 of the present invention; Figure 3 This is an image of the present invention after soil removal in Embodiment 1. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figure 1 As shown, the present invention proposes a method for identifying robust wheat seedlings during the wintering period based on remote sensing image data, comprising: In the target planting area, quadrats were identified and UAV remote sensing data covering the quadrats were obtained. The remote sensing image data included RGB visible light images, multispectral images and their imaging parameters. The multispectral images included red light band images and near-infrared band images. Perform radiometric correction, geometric correction, orthophoto stitching, and pixel-level registration on remote sensing image data; Based on the registered remote sensing image data, winter wheat vegetation mask was extracted and individual plant objects were obtained by segmentation within the quadrat. For each individual plant, leaf color characteristic C, vegetation index characteristic V, seedling height characteristic H, leaf area index characteristic L, and aboveground biological characteristic B are extracted from remote sensing image data. C, V, H, L, and B are standardized and fused to obtain the seedling vigor value P of each plant. The mean μ and standard deviation σ of the seedling vigor value P of each plant in the quadrat are calculated. Based on μ and σ, the individual plants are divided into vigorous seedlings, strong seedlings, and weak seedlings, and a quadrat level distribution map is generated. The results of each quadrat are extrapolated to the target planting area, and the spatial distribution map of vigorous seedling level of winter wheat during the overwintering period, as well as the strong seedling zoning and weak seedling cluster risk areas, are output.

[0028] To illustrate the specific implementation of the method more specifically, Examples 1 and 2 are provided. In these examples, the single plant refers to the tillering clump of winter wheat during its overwintering period.

[0029] Example 1: Identification of robust seedlings during the overwintering period in the Suixi wheat planting area 1. Overview of the test site The experiment was conducted in 2025 in a wheat-growing area of ​​Suixi County, Huaibei City, Anhui Province. The terrain was flat and the fields were contiguous. The soil was mainly sandy black soil, with significant winter temperature fluctuations and shading interference, resulting in a relatively high proportion of bare soil. The experimental fields adopted a single-season winter wheat planting method, with the planting density set according to the preset values, a row spacing of 20 cm, and a seeding rate of 12 kg / mu. Field management was carried out according to the local pre-winter management practices, without additional water and fertilizer stress treatments.

[0030] 2. Test methods This embodiment uses UAV remote sensing acquisition and image analysis to identify robust seedlings during the overwintering period. The method and steps are as follows: (1) Selection of quadrats: After eliminating the fields and road isolation zones within the target planting area, quadrats are randomly set up. Each quadrat is a 5m×5m rectangular area, and a total of 24 quadrats are set up. The center coordinates of the quadrats are recorded.

[0031] (2) Remote sensing data acquisition: The UAV is equipped with an RGB visible light camera and a multispectral camera to acquire remote sensing image data covering the sample plot. The multispectral image includes red light band image and near-infrared band image. The flight altitude, exposure parameters, pose information and imaging timestamp are recorded simultaneously. The flight altitude is set to 60m, with 80% forward overlap and 70% lateral overlap to obtain the image required for RGB and multispectral orthophoto stitching.

[0032] (3) Radiometric and geometric processing: Perform radiometric correction, geometric correction, orthorectification and pixel-level registration on remote sensing image data to align RGB orthorectification and multispectral orthorectification to the same pixel grid.

[0033] (4) Vegetation mask extraction and single plant object segmentation: NDVI is calculated based on the registered red and near-infrared bands, NDVI=(NIR-Red) / (NIR+Red), where NIR is the surface reflectance in the near-infrared band and Red is the surface reflectance in the red band; winter wheat vegetation mask is generated under NDVI constraint, and single plant objects are obtained by connecting the vegetation mask and refining the segmentation within the quadrat.

