A method and apparatus for identifying resistance to soil-borne diseases in vegetable seedlings
By acquiring orthogonal two-dimensional projection images of soil-borne disease resistance identification in vegetable seedlings, calculating the total root area and controlling the cutting of the root system, and combining quantitative hydration and disease index, the problems of individual differences and water stress in existing technologies have been solved, thus improving the accuracy and repeatability of resistance identification.
Patent Information
- Application Number
- CN202611343399.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-01
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for identifying resistance to soil-borne diseases in vegetable seedlings are difficult to quantitatively control the wound area and pathogen exposure of individual plants, leading to individual differences. Furthermore, they do not consider the differences in water absorption capacity of the remaining root system, introducing water stress variables. Current evaluations rely on a single canopy phenotype and ignore the initial root mass, resulting in low accuracy and repeatability of the results.
By acquiring orthogonal two-dimensional projection images of the hanging roots of the target seedlings, the total effective root area is calculated. The shearing blade is controlled to move to the dynamic cutting point to cut off the roots. The target coordinates of the gripping robot are adjusted in combination with the residual effective root area parameters to perform quantitative fluid replenishment. The comprehensive resistance evaluation index is calculated in combination with the disease index after the disease cycle.
It enables uniform control of the wound area of different seedlings, avoids water stress, improves the accuracy and objectivity of resistance identification, and truly reflects the pathological resistance level of plant varieties.
Smart Images

Figure CN123100408A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of agricultural plant protection and breeding, specifically to a method and apparatus for identifying resistance to soil-borne diseases in vegetable seedlings. Background Technology
[0002] Soil-borne diseases are a major factor affecting vegetable yield. Accurate disease resistance identification during the seedling stage plays a crucial role in screening disease-resistant vegetable varieties and controlling diseases. Currently, conventional methods for identifying soil-borne disease resistance in vegetable seedlings typically involve manual root cutting inoculation or direct irrigation inoculation. During inoculation, it is difficult to quantitatively control the contact area between the wound area of a single plant and the amount of pathogen suspension, leading to individual differences in the actual disease infection pressure experienced by plants at different developmental stages. After inoculation, the plants enter the disease culture stage, and most identification methods employ a uniform, timed irrigation pattern. This water management approach does not consider the individual differences in the water absorption capacity of the remaining root system after root cutting treatment.
[0003] A uniform amount of liquid replenishment can easily lead to excessively high substrate moisture content in seedlings with weak initial root systems, causing hypoxia, or insufficient water supply to seedlings with well-developed root systems, thus introducing non-pathological water stress variables into the disease identification process.
[0004] Furthermore, existing resistance evaluation mechanisms mainly rely on the disease index calculated from the plant canopy phenotype in the later stages of disease, without taking the initial developmental size of the plant's root system as a reference factor. Plants with larger initial biomass often exhibit a stronger ability to maintain their phenotype in the early stages of pathogen infection. If only the single canopy disease index in the later stages of disease is used for evaluation, it is easy to mask the true physiological disease resistance level of the target plant, ultimately leading to low accuracy and repeatability of resistance identification results.
[0005] Therefore, this invention proposes a method and apparatus for identifying resistance to soil-borne diseases in vegetable seedlings to address the shortcomings of existing technologies. Summary of the Invention
[0006] To address the shortcomings of existing technologies, this invention provides a method and apparatus for identifying resistance to soil-borne diseases in vegetable seedlings. This method solves the problems of existing resistance identification methods for soil-borne diseases in vegetable seedlings, which are difficult to quantitatively control the wound area and pathogen contact amount of individual plants, leading to individual differences in inoculation pressure; do not consider the differences in water absorption capacity of residual roots, which easily introduces water stress variables; and rely on a single canopy phenotype to ignore the initial root mass, resulting in low accuracy and repeatability of the evaluation results.
