A seedling root system disc identification method and system based on image processing

Abstract

The application belongs to the technical field of image processing, and particularly relates to a seedling root system disc identification method and system based on image processing, which comprises the following steps: collecting an original image and preprocessing, calculating an optimal segmentation threshold, extracting a root system region of interest, obtaining a root system skeleton through thinning processing, and calculating a support radius of each point of the skeleton and a discrete curvature of each skeleton point to obtain an average root system curvature by averaging; calculating a root system fractal dimension; constructing a seedling root system sample library, normalizing the average root system curvature and the fractal dimension, calculating two types of feature weight coefficients based on information entropy theory, and obtaining a disc root degree evaluation index through weighted summation; determining mild and severe disc root threshold values by combining sample library statistical characteristics and ROC curve analysis, mapping a classification judgment function to a grading signal, introducing a confidence degree checking mechanism to process boundary samples, and finally completing seedling automatic sorting and manual review triggering. The application improves the accuracy of seedling root system disc identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a seedling root system disc identification method and system based on image processing. BACKGROUND

[0002] Seedling quality evaluation is a key link in forestry production and landscaping engineering, and is directly related to the survival rate of afforestation, ecological restoration effect and later maintenance cost; among them, the discing of root system, i.e. the spiral growth or mutual entanglement of root system along the container wall due to growth restriction in the container, is one of the main reasons for the failure of root system to stretch, the decline of absorption capacity and even death after transplanting; therefore, rapid and accurate detection and grading of seedling root system at the production nursery link has important economic value and ecological significance for eliminating inferior seedlings and improving afforestation quality.

[0003] The existing seedling root system detection method mainly relies on manual destructive sampling or simple surface analysis based on machine vision; manual detection needs to take out or cut the roots of seedlings from the container, which not only has low efficiency and high labor intensity, but also causes irreversible physical damage to the seedlings, limiting its application in large-scale assembly line; and the existing machine vision detection method is mostly based on traditional image edge detection or gray threshold segmentation technology, which indirectly judges the degree of discing by calculating the projection area or edge circumference of the root system; however, due to the complex soil background, low contrast between root system and soil color, and complex nonlinear topological morphology of discing structure, the existing algorithm is difficult to accurately extract the skeleton of single root system from the adhered soil particle clusters, resulting in low recognition rate at the initial stage of discing or hidden entanglement, and easy to produce missed judgment or misjudgment.

[0004] Therefore, how to solve the problem of difficult root system feature extraction under complex soil background, and improve the recognition accuracy and robustness of nonlinear discing structure, has become a technical problem to be solved in the current seedling nondestructive detection field; the existing technology is limited by the insufficient analysis ability of image processing algorithm for complex texture features, and it is difficult to realize accurate quantification of subtle root system entanglement under the premise of ensuring nondestructive detection. SUMMARY

[0005] To solve the above technical problem of difficult root system feature extraction under complex soil background, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a seedling root disc identification method based on image processing, comprising: collecting an original image of a seedling to be tested; pre-processing the original image to calculate an optimal segmentation threshold for obtaining a root region of interest; performing thinning processing on the root region of interest to obtain a single-pixel-width root skeleton; calculating the support radius of each point on the root skeleton by a distance transform method, defining and calculating the discrete curvature of each skeleton point based on the theory of differential geometry, and taking the arithmetic average of all discrete curvatures to obtain the average root curvature; calculating the fractal dimension of the root skeleton by a box counting method; constructing a seedling root sample library containing different tree species, seedling ages and disc degrees; performing minimum-maximum value normalization processing on the average root curvature and the fractal dimension based on the sample library; calculating the weight coefficients of the two types of normalized features based on the information entropy theory, and obtaining a disc degree evaluation index through weighted summation; determining the mild disc root threshold and the severe disc root threshold based on the statistical distribution characteristics of the sample library combined with ROC curve analysis; mapping the disc degree evaluation index to a classification signal through a classification judgment function, and introducing a confidence check mechanism to process threshold boundary samples; and controlling an execution mechanism to complete the automatic sorting and manual review triggering of the seedling according to the classification signal and the confidence check result.

[0007] The method realizes non-destructive, accurate and automatic identification and sorting of seedling root discs, effectively solves the problems of low efficiency, easy damage to seedlings in manual detection, and low recognition rate, high false alarm rate in traditional machine vision detection of disc roots; through double-dimensional feature extraction and adaptive weighted fusion, the micro-bending and macro-winding characteristics of the disc roots are accurately captured, and the optimized threshold and confidence check mechanism are combined to ensure the reliability of the classification results; it is suitable for large-scale pipeline production and can complete seedling classification and sorting without manual intervention, significantly improving the detection efficiency and reducing the labor intensity, providing strong support for improving the afforestation quality, and having strong practicality.