[0034] 3. Calculation of individual seedling vigor and quadrat grading (1) Extraction of individual plant characteristics: For each individual plant, extract leaf color characteristic C, vegetation index characteristic V, seedling height characteristic H, leaf area index characteristic L, and aboveground biomass characteristic B from remote sensing image data. Leaf color characterization C: Transform the RGB image to the Lab color space in the registration coordinate system and extract a * With b * Channel, for L * The channel employs dual-anchor point brightness normalization correction using both the wheat vegetation reference set and the background reference set within the sample plots, and also applies to a... * b* After background bias elimination, the color mean and color dispersion are calculated and fused within the single plant object region to obtain C; Vegetation index representation V: NDVI is taken as V; The seedling height characteristic H is calculated by the canopy height model, which is a DSM and DTM difference raster. The quantile value of the canopy height raster within the coverage area of ​​a single plant is taken as H. Leaf area index (L) is a measure jointly mapped from NDVI and canopy coverage (F), where F is the percentage of vegetation pixels within the vegetation cover. L = a1·NDVI + a2·F + a3. In the Suixi experimental area, 20 of the 24 5m × 5m quadrats were selected as calibration quadrats and 4 as validation quadrats. For each calibration quadrat, after RGB and multispectral data acquisition by a UAV and subsequent radiometric correction, orthophoto stitching, and pixel-level registration, the mean NDVI value within the quadrat was calculated. The canopy coverage was obtained by statistical analysis of the vegetation cover. (Vegetation pixel percentage, value 0-1); On the same day, the corresponding quadrat was measured on the ground using a leaf area index meter to obtain the measured leaf area index. ;by For dependent variable, and A linear mapping was established for the independent variable, and the coefficients were obtained by least squares fitting. Finally, the formula for calculating the leaf area index of winter wheat during the overwintering period in Suixi District was determined. a1=2.06, a2=1.27, and a3=-0.12 are the preset calibration coefficients for this embodiment. When these coefficients are used in four validation plots, the correlation coefficients between the predicted and measured values ​​are... The accuracy reached 0.86, and the root mean square error (RMSE) was 0.19 (LAI units), meeting the accuracy requirements for canopy leaf quantity characterization in distinguishing robust seedlings during the overwintering period.

[0035] The aboveground biomass characterization value B is mapped jointly by NDVI, F, and H: B = b1·NDVI + b2·F + b3·H + b4. In the 24 5m×5m quadrats deployed in the Suixi experimental area, 20 quadrats were selected as calibration quadrats and 4 as validation quadrats. For each calibration quadrat, after RGB and multispectral acquisition by UAV and completion of radiometric correction, orthorectification stitching, and pixel-level registration, the mean NDVI value within the quadrat was calculated. Canopy coverage is statistically determined by vegetation cover. (Vegetation pixel ratio), and the mean value of seedling height within the quadrat was calculated using the canopy height model. (Expressed as the mean of the canopy height quantile of the area covered by a single plant within the quadrat); Subsequently, on the same day, 1m×1m sub-quadrats were randomly selected within each calibration quadrat for aboveground biomass measurement. All aboveground plants were collected using a ground-level cutting method, and their fresh weight was measured on-site. Representative sub-samples were then dried to constant weight, and the aboveground dry biomass per unit area was calculated. (g / m²); with For dependent variable, , , A linear mapping was established for the independent variable, and coefficients were obtained by least squares fitting. Finally, the formula for calculating the aboveground biomass of winter wheat during the overwintering period in Suixi District was determined. b1=310.5, b2=185.2, b3=42.8, and b4=-96.0 are the preset calibration coefficients for this embodiment, and H is in centimeters. When the above coefficients are used in four validation plots, the correlation coefficient between the predicted B and the measured dry biomass BM is determined. The accuracy reached 0.84, and the root mean square error (RMSE) was 18.7 g / m², meeting the accuracy requirements for biomass characterization in distinguishing robust seedlings during the overwintering period.

[0036] (2) P calculation: Normalize C, V, H, L, and B to obtain C′, V′, H′, L′, and B′ respectively, and calculate according to P=W1·C′+W2·V′+W3·H′+W4·L′+W5·B′, where W1~W5 are preset non-negative weights and their sum is 1; in this embodiment, W1=0.18, W2=0.22, W3=0.20, W4=0.20, and W5=0.20.

[0037] (3) Grading within the quadrat: Calculate the mean μ and standard deviation σ of all individual plants P within the quadrat, and grade them according to the following criteria: when P ≥ μ + σ, they are vigorous seedlings; when μ - σ < P < μ + σ, they are strong seedlings; when P ≤ μ - σ, they are weak seedlings. Generate a quadrat grade distribution map. In this embodiment, taking quadrat S-07 as an example, a total of N = 168 individual plants (tillering clumps) were obtained after dividing the 5m × 5m quadrat. Calculate the seedling vigor value P for all individual plants and obtain the quadrat mean μ = 0.647, standard deviation σ = 0.082, threshold μ + σ = 0.729, and μ - σ = 0.565. According to the grading criteria, 34 plants with P ≥ 0.729 are considered vigorous seedlings; 29 plants with P ≤ 0.565 are considered weak seedlings; and the rest are considered strong seedlings.