[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying resistance to soil-borne diseases in vegetable seedlings includes the following steps: Obtain orthogonal two-dimensional projection images of the hanging root system of the target seedling, extract the effective outline of the root system, and calculate the total area of the comprehensive effective root system using the root mean square model; The target wound area is calculated based on the total effective root area and the vertical coordinate of the dynamic cutting point. The shearing tool is controlled to move to the execution height corresponding to the vertical coordinate of the dynamic cutting point to perform root truncation. The residual effective root area parameter is calculated and associated with the RFID tag identification code of the culture unit carrying the target seedling. The immersion inoculation was performed by adjusting the target coordinates of the gripping robot arm based on the residual effective root area parameter. Read the RFID tag identification code, call back the associated residual effective root area parameters, calculate the target micro-irrigation amount, and perform quantitative liquid replenishment; The disease index is extracted after the end of the disease cycle, and the comprehensive resistance evaluation index is calculated and output by combining the residual effective root area parameter with the disease index.
[0008] Preferably, the effective contour of the root system is extracted, including: For each pixel of an orthogonal two-dimensional projected image, the second-order partial derivatives are calculated to construct a second-order partial derivative matrix, and the first and second eigenvalues are solved. The ratio of the first feature value to the second feature value is calculated as the morphological identification factor. When the absolute value of the morphological identification factor is less than the tubular threshold constant and the absolute value of the second feature value is greater than the background noise threshold constant, the current pixel is determined to belong to the tubular root tissue, and the pixel domain of tubular features is retained. The tubular threshold constant is determined by the prior statistical distribution of the aspect ratio of the target plant root system, and the background noise threshold constant is calibrated based on the basic dark current noise of the environment in which the visual acquisition mechanism that acquires the orthogonal two-dimensional projection image is located. The Otsu's method is applied to the pixel domain of the tubular feature to perform global threshold segmentation and generate a binary mask matrix as the effective contour of the root system.
[0009] Preferably, the total effective root area is calculated using the root mean square model, including: Calculate the effective pixel area of the binarized mask matrix in each orthogonal direction; The total effective root area is obtained by taking the square root of the sum of the squares of the areas of each effective pixel. When the total effective root area is greater than the minimum survival area threshold, it is temporarily stored in the internal cache as a quantitative benchmark. The minimum survival area threshold is determined based on the historical lower limit of the survival area of the corresponding plant variety in the current growth period.
[0010] Preferably, the calculation of the target wound area and the vertical coordinates of the dynamic cutting point based on the total effective root area includes: The target wound area is obtained by multiplying the total effective root area by the equivalent inoculation pressure coefficient. Extract the effective pixel width from the effective contour of the root system to construct a comprehensive vertical pixel distribution function. Take the vertical coordinate corresponding to the bottom of the hanging root system as the starting optimization position and perform discrete accumulation and summation operation from bottom to top along the vertical direction, that is, accumulate the effective pixel width row by row. When the accumulated area value is equal to or just exceeds the target wound area, record the current vertical coordinate as the vertical coordinate of the dynamic cutting point.
[0011] Preferably, the immersion inoculation is performed by adjusting the target coordinates of the gripping robot based on the residual effective root area parameter, including: The residual effective root area parameter is combined with the allometric growth index and the three-dimensional mapping coefficient to perform a power function operation to solve for the residual root volume. Divide the volume of the remaining root system by the effective internal bottom area of the root soaking container to obtain the height of the liquid level rise. Subtract the height of the liquid bulge from the reference coordinates of the liquid surface and add the immersion depth of the wound infection to obtain the target coordinates of the gripping robot for descent.
[0012] Preferably, the process of reading the RFID tag identification code and calling back the associated residual effective root area parameter to calculate the target micro-irrigation amount and perform quantitative liquid replenishment includes: The target micro-irrigation amount is calculated by multiplying the residual effective root area parameter by the unit root water absorption coefficient and adding the substrate base evaporation compensation amount. The target micro-irrigation volume is checked for upper and lower limits. When the calculated target micro-irrigation volume is lower than the set minimum water volume for survival, a forced assignment process is triggered, and the target micro-irrigation volume is forcibly assigned to the minimum water volume for survival. When the target micro-irrigation volume is higher than the maximum water holding capacity of the cultivation substrate, the target micro-irrigation volume is limited to the maximum water holding capacity.