[0008] Preferably, the pre-processing of the original image comprises: ; wherein, is the brightness value of the enhanced coordinates , is a Gaussian enveloping function and satisfies , is a normalization constant to ensure that the integral value of the Gaussian enveloping function is 1; is the standard deviation of the th scale; is a logarithmic function with a base of 10; represents convolution operation, is the optimal number of Gaussian scales verified by experiments to adapt to the seedling root scene.

[0009] Preferably, the calculation of the optimal segmentation threshold comprises: ; wherein, It is the optimal segmentation threshold; Indicates in set In, the objective function To obtain the maximum value The value; The gray level to be traversed, with a value range from 0 to 255; , The segmentation thresholds are respectively The percentage of pixels in the foreground (root system) and background (soil); , These are the average grayscale values ​​of the foreground and background, respectively. This represents the average grayscale value of the enhanced grayscale image.

[0010] This method determines the optimal segmentation threshold by traversing the entire grayscale level and maximizing the inter-class difference between the foreground and background. This accurately distinguishes root systems from soil regions, effectively addressing the issues of low contrast and blurred boundaries between the two in complex soil backgrounds. This approach adaptively finds the most suitable segmentation standard without manual intervention, resulting in clearer root edges, reduced segmentation errors caused by tiny soil particles adhering to the roots, and ensuring that the extracted root regions of interest are pure and undisturbed. This provides a high-quality data foundation for subsequent skeleton construction and feature calculation.

[0011] Preferably, the step of calculating the support radius of each point on the root system skeleton using the distance transformation method includes: calculating the support radius of the first point on the root system skeleton using the distance transformation method. The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the edge of the root system That is, the first The support radius of each skeleton point; if it appears In the case of extreme fine branches, the average of the support radii of three skeleton points (one adjacent to the front and one to the rear) is taken as the first... The support radius of each skeleton point.

[0012] Preferably, the step of defining and calculating the discrete curvature of each skeleton point based on differential geometry theory includes: In the formula, It is the first Discrete curvature of each skeleton point, in radians per pixel; It is the first Forward vectors of each skeleton point; It is the first The backward vector of each skeleton point; This represents the vector dot product operation; Represents the magnitude of the vector; It is the inverse cosine function; For the first The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the edge of the root system.

[0013] This method uses distance transformation to obtain the shortest distance from each point in the framework to the root edge as the support radius, which can accurately reflect the root thickness at the corresponding location. In cases of extremely thin branches, the average value of three adjacent points is used to supplement the data, avoiding anomalies in subsequent calculations and ensuring data integrity. Based on this, discrete curvature is calculated using differential geometry principles. By weighting the root thickness, the bending characteristics of fine roots are highlighted. Since fine roots are more easily restricted by the container and tend to coil, this design can significantly improve the sensitivity of early root coiling identification, accurately capture abrupt changes in root growth direction, and make the quantification of micro-bending degree more consistent with the actual situation of root coiling, providing accurate micro-feature support for determining the degree of root coiling.

[0014] Preferably, the calculation of the fractal dimension of the root system skeleton using box counting includes: In the formula, It is the fractal dimension of the root system; To cover the root system region of interest The grid side length; Complete coverage of the root system skeleton at each scale Minimum number of grid cells required; It is a logarithmic function with base 10; For limit operations; using the least squares method Perform linear regression, setting the goodness of fit to be greater than or equal to 0.98. The absolute value of the slope of the regression line is the fractal dimension of the root system. , The value ranges from 1.2 to 1.8.

[0015] This method calculates the fractal dimension using box counting, covering the root system's region of interest with a multi-scale grid, and combining linear regression and goodness-of-fit requirements to ensure reliable results. This method comprehensively captures the macroscopic topological entanglement characteristics of the root system. The numerical changes in the fractal dimension intuitively reflect the density and complexity of the root system's entanglement in space. The limited range of values ​​makes the results more referential, effectively overcoming the shortcomings of traditional single-feature methods that cannot characterize macroscopic entanglement states. Combined with microscopic bending features, it forms a two-dimensional support, making the evaluation of root entanglement more comprehensive and objective, and improving the robustness and accuracy of the overall identification method.

[0016] Preferably, the calculation of the weight coefficients for the two types of normalized features based on information entropy theory includes: In the formula, It is the first Information entropy of features; This refers to the number of samples in the seedling root system sample bank. It is an index of samples in the seedling root sample bank; For the first The sample at the th The proportion of the normalized value of a feature to the total sum of the normalized values ​​of that feature in all samples; It is an index of features. Represents the normalized fractal dimension feature, Represents the normalized mean curvature ; Based on the natural constant Logarithmic function with base 0; information entropy The range of values ​​is .