[0038] 4. Spatial consistency correction, risk areas of weak seedling clusters, and extrapolation output (1) Spatial consistency constraint correction: The crop row direction angle θ is estimated based on the vegetation mask, and the center coordinates of the single plant object are rotated to the row direction coordinate system (u,v), with the u axis along the crop row direction and the v axis along the direction perpendicular to the crop row direction; a single plant adjacency graph is constructed in the row zone according to the row direction adjacency relationship, and the same zone nearest neighbor consistency penalty and isolated weak seedling and isolated strong seedling removal rules are applied to the grading results. The single plant grade is corrected based on the cost minimization result in the adjacency graph.

[0039] Taking the above-mentioned quadrat S-07 (5m×5m) as an example, based on the winter wheat vegetation mask in this quadrat, the skeleton line segments are first extracted and the main direction of the skeleton line segments is weighted and fitted to obtain the crop row direction angle θ=12.4° (counterclockwise relative to the x-axis of the registration coordinate system); for the N=168 individual plants obtained by segmenting within the quadrat, their center coordinates are taken ( , ) and press = cosθ+ sinθ = sinθ+ The system is rotated to a row coordinate system (u, v) using cosθ. Then, the individual plant is divided into 23 rows with a row band width of ∆v = 0.22m along the v-axis. Within each row band, plants are sorted by u from smallest to largest, and a single plant adjacency graph is constructed by connecting K=2 nearest neighbors before and after each plant. For plant A (initially classified as a weak seedling, P=0.57) located in the 9th row band in the initial classification results, 3 of its 4 nearest neighbors in the same band are medium-sized seedlings and 1 is a weak seedling, satisfying the isolated weak seedling criterion (n=0.57). i =1 <n min =2 and the inconsistency rate q i =0.75≥η=0.6), therefore, the grade of A is corrected to strong seedling by majority vote; for another single plant object B in the same zone (initially judged as strong seedling, P=0.74), only one of its four nearest neighbors is a vigorous seedling, the rest are strong seedlings, and B is located in the high reflection noise area at the row boundary, which satisfies the isolated strong seedling criterion, so B is corrected to a strong seedling; at the same time, a cluster of continuous weak seedling objects in the 15th row zone (all five adjacent objects are weak seedlings) is not isolated and the weak seedling grade is maintained; after completing the above-mentioned neighbor consistency penalty and isolated point removal in the same zone, the final single plant grade map is obtained by minimizing the number of grade changes before and after correction and minimizing the cost of consistency with neighbors, thereby eliminating scattered misjudged points in the row zone at the quadrat scale and retaining real continuous weak seedling patches, providing a stable seed distribution for subsequent statistics of weak seedling cluster risk areas and global extrapolation.

[0040] (2) Weak seedling cluster risk area: After the correction is completed, the center point of the weak seedling is used as the event point set, and the number of weak seedlings is counted in the neighborhood with a radius of r=2.0m to obtain D(x), and D0=8 is taken; when D(x)≥D0 and the current position of the quadrat meets the judgment condition of the Class III strong seedling quadrat, it is marked as a weak seedling cluster risk area.

[0041] Taking plot S-07 as an example, grid points were selected in a contiguous area of ​​weak seedlings in the southwest corner of the plot. The number of weak seedling events D ( ) was counted within its 2.0m neighborhood. =11, and the corresponding quadrats in this contiguous area meet the criteria for Class III robust seedling quadrats (S R =29 / 168=0.173<S R1 And S I =0.41<S I1 Therefore, the position Continuous grid cells within the quadrat satisfying D(x)≥8 are marked as areas of risk for weak seedling clustering; while grid points are taken in another scattered area of ​​weak seedlings in the middle of the quadrat. Within its 2.0m neighborhood, only the number of weak seedling events D ( Since D(x) = 5 < D0 = 8, the density threshold is not met, so it is not marked as a risk area. Finally, all grid connected regions that satisfy D(x) ≥ D0 and belong to the sample plot with strong seedlings of level III are merged, and the holes are filled and the boundaries are smoothed to obtain the spatial range of the risk area of ​​weak seedlings in the sample plot, which is used to locate the priority area for subsequent zonal fertilization and replanting.