[0013] The minimum water requirement for survival is determined by the critical water content at which plants wilt, while the maximum water holding capacity is determined by the porosity and volume characteristics of the cultivation substrate. The final target micro-irrigation volume is converted into a pulse width modulation signal to control the irrigation valve and the precision fluid pump to perform quantitative liquid replenishment to the cultivation unit.
[0014] Preferably, the disease index extracted after the end of the disease cycle includes: The region of interest (ROI) is cropped and normalized on the canopy digital image matrix acquired after the onset of the disease cycle to generate the input image matrix; The disease semantic segmentation neural network model is invoked to process the input image matrix and output a two-dimensional category label matrix; The disease index is forcibly assigned a value of one hundred when the sum of the diseased and healthy canopy pixel areas in the two-dimensional category label matrix is calculated.
[0015] When the sum of the area of diseased canopy pixels and the area of healthy canopy pixels is greater than or equal to the minimum lower limit of canopy area, the ratio of the area of diseased canopy pixels to the sum of the two areas is calculated and multiplied by one hundred to obtain the disease index. The minimum lower limit of canopy area is determined based on the field of view of the visual camera that acquires the canopy digital image matrix and the remaining cross-sectional area of the plant after natural wilting.
[0016] Preferably, the disease semantic segmentation neural network model adopts a fully convolutional network architecture, including: The encoder module, feature splicing module, and decoder module are cascaded together in sequence. Each discrete pixel in the two-dimensional category label matrix corresponds to a background region with a value of zero, a healthy canopy region with a value of one, and a diseased, yellowed, and withered region with a value of two.
[0017] Preferably, a comprehensive resistance evaluation index is calculated and output by combining the residual effective root area parameter and the disease index, including: Calculate the difference between the residual effective root area parameter and the statistical average of the residual effective root area parameter of the same batch of identified samples, divide the difference by the statistical average, multiply the result by the root volume compensation weight coefficient, and then add it to the value. Multiply the result of subtracting the disease index from the value of 100 by the sum of the above results to obtain the comprehensive resistance evaluation index.
[0018] This invention also provides a device for identifying resistance to soil-borne diseases in vegetable seedlings, comprising: The shifting mechanism is equipped with a culture unit, and the outer wall of the culture unit is fitted with an RFID tag containing a unique identification code; It is equipped with a linear air knife assembly and a pneumatic pressure regulating valve arranged in a vertical direction, which is used to blow off the attached objects and remove impurities. The visual acquisition mechanism for acquiring orthogonal two-dimensional projection images is equipped with a first camera and a second camera arranged horizontally at a 90-degree angle, and is equipped with a photoelectric switch and a backlight. The inoculation mechanism for performing root cutting and immersion inoculation is equipped with a shearing cutter with a linear servo drive module, a gripping manipulator, and an immersion container with a liquid level limit structure. The irrigation mechanism for performing quantitative replenishment is equipped with a reader, a precision fluid pump, and an irrigation valve; A disease monitoring facility equipped with color industrial cameras for acquiring digital image matrices of the canopy; A control system that establishes bidirectional data communication with the displacement mechanism, cleaning mechanism, visual acquisition mechanism, inoculation mechanism, irrigation mechanism, and disease monitoring mechanism via an industrial bus and has an internal storage unit.
[0019] This invention provides a method and apparatus for identifying resistance to soil-borne diseases in vegetable seedlings. It has the following beneficial effects: 1. This invention calculates the total effective root area by acquiring orthogonal projection images of the root system, and calculates dynamic cutting points by accumulating the effective pixel width row by row along the vertical direction. It controls the shearing blade to perform truncation at the corresponding height, so that the proportion of the wound area to the total root area of different seedlings remains consistent. This eliminates the error caused by the difference in the initial development size of the plant and ensures the uniformity of the inoculation standard of each plant.