[0017] This method calculates the weight coefficients of two types of normalized features based on information entropy theory. It automatically quantifies the ability of the normalized fractal dimension and average curvature to distinguish the degree of root entanglement, completely avoiding subjective bias and experience dependence caused by manually setting weights. The value of information entropy can intuitively reflect the distribution differences of various features in samples with different degrees of root entanglement. The weight coefficients derived from this allow features with higher sensitivity to root entanglement to receive a higher proportion, making the calculation of the root entanglement evaluation index more closely match the representation needs of actual root entanglement morphology. This highlights the core role of macroscopic entanglement and microscopic bending features, while also adapting to the characteristic differences of samples from different tree species and seedling ages, significantly improving the relevance and reliability of the evaluation index.

[0018] Preferably, the calculation of the weight coefficients for the two types of normalized features based on information entropy theory further includes: using the difference coefficient... Quantify feature discrimination, set The difference coefficients are normalized to obtain the weights of the two types of features. The relation is: ;when Time corresponding weight ,when Time corresponding weight ,and .

[0019] Preferably, the method for obtaining the confidence level includes: In the formula, It is the confidence level; It is an indicator for evaluating the degree of entanglement; It is a root strength evaluation index The nearest classification threshold; For mild encirclement threshold, The threshold for severe encirclement; It is the absolute value symbol; if In 0 to Between, that is, 0 to , Between the midpoints, the nearest classification threshold is ;like In Between 1 and 1, that is , If the midpoint is between 1 and 0, then the nearest classification threshold is... .

[0020] Secondly, the present invention provides a seedling root system identification system based on image processing, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned seedling root system identification method based on image processing is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned image processing-based seedling root system identification method and stored in the memory so that it can be loaded and executed by the processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0022] The beneficial effects of this invention are as follows: Through a complete technical solution involving image preprocessing, dual-dimensional feature extraction, adaptive weighted fusion, and hierarchical early warning, it achieves non-destructive, accurate, and automated identification and sorting of root clumps in seedlings, offering significant advantages over existing technologies. Firstly, the use of multi-scale Retinex illumination compensation and maximum inter-class variance method in collaborative preprocessing effectively solves the problem of low contrast between roots and soil in complex soil backgrounds, accurately extracting pure root regions of interest and avoiding the shortcomings of traditional segmentation algorithms in separating adhered soil particles, thus providing high-quality data support for subsequent feature calculations. Secondly, the innovative fusion of dual-dimensional parameters—average root curvature and fractal dimension—combined with information entropy theory to adaptively determine feature weights overcomes the inadequacy of traditional single features in comprehensively representing root clump morphology. The system significantly improves the sensitivity of identifying early-stage and hidden entanglement scenarios, ensuring accurate differentiation between mild and severe root entanglement. Thirdly, by statistically analyzing extreme values ​​and optimizing thresholds using a large-scale sample library containing different tree species, seedling ages, and root entanglement levels, and combining this with the principle of maximizing the Youden exponent of the ROC curve to determine classification thresholds, a confidence verification mechanism is introduced to handle samples with ambiguous threshold boundaries. This keeps both the overall false alarm rate and false alarm rate low, balancing automated detection efficiency with grading accuracy. Fourthly, it avoids irreversible damage to seedlings, and its detection and sorting efficiency is adapted to the needs of large-scale production lines. It can quickly complete the grading and manual verification of normal, mildly entangled, and severely entangled seedling samples, significantly reducing labor intensity, eliminating inferior seedlings, and improving afforestation survival rates, demonstrating significant application prospects. Attached Figure Description

[0023] Figure 1The flowchart illustrates a seedling root system entanglement identification method based on image processing according to the present invention.

[0024] Figure 2 The diagram illustrates the distribution statistics of the root retraction degree evaluation index of a seedling root retraction degree identification method based on image processing in this invention.

[0025] Figure 3 The diagram illustrates a comparison between the detection methods of this invention and existing methods for identifying seedling root systems based on image processing. 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, not all, of the embodiments of the present invention. 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] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] This invention discloses a method for identifying root clusters in seedlings based on image processing, referring to... Figure 1 This includes steps S1-S4:

[0029] S1. Based on the image of the seedling to be tested, obtain the preprocessed root system region of interest.

[0030] It should be noted that the overall logic of this step is to eliminate soil background interference by removing noise and using segmentation algorithms to extract a clean root image region, providing standardized data input for subsequent morphological analysis. The operation involves image transformation and threshold segmentation, and the formula construction logic is based on Retinex theory and the maximum inter-class variance method. By compensating for the illumination component and determining the adaptive threshold, the contrast between the root system and the soil is enhanced.