[0042] (3) Extrapolation: Under the constraint of winter wheat vegetation cover, the quadrat grade results are used as seeds to perform geodesic expansion morphological reconstruction extrapolation on the grade area, and the overlapping propagation area is judged according to the principle of minimum morphological distance. Finally, the target area grade spatial distribution map is generated by purification through opening and closing operations.

[0043] 5. Result Examples and Disposal Identification Output This embodiment outputs a spatial distribution map of winter wheat grades during the overwintering period in the Suixi area, and generates treatment markers in areas with a risk of weak seedling clusters: For weak seedlings with yellowing due to lack of fertilizer, apply 8-10 kg / mu of urea and 5-8 kg / mu of diammonium phosphate during the peak tillering period before winter, combined with irrigation. For areas with sparse or weak seedlings, provide assistance in identifying and replanting seedlings; For areas with excessive seedling growth, avoid applying nitrogen and instead use soil compaction or deep cultivation to control excessive growth and conserve moisture. For Grade I strong seedling areas, nitrogen fertilizer is generally not applied before winter.

[0044] like Figure 2 and Figure 3As shown in the comparison, in this embodiment, the Normalized Difference Vegetation Index (NDVI) is calculated based on multispectral red and near-infrared bands, and a winter wheat vegetation mask is generated by threshold segmentation. After morphological opening and closing operations are combined to purify the mask, single-plant object segmentation is performed under the mask constraint. It can be seen that: Figure 2 In the visible light spectrum, bare soil particles, straw residues, and shadow textures intertwine with sparse seedlings, making it difficult to reliably distinguish the seedlings from the background. Figure 3 The soil background was effectively removed, and only pixels consistent with winter wheat vegetation were retained, making the ROI extracted by subsequent C, V, H, L, and B characterization metrics purer and reducing false positives and false negatives in single-plant segmentation. This improved the accuracy and stability of the classification of strong and weak seedlings and the location of risk areas for weak seedling clusters in low-coverage scenarios during the overwintering period.

[0045] Example 2: Identification of robust seedlings during the overwintering period in Mengcheng wheat-growing area 1. Overview of the test site The experiment was conducted in 2024 in a wheat-growing area of ​​Mengcheng County, Bozhou City, Anhui Province. The fields were contiguous and flat, with frequent low temperatures and morning frosts in winter. During the overwintering period, residual frost caused high brightness interference, and local differences in soil fertility led to patches of weak seedlings. The experimental fields adopted conventional winter wheat cultivation methods, with the same planting density as in Example 1, a row spacing of 20 cm, and a seeding rate of 12 kg / mu. Field management was carried out according to local conventional pre-winter management practices.

[0046] 2. Test methods (1) Selection of quadrats: 30 5m×5m quadrats are randomly set up in the target planting area, and the center coordinates of the quadrats are recorded.

[0047] (2) Remote sensing data acquisition: The UAV is equipped with RGB and multispectral cameras to acquire remote sensing image data covering the sample plot. The multispectral images include red light and near-infrared bands. The flight altitude is set to 80m, with 80% forward overlap and 75% lateral overlap. Exposure parameters and pose information are recorded simultaneously.

[0048] (3) Radiometric and geometric processing: Perform radiometric correction, geometric correction, orthorectification stitching and pixel-level registration to obtain RGB orthorectified and multispectral orthorectified data on the same grid.

[0049] (4) Vegetation mask and single plant segmentation: NDVI is calculated and threshold segmentation is performed to obtain winter wheat vegetation mask; single plant object segmentation is performed on the vegetation mask within the quadrat to obtain a set of single plant objects.

[0050] 3. Calculation of individual seedling vigor and quadrat grading (1) Characteristic extraction: C, V, H, L and B were extracted from the single plant object, and the extraction method was the same as in Example 1; the frost residue area was suppressed as an interference source in the subsequent reliable fusion.