[0020] 2. This invention achieves accurate variable control during the inoculation and disease development stages. During inoculation, the system calculates the height of the liquid surface rise after immersion in the liquid based on the residual effective root area, and adjusts the robot's downward coordinates accordingly to ensure that the liquid immersion depth on the wound surface of each plant is the same. During disease development, the system calculates the micro-irrigation amount based on the actual residual root area of each plant and performs upper and lower limit threshold checks, avoiding root hypoxia in weak seedlings or water stress in strong seedlings caused by uniform quantitative irrigation, and eliminating the interference of water variables on the disease development process.
[0021] 3. This invention improves the accuracy and objectivity of comprehensive resistance evaluation results. Based on the extraction of the canopy disease index in the later stage of disease, the residual effective root area parameter is used as an evaluation correction factor in the calculation. By comparing the difference between the individual residual root area and the batch statistical average, a volume compensation weight is constructed, which corrects the lag error in phenotypic changes shown by plants with large initial biomass in the early stage of disease, and truly reflects the pathological resistance level of the target plant variety. Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the structure of the vegetable seedling soil-borne disease resistance identification device of the present invention; Figure 2 This is a flowchart of the identification method of the present invention; Figure 3 This is a flowchart of the plant pretreatment and comprehensive root area extraction process of the present invention; Figure 4 This is a flowchart of the adaptive cutting inoculation and parameter feedforward micro-irrigation control of the present invention; Figure 5 This is a flowchart of the disease data collection and comprehensive resistance evaluation process of the present invention; Figure 6 This is a line graph illustrating the dynamic cutting optimization logic of the present invention. Figure 7 This is a line graph showing the distribution of the comprehensive resistance evaluation index of the present invention. Figure 8 This is a perspective view of the present invention.
[0023] Among them, 1. Transfer mechanism; 2. Culture unit; 3. Impurity removal mechanism; 4. Visual acquisition mechanism; 5. Inoculation mechanism; 6. Irrigation mechanism. Detailed Implementation
[0024] The technical solutions in 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.
[0025] See attached document Figure 1 and Figure 8 The present invention provides a device for identifying resistance to soil-borne diseases in vegetable seedlings, which may include: a transfer mechanism 1, a culture unit 2, a cleaning mechanism 3, a visual acquisition mechanism 4, an inoculation mechanism 5, an irrigation mechanism 6, and a control system.
[0026] The shifting mechanism 1 constitutes the main operating backbone of the identification device. Based on the specific mechanical structure shown in the three-dimensional diagram of the device, the shifting mechanism 1 includes a vertically arranged main support backplate, with a rotating crank disc and a multi-link swing arm transmission mechanism mounted on its front. Multiple culture units 2 are arranged linearly in a row, specifically resembling square open-mouthed cup-shaped containers, mounted on a horizontal support rail or bracket at the top of the shifting mechanism 1. Each culture unit 2 has an independent structure and is equipped with a unique RFID tag on its outer wall. The shifting mechanism 1 drives the transmission rail continuously or in steps according to a set periodic frequency via the mechanical transmission of the crank and connecting rod, causing the mounted culture units 2 to perform repositioning along a closed-loop trajectory within the indoor space.
[0027] The impurity removal mechanism 3 is located at the downstream node of the removal process of the cultivation unit 2. A square guide funnel is fixed on the upper part of the impurity removal mechanism 3, and a V-shaped opening and closing guide hood is assembled on the lower part. It has a duckbill-shaped opening and closing structure. The two sides of the V-shaped opening and closing guide hood are connected to the external inclined hinged connecting rod. The linear air knife assembly arranged in the vertical direction is concealed inside the V-shaped opening and closing guide hood. When the plant roots enter through the guide funnel, the connecting rod drives the V-shaped opening and closing guide hood to close and release the advection air curtain in the sealed cavity to prevent water droplets or substrate splashing during impurity removal.