[0031] Specifically, an industrial camera is used to acquire side views of the seedlings under test to cover the complete growth morphology of the root system, ensuring that the images can clearly capture root details, and the original images are obtained after acquisition. Image acquisition equipment is a well-known technology in this field and will not be described in detail here.

[0032] Furthermore, regarding the original image Preprocessing operations are performed to eliminate interference and separate the root system from the background. First, the multi-scale Retinex algorithm is used to compensate for the effects of uneven illumination. By simulating the human visual system's perception of illumination at different scales, the dynamic range of the image is compressed and the local details of the root system are enhanced. The formula for calculating the brightness value of the enhanced image is as follows:

[0033] ;

[0034] In the formula, It is the enhanced coordinates The brightness value at that location, It is a Gaussian enclosed function and satisfies , The normalization constant that ensures the integral value of the Gaussian encirclement function is 1 is calculated using a well-known technique, which will not be elaborated here. For the first Standard deviation of each scale; It is a logarithmic function with base 10; This represents the convolution operation. To experimentally verify the optimal number of Gaussian scales suitable for seedling root system scenarios, an example is provided. And corresponding , , The dedicated parameter combinations enable enhancement of root and branch details, balance of details and background, and global illumination compensation, respectively.

[0035] Specifically, the enhanced grayscale image is subjected to adaptive threshold segmentation using the maximum inter-class variance method, and the optimal segmentation threshold is determined by maximizing the inter-class difference between the foreground and the background. The relationship is as follows:

[0036] ;

[0037] In the formula, It is the optimal segmentation threshold; Indicates in set In, the objective function To obtain the maximum value The value; The gray level to be traversed, with a value range from 0 to 255; , The segmentation thresholds are respectively The percentage of pixels in the foreground (root system) and background (soil); , These are the average grayscale values ​​of the foreground and background, respectively. The average grayscale value of the enhanced grayscale image; based on the optimal segmentation threshold. Generate binarized image The pixels belonging to the root system are assigned Pixels belonging to the soil Finally adopted The structuring element performs a morphological opening operation on the binarized image, first eroding to remove tiny soil particle noise of 3 pixels or less, then dilating to preserve the complete topological structure of the root system, finally obtaining the preprocessed root system region of interest data. .

[0038] S2. Based on the region of interest of the root system, construct the root system skeleton, obtain the discrete curvature of each skeleton point and the average curvature of the root system; obtain the fractal dimension of the root system based on the box counting method.

[0039] It should be noted that the overall logic of this step is to transform the pixels of the two-dimensional root system image into quantifiable geometric descriptors. The core is to extract two types of key geometric morphological feature parameters: the average curvature of the root system and the fractal dimension. The average curvature of the root system is used to characterize the microscopic bending degree of the root system, and the fractal dimension is used to characterize the macroscopic topological entanglement complexity of the root system. By simultaneously capturing the spatial distribution state of the root system through dual-dimensional features, the technical problem that traditional single features cannot fully characterize the morphology of the root system is solved, providing core feature support for subsequent determination of the degree of root entanglement. The operation involves skeleton extraction, discrete curvature calculation and fractal dimension analysis. The formula construction logic is based on distance transformation theory and differential geometry principles. An innovative discrete curvature operator that integrates the weight of root thickness is designed to accurately capture the abrupt change in root growth direction caused by root entanglement. At the same time, the box counting method is used to quantify the filling complexity of the root topological entanglement.

[0040] Specifically, the root system region of interest obtained in step S1 The process involves refining the image by using the Zhang-Suen fast refining algorithm to iteratively remove redundant pixels at the edges until a root skeleton with a width of one pixel is obtained. This process strictly preserves the branching structure and intersection information of the root system, ensuring that the skeleton breakage rate is less than or equal to 0.5%, providing an accurate topological basis for subsequent calculations. The Zhang-Suen fast refinement algorithm is an existing technology and will not be elaborated here. 0.5% is an example value and can be set according to actual needs.