[0051] Taking quadrat M-12 as an example, N=214 individual plants (tillers) were obtained after segmentation within the quadrat. For each individual plant, C, V, H, L, and B were extracted as follows: First, NDVI was calculated using multispectral red and near-infrared bands as the vegetation index characterization V, and canopy coverage F was statistically determined using vegetation masking. Second, the canopy height model was obtained from the difference between DSM and DTM, and the 90th percentile value of the canopy height raster within the coverage area of ​​each individual plant was taken as the seedling height characterization H. Then, the leaf area index characterization L was calculated as L=2.14·NDVI+1.19·F-0.10; and the aboveground biomass characterization B (H in centimeters) was calculated as B=295.0·NDVI+170.6·F+39.5·H-88.0. Simultaneously, the RGB image was converted to the Lab color space, and L... * The channel uses a dual anchor point system of vegetation and background reference sets for brightness normalization correction, and performs brightness normalization correction on channel a. * b * After background bias elimination, the mean color and color dispersion are calculated and fused within the individual plant area to obtain the leaf color characterization C. Specifically, due to residual morning frost during the wintering period in the Mengcheng area causing bright patches on the north side of the sample plot, a frost residue mask is constructed within the vegetation mask. Pixels covered by this mask are assigned a lower baseline suppression coefficient (β) in subsequent reliable fusion. f =0.35), so that its pixel contribution to C and V is suppressed by ρ when calculating C, V, H, L, B of a single plant, thereby avoiding the misjudgment of strong seedlings due to abnormal leaf color and index caused by frost brightness.

[0052] (2) P calculation: After normalizing C, V, H, L, and B, calculate according to P=W1·C′+W2·V′+W3·H′+W4·L′+W5·B′. In this embodiment, W1=0.20, W2=0.20, W3=0.20, W4=0.20, and W5=0.20.

[0053] (3) Intra-plot grading: Calculate μ and σ within the plot, and generate a plot grading distribution map according to P≥μ+σ for strong seedlings, μ-σ<P<μ+σ for medium seedlings, and P≤μ-σ for weak seedlings.

[0054] Taking quadrat M-12 (5m×5m) in Example 2 of Mengcheng as an example, a total of N=214 individual plants (tillering clumps) were obtained within the quadrat after being constrained by vegetation and segmented into individual plants. The seedling vigor value P was calculated for all individual plants, and the average value μ=0.612, standard deviation σ=0.078, threshold μ+σ=0.690, and μ-σ=0.534 were obtained. According to the grading standard, 41 plants were judged as vigorous seedlings with P≥0.690; 36 plants were judged as weak seedlings with P≤0.534; and the rest were judged as strong seedlings with P≤0.534. The grading results of each individual plant were then filled back into its spatial location within the quadrat to form a quadrat grading distribution map.

[0055] 4. Interference suppression, reliable fusion, spatial consistency correction, and weak seedling risk zone (1) Interference suppression and credibility fusion: A shadow mask, a bare soil mask and a frost residue mask are constructed within the vegetation mask; the credibility weight ρ is calculated for each pixel. ρ is obtained by weighting the spectral consistency score, texture stability score and index anomaly score according to the preset weight and normalizing them to 0 to 1, and combined with the basic suppression coefficient of shadow / bare soil / frost; when calculating C, V, H, L and B of a single plant object, the contribution of the pixel is weighted by ρ, and W1 to W5 are adaptively redistributed according to the mean ρ in a single plant object, so that the influence of frost bright and shadow low bright pixels on leaf color and index is suppressed.