[0028] The visual acquisition mechanism 4 is located at the end of the process of the impurity removal mechanism 3. As shown in the figure, the external shape of the visual acquisition mechanism 4 is a rectangular box housing. A cylindrical light-transmitting acquisition window or lens cover is provided on its front side. Inside, a first camera and a second camera are arranged horizontally at a 90-degree angle, and a corresponding backlight is provided on the opposite side of the first camera and the second camera.
[0029] The inoculation mechanism 5 is located adjacent to the vision acquisition mechanism 4. The inoculation mechanism 5 is equipped with a shearing blade with a linear servo drive module and an immersion container with a liquid level limiting structure. Referring to the specific modular features in the attached diagram, the main body of the inoculation mechanism 5 is rectangular, with a manual adjustment knob with anti-slip texture at its rear end for mechanical zero-point calibration or fine-tuning of the immersion depth. Its side has a row of circular sensor quick-connect holes or status indicator holes. The top of the inoculation mechanism 5 is pivotally hinged to a robust rocker arm extending from the displacement mechanism 1. The vertical swing of the link, in conjunction with the internal linear module, achieves the downward movement of shearing and immersion inoculation.
[0030] The irrigation mechanism 6 is installed at the preset position of the closed-loop transmission trajectory of the displacement mechanism 1, as shown in the figure. The irrigation mechanism 6 specifically includes a long strip-shaped fluid distribution manifold block that spans and is suspended above the culture unit 2. It is obliquely fixed to the main frame by the support rods on both sides. The bottom surface of the fluid distribution manifold block is equipped with multiple downward-sloping nozzles arranged in a row, such as oblique atomizing nozzles or micro-droplet arrows as shown in the figure, which are used to perform precise liquid replenishment to the square culture unit 2 passing below. The irrigation mechanism 6 is also equipped with a reader and irrigation valve connected to the control system.
[0031] The control system establishes a bidirectional data communication connection with the shifting mechanism 1, the impurity removal mechanism 3, the vision acquisition mechanism 4, the inoculation mechanism 5, and the irrigation mechanism 6 via an industrial bus. The control system has an internal storage unit for recording the identification codes corresponding to the RFID tags and various root morphology parameters calculated in real time during the process.
[0032] See attached document Figure 2 This invention provides a method for identifying resistance to soil-borne diseases in vegetable seedlings, comprising the following steps: S10, the seedling material to be identified is cultivated inside the cultivation unit 2, and spatially repositioned by the transfer mechanism 1 to balance the environmental conditions during the plant's growth period. When the plant develops to the set inoculation physiological stage, the target seedling, along with its roots, is removed from the cultivation unit 2 and rinsed off the cultivation substrate attached to the root surface by a low-pressure water flow. After rinsing, the seedlings, with their roots hanging naturally, pass through the impurity removal mechanism 3. The linear air knife assembly releases a micro-pressure advection air curtain in the vertical direction, blowing off free water droplets and loose substrate particles attached to the root hairs. S20, after impurity removal, the seedlings are conveyed into the vision acquisition mechanism 4. The first and second cameras, with the assistance of a backlight, synchronously acquire orthogonal two-dimensional projection images of the overhanging root system. The control system performs feature enhancement and binarization segmentation on the orthogonal two-dimensional projection images, extracts the effective contour of the root system, and then calculates the comprehensive effective total root area that eliminates single-view occlusion errors. S30: The control system calculates the target wound area based on the total effective root area and the preset equivalent inoculation pressure coefficient, and constructs a longitudinal pixel distribution function along the vertical axis. The corresponding dynamic cutting point coordinates are determined through one-dimensional integral calculation. The control system outputs pulse control commands to drive the shearing blade of the inoculation mechanism 5 to move to the corresponding height to perform root severance. After severance, the control system synchronously calculates the residual effective root area parameter of the seedling and associates this parameter with the RFID tag identification code of the corresponding culture unit 2, storing it in the control system's storage unit. S40, the inoculation mechanism 5 dynamically adjusts the downward immersion target coordinate of the gripping robot in the root soaking container based on the calculated residual effective root area parameters, and performs standardized quantitative root soaking within a predetermined time. After inoculation, the seedlings are returned to the original culture unit 2 and reattached to the transfer track of the transfer mechanism 1 to continue operation. When culture unit 2 reaches the irrigation mechanism 6, the reader reads the identification code, controls the system to recall the associated residual effective root area parameters and calculate the feedforward water compensation volume, generating a pulse width modulation signal to control the irrigation valve to perform quantitative liquid replenishment to culture unit 2. After a set pathogen incubation and disease development cycle, S50 collects phenotypic and physiological data of plants in each culture unit 2 under a unified environmental standard. The obtained disease indicators such as yellowing rate and wilting index are compared with the baseline disease threshold of the standard control group, and the quantitative evaluation results of the resistance level against the targeted soil-borne disease are output.