[0041] Furthermore, the root system skeleton was calculated using the distance transformation method. Upper The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the edge of the root system This is used to reflect the root thickness at the corresponding skeleton point; if it appears In the case of extreme fine branches, the average of the support radii of three skeleton points (one adjacent to the front and one to the rear) is taken as the first... The support radius of the first skeleton point is determined to avoid the problem of the denominator being zero in subsequent calculations; the first skeleton point is defined based on differential geometry theory. The discrete curvature of each skeleton point is expressed by the following formula:

[0042] ;

[0043] In the formula, It is the first Discrete curvature of each skeleton point, in radians per pixel; It is the first The forward vector of the skeleton point, specifically from the . The skeletal point points to the first The vector of the skeletal point, whose direction corresponds to the root system at the th to The growth direction of the point; It is the first The backward vector of the skeleton point, specifically from the _ ... The skeletal point points to the first The vector of the skeletal point, whose direction corresponds to the root system at the th to The growth direction of the point; This represents the vector dot product operation; Represents the magnitude of the vector; It is the inverse cosine function; For the first The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance to the edge of the root system is calculated; then, the arithmetic mean of the discrete curvatures of all skeleton points is taken to obtain the root system mean curvature. To quantify the overall degree of curvature.

[0044] It should be noted that, based on the definition of differential geometry theory, the first... The physical meaning of the relational formula for the discrete curvature of each skeleton point is to measure the degree of root bending through the change in the angle between the local tangent vectors, while introducing the first... The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the edge of the root system The reciprocal of the weighting is used to give higher weight to the bending features of fine roots, i.e., the shortest Euclidean distance at the root edge. The purpose of giving higher weight to them is because fine roots are more likely to coil due to the constraints of container space. This design can significantly improve the feature sensitivity in the early stage of coiling. The larger the angle, the more severe the bending of the root system at that point, and the more likely it is to be the turning point of coiling.

[0045] Furthermore, the fractal dimension of the root system is extracted as a core parameter characterizing the macroscopic entanglement complexity, and the box counting method is used to complete the calculation. The relationship is as follows:

[0046] ;

[0047] In the formula, It is the fractal dimension of the root system; To cover the root system region of interest The grid side length is set, starting from half the maximum side length of the image and gradually decreasing to 1 pixel, generating a total of 11 grid scales to ensure coverage integrity. Complete coverage of the root system skeleton at each scale Minimum number of grid cells required; It is a logarithmic function with base 10; For limit operations, the mathematical meaning is that when the grid side length... When it approaches 0 infinitely, The limit value; obtained by least squares method Perform linear regression, setting the goodness of fit to be greater than or equal to 0.98. The absolute value of the slope of the regression line is the fractal dimension of the root system. , The value ranges from 1.2 to 1.8. The larger the value, the higher the density of root entanglement in space, and the more severe the root entanglement. The least squares method is a well-known technique and will not be elaborated here. 0.98 is an example value and can be set according to actual needs.

[0048] It should be noted that in the practical application of this technology, it is not necessary to infinitely reduce the mesh side length. Through 11 scales of grid side length Calculate the corresponding After the data is matched, the least squares method is used to perform linear regression fitting, which can approximate the result of the limit calculation, ensuring both calculation accuracy and reducing computational complexity.

[0049] S3. Based on the average curvature of the root system and the fractal dimension of the root system, obtain the root consolidation evaluation index through weighted mapping.

[0050] It should be noted that the overall logic of this step is to integrate two core geometric morphological feature parameters: root mean curvature and fractal dimension. Root mean curvature reflects microscopic bending characteristics, while fractal dimension reflects macroscopic entanglement characteristics. Through feature normalization, information entropy weighting, and multi-feature fusion, these geometric parameters of different dimensions are transformed into an intuitive and unified evaluation index for root entanglement. This solves the technical problem that traditional single features cannot comprehensively characterize the severity of root entanglement. That is, a single feature can only reflect the characteristics of one dimension of root entanglement. For example, fractal dimension alone cannot accurately identify the microscopic bending of mild root entanglement, and average curvature alone cannot quantify the macroscopic entanglement of severe root entanglement. Dual-feature weighted fusion can achieve a comprehensive quantitative evaluation of root entanglement phenomena. The operation involves feature normalization and weighted summation. The formula construction logic is based on information entropy theory. By calculating the information entropy of each feature, its ability to distinguish root entanglement phenomena is quantified, and then the weight coefficient is determined. This allows features with higher sensitivity to root entanglement to occupy a higher proportion in the evaluation, improving the accuracy of the evaluation index.

[0051] Specifically, a root system compressibility evaluation model is constructed based on this sample database. First, the average curvature of the root system calculated in step S2 is evaluated. and fractal dimension Minimum-maximum value normalization is performed to address the issue of direct fusion caused by the different dimensions and large differences in numerical ranges of the two types of features. Specific operations include: firstly, constructing a root system sample library containing different tree species, seedling ages, and different degrees of root entanglement (normal samples, mildly entangled samples, and severely entangled samples), setting the sample size to [number missing]. For example, setting To ensure sample coverage of the entire root system scenario, the maximum and minimum values ​​of fractal dimension and root mean curvature were statistically obtained from the sample database. This provides a reliable extreme value reference for feature normalization. Based on this, the fractal dimension and root mean curvature are uniformly mapped to... The interval is used to obtain the normalized fractal dimension. and normalized mean curvature .