[0056] Taking plot M-12 (5m×5m) as an example, after registration and vegetation mask extraction, a shadow mask, a soil exposure mask, and a frost residue mask were further constructed within the vegetation mask: the shadow mask was constructed from the luminance channel L after RGB to Lab conversion. * Below the sample plot L * The soil bare mask consists of pixels whose median value is less than 1.5 times the MAD. The soil bare mask consists of pixels whose NDVI is below 0.15 and do not belong to the vegetation mask boundary buffer zone. The frost residue mask consists of pixels whose L value is higher than the median L of the quadrat plus 2.0 times the MAD and whose NDVI is abnormally high or low. For each pixel x within the vegetation mask, a confidence weight ρ(x) is calculated. Specifically, the deviation of the pixel's multispectral vector [Red, Green, NIR] relative to the neighborhood robust center is calculated within a 3×3 neighborhood to obtain the spectral consistency score S(x). The MAD of the grayscale gradient magnitude within the neighborhood is calculated to obtain the texture stability score T(x). The normalized deviation of NDVI relative to the neighborhood median is used to obtain the exponential anomaly score A(x). Then, ρ(x) = clip(W) is used to calculate the weight. s ·S(x)+W t ·T(x)+W v The fusion weights in the range of 0 to 1 are obtained by multiplying the shadow pixels by β. s =0.60, frost-residual pixels multiplied by β f =0.35, soil exposed pixels are set to 0; for example, a frost-exposed high-brightness pixel x on the north side of the sample plot. f When not suppressed, due to L * Excessive levels of C caused a yellowish tint to the leaf color and localized anomalies in NDVI. The calculated S(x) f )=0.52、T(x f )=0.48、A(x f =0.73, according to the preset weight W s =0.4, W t =0.3、W v=0.3 gives the original value of ρ as approximately 0.39, then multiply by β. f After =0.35, we get ρ(x) f The value of x = 0.14, thus reducing the contribution of this pixel in calculating the C and V of a single plant; while a normal vegetation pixel x in the middle of the quadrat... n S(x) n )=0.86、T(x n )=0.79、A(x n )=0.12, thus obtaining ρ(x) n The contribution was 0.83, maintaining a high level. Subsequently, for each individual plant within the quadrat, C, V, H, L, and B were calculated by weighting the pixel contribution with ρ(x) within its masked area, and the average confidence level of that individual plant was calculated. ,when When the value is below the preset threshold (0.55), adaptive redistribution is performed on the P fusion weights W1 to W5. For example, the weight of leaf color W1, which is easily affected by light, is reduced from 0.20 to 0.12, the weight of index W2 is reduced from 0.20 to 0.16, and the weights of structural quantity seedling height and LAI W3 and W4 are increased to 0.24 and 0.24 respectively (W5 remains at 0.24 or is allocated according to the rules). This reduces the dominant role of C and V on P in individual plants with strong interference from frost-induced high brightness and shadow-induced low brightness, and ultimately suppresses the misjudgment of strong seedlings caused by leaf color and index drift due to frost and shadow.

[0057] (2) Spatial consistency correction: Establish a row coordinate system (u,v) according to the row angle θ, construct an adjacency graph and perform cost minimization correction, remove isolated weak seedlings or isolated strong seedlings in the row zone, and obtain more continuous grading results.

[0058] (3) Risk area for weak seedlings: take r=2.5m, D0=10 plants / m 2 The number of weak seedlings in the neighborhood is counted to obtain D(x). When D(x)≥D0 and the location of the quadrat meets the criteria for a Class III strong seedling quadrat, it is marked as a risk area for weak seedling aggregation.

[0059] 5. Extrapolation and Disposal Identification Output Using quadrat grades as seeds, geodesic dilatation morphological reconstruction extrapolation was performed under vegetation mask constraints. Overlapping propagation areas were determined according to the principle of minimum morphological distance. Finally, opening and closing operations were used to purify and output a spatial distribution map of winter wheat grades during the overwintering period in the Mengcheng area. Simultaneously, treatment labels were added to different regions in the output layer. For weak seedlings with yellowing due to nutrient deficiency: Apply 8-10 kg / mu of urea during the peak tillering period before winter, along with 5-8 kg / mu of diammonium phosphate; In areas with sparse or weak seedlings: check and replant seedlings; In areas with stunted growth and weak seedlings: cultivate the soil promptly after winter irrigation or rain; In areas prone to excessive seedling growth: avoid applying nitrogen and instead compact the soil or perform deep cultivation to control excessive growth and conserve moisture; Grade I strong seedling area: Generally no nitrogen fertilizer is applied before winter.

[0060] As can be seen from Examples 1 and 2 above, the method proposed in this invention can achieve stable segmentation of single plants and multidimensional collaborative characterization of leaf color C, index V, seedling height H, leaf area index L, and biomass B based on UAV RGB and multispectral remote sensing data in two significantly different overwintering field scenarios in Suixi and Mengcheng. Through standardized fusion of P and statistical grading within quadrats, objective discrimination results of strong seedlings, medium seedlings, and weak seedlings are obtained. Under the effects of spatial consistency constraint correction, local density determination of weak seedlings, and morphological reconstruction extrapolation, isolated misjudgments and zoning breaks caused by interference from shadows, bare soil, and frost residues are suppressed. Finally, a continuous spatial distribution map of strong seedling levels and risk areas of weak seedling aggregation are formed, providing a directly executable zoning basis for precise management such as fertilization according to seedling conditions, moisture retention and replanting during the overwintering period.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0062] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying robust wheat seedlings during the wintering period based on remote sensing image data, characterized in that, include: In the target planting area, quadrats were identified and UAV remote sensing data covering the quadrats were obtained. The remote sensing image data included RGB visible light images, multispectral images and their imaging parameters. The multispectral images included red light band images and near-infrared band images. Perform radiometric correction, geometric correction, orthophoto stitching, and pixel-level registration on remote sensing image data; Based on the registered remote sensing image data, winter wheat vegetation mask was extracted and individual plant objects were obtained by segmentation within the quadrat. For each individual plant, leaf color characteristic C, vegetation index characteristic V, seedling height characteristic H, leaf area index characteristic L, and aboveground biological characteristic B are extracted from remote sensing image data. C, V, H, L, and B are standardized and fused to obtain the individual plant seedling vigor value P. The mean μ and standard deviation σ of the individual plant seedling vigor value P within the quadrat are calculated. Based on μ and σ, individual plants are classified into vigorous seedlings, strong seedlings, and weak seedlings, and a quadrat grade distribution map is generated. The results of each quadrat are extrapolated to the target planting area, and a spatial distribution map of winter wheat seedling condition grade and a risk area for weak seedling aggregation during the overwintering period are output.

2. The method for identifying robust wheat seedlings during the wintering period based on remote sensing image data according to claim 1, characterized in that, In the spatial distribution map of quadrat grades, the grading criteria for vigorous seedlings, strong seedlings, and weak seedlings are as follows: when P ≥ μ + σ, the seedlings are graded as vigorous seedlings; when μ - σ < P < μ + σ, the seedlings are graded as strong seedlings; and when P ≤ μ - σ, the seedlings are graded as weak seedlings.

3. The method for identifying robust wheat seedlings during the wintering period based on remote sensing image data according to claim 1, characterized in that, The seedling potential value P of a single plant is calculated as follows: C, V, H, L, and B are normalized to obtain C′, V′, H′, L′, and B′, respectively, and then calculated according to P=W1·C′+W2·V′+W3·H′+W4·L′+W5·B′, where W1~W5 are preset non-negative weights, and satisfy W1+W2+W3+W4+W5=1.

4. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 1, characterized in that, Before outputting the distribution map of robust seedling grades, spatial consistency constraint correction is performed. Specifically, the crop row angle θ is estimated based on the wheat vegetation mask, and the center coordinates of individual plants are rotated to the row coordinate system. The row coordinate system is a two-dimensional orthogonal coordinate system (u,v) established based on the crop row angle θ, where the u-axis is along the crop row direction and the v-axis is along the direction perpendicular to the crop row. An adjacency graph of individual plants is constructed in the row zone according to the row adjacency relationship. The same-zone nearest neighbor consistency penalty and isolated vigorous seedling and isolated weak seedling removal rules are applied to the individual plant grading results. The individual plant grade is corrected based on the cost minimization result in the adjacency graph. After the correction is completed, the center points of vigorous and weak seedlings are used as event point sets. The number of weak seedlings is counted in the neighborhood of the preset radius r to obtain the local density of weak seedlings D(x). When D(x)≥D0 and the current position of the sample plot meets the judgment condition of the Class III robust seedling sample plot, the current position is marked as the weak seedling aggregation risk area, where r and D0 are preset parameters.

5. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 1, characterized in that, While generating the spatial distribution map of robust seedling grades, the proportion S of robust seedlings is calculated. R With the seedling index S I， First, determine the seedlings as vigorous, strong, and weak in different zones. Then, classify the quadrats within the strong seedling zone and output treatment labels. Simultaneously, classify the vigorous and weak seedling zones and output treatment labels: where S R S is the ratio of the number of robust seedlings in the quadrat to the total number of seedlings. I Press S I =(P z -P r ) / μ calculation, P z P represents the average seedling vigor value P of a single strong seedling. r The average seedling vigor value P of weak seedlings; set the threshold S for judging strong seedlings. R0 S I0 When S R ≥S R0 And S I ≥S I0 At that time, the quadrat is assigned to the robust seedling zone and a grading threshold S is set within the current robust seedling zone. R1 S R2 S I1 S I2 And satisfy S R2 >S R1 ≥S R0 S I2 >S I1 ≥S I0 When S R ≥S R2 And S I ≥S I2 When S is identified as a Grade I robust seedling quadrat, a treatment label is output. R ≥S R2 And S I1 ≤S I <S I2 When S is identified as a Grade II robust seedling quadrat, a treatment label is output. R1 ≤S R <S R2 And S I1 ≤S I <S I2 When S is identified as a Grade III robust seedling quadrat, it is output; when S R <S R0 or S I <S I0 At that time, the quadrats were designated as areas of risk for weak seedling clustering, and the average leaf color characteristic value within the quadrats was used as the basis for classification. Mean value of aboveground biomass Mean value of leaf area index Mean of seedling height Weak seedlings were categorized, and a categorization threshold C was set. W B W L W H W With M D0 C W B W L W H W With M D0 These are the threshold values ​​for weak seedling yellowing, weak seedling low biomass, weak seedling low leaf area index, weak seedling short and tall, and missing and sparse seedlings. M D The ratio of the total number of individual plants in a quadrat to the area of ​​the quadrat is given by the following formula: ≥C W and ≤B W When identified as weak seedlings exhibiting chlorosis due to nutrient deficiency, a treatment label is issued, and when M... D ≤M D0 When the seedlings are identified as sparse and weak, a treatment label is generated. The remaining weak seedling plots are identified as slow-growing weak seedlings, and a treatment label is generated.