[0052] Furthermore, based on information entropy theory, the weight coefficients of the two types of normalized features are calculated to highlight the influence weight of features highly sensitive to the rooting phenomenon. First, the weight coefficients of the first type of normalized features are calculated. The information entropy of a feature is expressed as follows:

[0053] ;

[0054] In the formula, It is the first Information entropy of features; This refers to the number of samples in the seedling root system sample bank. It is an index of samples in the seedling root sample bank; For the first The sample at the th The proportion of the normalized value of a feature to the total sum of the normalized values ​​of that feature in all samples; It is an index of features. Represents the normalized fractal dimension feature, Represents the normalized mean curvature ; Based on the natural constant Logarithmic function with base 0; information entropy The range of values ​​is Information entropy The closer the value is to 1, the more evenly the feature is distributed across different samples, and the weaker its ability to distinguish the degree of root entropy; Information entropy The closer the value is to 0, the more significant the difference in this feature is across samples with different degrees of root penetration, and the stronger its discriminative power.

[0055] Specifically, through the coefficient of difference Quantify feature discrimination, set Coefficient of difference The larger the number, the higher the number of... The higher the discriminative power of the features, the higher the weights of the two types of features are obtained by normalizing the difference coefficients. The relation is: ;when Time corresponding weight ,when Time corresponding weight ,and Finally, a root-binding measurement model was constructed based on weighting coefficients to obtain the root-binding degree evaluation index. The relationship is as follows:

[0056] ;

[0057] In the formula, It is a root strength evaluation index, with a value range of [value range missing]. , fractal dimension after normalization The weight, Normalized mean curvature The weights of the formula are as follows: the normalized fractal dimension describes the macroscopic complexity of the root system's entanglement, and the normalized average curvature describes the microscopic severity of the root system's bending. The weights determined by information entropy organically integrate the two, achieving a comprehensive quantitative evaluation of the root entanglement phenomenon from both macroscopic and microscopic perspectives. The closer the value of the root entanglement evaluation index is to 1, the more severe the root entanglement phenomenon is; the closer it is to 0, the more normal the root system's growth state is.

[0058] S4. Based on the root coil evaluation index, preset classification thresholds are used to achieve graded early warning for the identification of root coil in seedlings.

[0059] It should be noted that the overall logic of this step is to apply the root strength evaluation index obtained in step S3. The system transforms the data into discrete hierarchical decision signals, addressing the issue that continuous values ​​cannot directly guide automated sorting on production lines. Simultaneously, a confidence verification mechanism handles ambiguous samples at threshold boundaries, balancing automation efficiency and detection accuracy. The operation involves threshold decision-making and signal mapping, with the formula constructed based on the statistical distribution of the root-root index in the sample library. Combined with ROC curve analysis, the optimal classification threshold is determined, achieving the best balance between false negatives and false positives. This ensures the reliability and practicality of the grading results, ultimately providing clear control signals for production line sorting operations.

[0060] Specifically, firstly, based on the seedling root system sample bank constructed in step S3, the root resilientness evaluation index is statistically analyzed. By analyzing the distribution characteristics of samples at different plaque levels (normal, mild, and severe plaque) and combining this with ROC curve analysis to determine the optimal classification threshold, the Youden exponent maximization principle of the ROC curve can be used to screen for the mild plaque threshold that best balances the false negative and false positive rates. With severe enlargement threshold The principle of maximizing the Yoden exponent is a well-known technique and will not be elaborated upon here; for example, setting... This corresponds to the dividing point between normal samples and mildly congested root samples. This threshold combination, corresponding to the dividing point between mild and severe infestation samples, has been experimentally verified to result in both a low overall false alarm rate and a low false alarm rate. A classification decision function is introduced. Identify and classify the root system status of seedlings, and use root resilient strength evaluation indicators. The corresponding hierarchical signal is mapped to the following piecewise function relationship: ,like If the sample is normal, it can be directly removed from the nursery. ,like If the sample is found to have mild root entanglement, it needs to be re-examined for root spread. ,like If the sample is classified as severely entrenched, it will be directly eliminated; where, the classification judgment function is... The output is a hierarchical signal. The root compaction evaluation index obtained in step S3 is... For mild encirclement threshold, This is the threshold for severe root entanglement.

[0061] To further improve the reliability of judgments and avoid the influence of entrenched judgment indicators To address misjudgments in grading due to proximity to thresholds, a confidence level verification mechanism is introduced. A confidence level is defined to quantify the reliability of the grading results. The judgment logic involves first determining the root strength evaluation index. The nearest classification threshold The specific operations include: If In 0 to Between, that is, 0 to , Between the midpoints, the nearest classification threshold is ;like In Between 1 and 1, that is , If the midpoint is between 1 and 0, then the nearest classification threshold is... Then through calculation The confidence level is determined by the absolute distance from the nearest classification threshold, as shown in the following formula:

[0062] ;

[0063] In the formula, It is the confidence level; It is an indicator for evaluating the degree of entanglement; It is a root strength evaluation index The nearest classification threshold; For mild encirclement threshold, The threshold for severe encirclement; It is the absolute value symbol; when absolute distance The larger it is, the more it indicates The more clearly defined the classification category, the higher the confidence level and the more reliable the classification result; when the absolute distance The smaller the size, the more it indicates At the boundary between the two classification categories, the lower the confidence level, the higher the uncertainty of the classification result.

[0064] Specifically, a pre-set reliability threshold The example setting is 0.1, which can be flexibly adjusted according to the accuracy requirements of the production line; when the calculated confidence level is lower than the confidence level threshold... At that time, the system automatically outputs an early warning signal, prompting staff to manually verify the seedling to ensure the accuracy of the boundary sample grading; finally, the system makes a classification judgment based on the classification function. The output value controls the corresponding actuator, such as a conveyor belt distributor, to transport seedlings with different grading results to their respective processing areas: Classification Judgment Function When the output is 0, the seedlings are transported to the normal seedling exit channel; when the output is 1, they are transported to the mild root entanglement re-inspection channel; when the output is 2, they are transported to the severe root entanglement elimination channel; seedlings that trigger the warning signal are transported separately to the manual verification channel, thus completing the entire process of automatic identification, grading and sorting of seedling root entanglement.

[0065] Furthermore, based on the above operations, the root strength evaluation index for each sample can be obtained. Please refer to [link to relevant documentation]. Figure 2 As shown, Figure 2 This is a distribution statistics chart of the seedling root system reticulation identification method based on image processing in this invention. As shown in the chart, the root reticulation evaluation index for seedling samples with different degrees of reticulation exhibits a clear interval distribution characteristic: the root reticulation evaluation index for normal samples is mainly concentrated below 0.4, the root reticulation evaluation index for slightly reticulated samples is concentrated between 0.4 and 0.7, and the root reticulation evaluation index for severely reticulated samples is mostly above 0.7. The overlap of the root reticulation evaluation index distribution among the three types of samples is extremely low, indicating that the root reticulation evaluation index of this invention can effectively distinguish seedlings with different degrees of reticulation. Meanwhile, the current sample's root reticulation evaluation index is 0.319, which is within the distribution range of normal samples. Combined with the division of slightly reticulated and severely reticulated thresholds, it can be clearly determined that it is a normal sample. This further verifies the rationality and discriminative power of the root reticulation evaluation index and classification threshold in this method, providing accurate and reliable quantitative basis for subsequent automated grading of seedlings.

[0066] Furthermore, a comparison chart of the detection methods of this method and existing methods can be obtained; please refer to [link / reference]. Figure 3 As shown, Figure 3 This is a comparison of the detection methods of the seedling root system reticulation identification method based on image processing in this invention with existing methods. As shown in the figure, the detection performance indicators of this method are significantly better than those of traditional detection methods: the accuracy, precision, recall, and F1 score of traditional edge detection methods, traditional projected area methods, and traditional Hough transform are all in the range of 65-78%, while the accuracy of this method reaches 95%, and the precision, recall, and F1 score are all over 93%, with each indicator improving by more than 20% compared with traditional methods. This shows that this method effectively overcomes the defects of traditional methods in complex soil backgrounds and diverse root morphology scenarios, such as low recognition accuracy and poor stability, and has better accuracy and robustness in reticulation identification, fully verifying the practicality and advancement of the technical solution of this invention.

[0067] This invention also discloses a seedling root system identification system based on image processing, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a seedling root system identification method based on image processing according to the present invention. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art and will not be described in detail here.