6. The method for identifying robust wheat seedlings during the wintering period based on remote sensing image data according to claim 1, characterized in that, The seedling height characteristic H is calculated by the canopy height model, which is a grid of difference between the digital surface model and the digital ground model. The quantile value of the canopy height grid within the coverage area of ​​a single plant is used as the seedling height characteristic H of the single plant.

7. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 1, characterized in that, The leaf area index (L) is obtained by jointly mapping the vegetation index (V) and the canopy cover (F). The canopy cover (F) is the percentage of winter wheat vegetation pixels within the winter wheat vegetation mask. The vegetation index (V) includes NDVI, where NDVI = (NIR - Red) / (NIR + Red). NIR is the surface reflectance in the near-infrared band, which is collected by a multispectral camera in the near-infrared channel. Red is the surface reflectance in the red band, which is collected by a multispectral camera in the red channel. The result is calculated as L = a1·NDVI + a2·F + a3, where a1, a2, and a3 are preset calibration coefficients.

8. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 7, characterized in that, The aboveground biomass characterization quantity B is obtained by jointly mapping the vegetation index characterization quantity V, the canopy coverage quantity F, and the seedling height characterization quantity H, and is calculated according to B=b1·NDVI+b2·F+b3·H+b4, where b1, b2, b3, and b4 are preset calibration coefficients.

9. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 1, characterized in that, The leaf color characterization C is obtained as follows: the RGB image is transformed to the Lab color space in the registration coordinate system, and color a is extracted. * Channels and Colors b * Channel, based on the brightness statistics of non-vegetated areas within the sample plot, for brightness L * After performing illumination normalization correction on the channel, the color mean and color dispersion within the individual plant area are calculated and fused to obtain the leaf color characterization quantity C, where L * For the dual-anchor point brightness normalization correction channel based on the wheat vegetation reference set and background reference set within the sample plot, a * b is the red-green axis chromaticity channel characterizing the degree of leaf greening after background bias elimination. * To characterize the degree of leaf yellowing after background bias elimination, the yellow-blue axis chromaticity channel was extracted in combination with chromaticity invariants.

10. The method for identifying robust wheat seedlings during the overwintering period based on remote sensing image data according to claim 3, characterized in that, When extracting C, V, H, L, and B, interference suppression and credibility fusion are performed. Specifically, a shadow mask, a bare soil mask, and a frost residue mask are constructed within the vegetation mask. For each pixel, a credibility weight ρ is calculated. The credibility weight ρ is jointly determined by the pixel's spectral consistency, neighborhood texture stability, and exponential anomaly amplitude. Within the vegetation mask, the spectral consistency score is obtained by the deviation of the pixel's multispectral vector from the neighborhood robust center. The texture stability score is obtained by the fluctuation of the gradient texture in the neighborhood. The exponential anomaly score is obtained by the normalized deviation of NDVI from the neighborhood median. Combined with the preset basic suppression coefficients given by the shadow, bare soil, and frost masks, the spectral consistency score, texture stability score, and exponential anomaly score are weighted according to preset weights and normalized to 0 to 1 to jointly determine the pixel credibility weight ρ. When calculating C, V, H, L, and B for a single plant object, the pixel contribution is weighted by ρ, and adaptive redistribution is performed on W1 to W5 based on the mean of ρ within the single plant object.

Images