Claims

1. A method for identifying root system infestation in seedlings based on image processing, characterized in that, include: Collect original images of the seedlings to be tested; The original image is preprocessed to calculate the optimal segmentation threshold, which is used to obtain the root system region of interest. The region of interest of the root system is refined to obtain a root system skeleton with a width of one pixel; The support radius of each point on the root system skeleton is calculated using the distance transformation method, including: calculating the support radius of the first point on the root system skeleton using the distance transformation method. The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the edge of the root system That is, the first The support radius of each skeleton point; if it appears In the case of extreme fine branches, the average of the support radii of three skeleton points (one adjacent to the front and one to the rear) is taken as the first... The support radius of each skeleton point is defined and calculated based on differential geometry theory, including: In the formula, It is the first Discrete curvature of each skeleton point, in radians per pixel; It is the first Forward vectors of each skeleton point; It is the first The backward vector of each skeleton point; This represents the vector dot product operation; Represents the magnitude of the vector; It is the inverse cosine function; For the first The skeleton points are connected to the preprocessed root system region of interest. The shortest Euclidean distance at the root system edge is used to calculate the average root system curvature by taking the arithmetic mean of all discrete curvatures; the fractal dimension of the root system skeleton is calculated using box counting, including: In the formula, It is the fractal dimension of the root system; To cover the root system region of interest The grid side length; Complete coverage of the root system skeleton at each scale Minimum number of grid cells required; It is a logarithmic function with base 10; For limit operations; using the least squares method Perform linear regression, setting the goodness of fit to be greater than or equal to 0.

98. The absolute value of the slope of the regression line is the fractal dimension of the root system. , The value ranges from 1.2 to 1.8; A root system sample library containing different tree species, seedling ages, and root entanglement degrees is constructed. Based on this sample library, the average curvature and fractal dimension of the root system are normalized using minimum and maximum values. Weight coefficients of the two normalized features are calculated based on information entropy theory, and a root entanglement degree evaluation index is obtained through weighted summation. Based on the statistical distribution characteristics of the sample library, and combined with ROC curve analysis, thresholds for mild and severe root entanglement are determined. The root entanglement degree evaluation index is mapped to a grading signal using a classification judgment function, while a confidence verification mechanism is introduced to process threshold boundary samples. Based on the grading signal and the confidence verification results, the execution mechanism is controlled to complete the automated sorting and manual verification triggering of seedlings.

2. The method for identifying root systems and basalt patterns in seedlings based on image processing according to claim 1, characterized in that, The preprocessing of the original image includes: ； In the formula, It is the enhanced coordinates The brightness value at that location, It is a Gaussian enclosed function and satisfies , A normalization constant to ensure that the integral of the Gaussian encirclement function is 1; For the first Standard deviation of each scale; It is a logarithmic function with base 10; This represents the convolution operation. The optimal number of Gaussian scales was determined through experiments to be suitable for seedling root system scenarios.

3. The method for identifying root systems and basalt patterns in seedlings based on image processing according to claim 1, characterized in that, The calculation of the optimal segmentation threshold includes: ； In the formula, It is the optimal segmentation threshold; Indicates in set In, the objective function To obtain the maximum value The value; The gray level to be traversed, with a value range from 0 to 255; , The segmentation thresholds are respectively The percentage of pixels in the foreground (root system) and background (soil); , These are the average grayscale values ​​of the foreground and background, respectively. This represents the average grayscale value of the enhanced grayscale image.

4. The method for identifying root systems and basalt patterns in seedlings based on image processing according to claim 1, characterized in that, The weight coefficients for calculating the two types of normalized features based on information entropy theory include: ； In the formula, It is the first Information entropy of features; This refers to the number of samples in the seedling root system sample bank. It is an index of samples in the seedling root sample bank; For the first The sample at the th The proportion of the normalized value of a feature to the total sum of the normalized values ​​of that feature in all samples; It is an index of features. Represents the normalized fractal dimension feature, Represents the normalized mean curvature ; Based on the natural constant Logarithmic function with base 0; information entropy The range of values ​​is .

5. The method for identifying root systems and basalt patterns in seedlings based on image processing according to claim 1, characterized in that, The calculation of the weight coefficients for the two types of normalized features based on information entropy theory also includes: Through the coefficient of difference Quantify feature discrimination, set The difference coefficients are normalized to obtain the weights of the two types of features. The relation is: ;when Time corresponding weight ,when Time corresponding weight ,and .

6. The method for identifying root systems and basalt patterns in seedlings based on image processing according to claim 1, characterized in that, The method for obtaining the confidence level includes: ； In the formula, It is the confidence level; It is an indicator for evaluating the degree of entanglement; It is a root strength evaluation index The nearest classification threshold; For mild encirclement threshold, The threshold for severe encirclement; It is the absolute value symbol; if In 0 to Between, that is, 0 to , Between the midpoints, the nearest classification threshold is ;like In Between 1 and 1, that is , If the midpoint is between 1 and 0, then the nearest classification threshold is... .

7. A seedling root system identification system based on image processing, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a seedling root system identification method based on image processing according to any one of claims 1-6.

Images