[0033] See attached document Figure 3 Regarding the transfer and pretreatment operations during the seedling stage and the pre-inoculation stage, the specific implementation of the above step S10 includes the following process.
[0034] S11, the seedling material to be identified is planted in cultivation unit 2. Cultivation unit 2 is equipped with a detachable inner seedling cylinder and an outer protective sleeve. The inner seedling cylinder supports the cultivation substrate and plant roots. The outer protective sleeve provides mechanical support and isolation. An RFID tag is fixedly installed on the outer wall of cultivation unit 2. The control system generates a unique corresponding identification code and writes the code into the RFID tag using an external RFID writing device, establishing a data mapping relationship between the individual plant and the container, facilitating data traceability and status retrieval for individual plants in subsequent processes.
[0035] S12, the assembled cultivation unit 2 is placed onto the transfer track of the shifting mechanism 1 for cultivation. The control system sends a control signal to the shifting mechanism 1, driving the transfer track to run continuously or in steps, causing the carried cultivation unit 2 to change position along a closed-loop trajectory. From an engineering perspective, due to the influence of light source angle, heat source distribution, and ventilation vent location within the greenhouse, there is an unavoidable spatial gradient distribution of temperature, humidity, and light. Through the dynamic flow of this step, the differences caused by spatial location can be evenly distributed among the seedling materials. Specifically, the cumulative environmental parameters acquired by the cultivation unit 2 throughout the entire cultivation cycle satisfy the following integral relationship: ; In the formula, Represents the cumulative environmental parameters obtained from a single culture unit 2; This represents the total duration of the identification process; Represents the current running time; The sorting number representing culture unit 2; The spatial coordinates representing the change of a specific culture unit 2 over time; This represents the environmental distribution function within the greenhouse space. Through the continuous operation of the shifting mechanism 1, the spatial positions of different cultivation units 2 traverse the same area in the time dimension, making the expected values of environmental resources acquired by each cultivation unit 2 within the same time period tend to be consistent, thereby effectively reducing the interference of environmental gradient differences existing in traditional static seed trays on subsequent disease phenotypic data.
[0036] For the specific mechanical transmission structure of the shifting mechanism 1, those skilled in the art can use a servo motor to drive the guide rail or a chain transmission assembly. Its underlying mechanical structure is a well-known technology in the field and will not be described in detail here.
[0037] S13, When the target seedling material reaches the inoculation physiological stage, the inner seedling tube, along with the seedling, is pulled upwards to detach from the outer protective sleeve. The root system is rinsed with low-pressure water to remove the cultivation substrate covering the roots. After rinsing, the root collar of the seedling is secured using a suspension clamping assembly on the transport track. Specifically, the suspension clamping assembly can be positioned from below the cotyledon node to above the primary root at the hypocotyl, providing vertical support without damaging the plant's vascular tissue. At this point, the lower roots of the seedling naturally droop under gravity, forming a suspended posture.
[0038] S14, the seedling roots, maintaining a suspended posture, enter the cleaning mechanism 3 along the transport track. The cleaning mechanism 3 is equipped with a linear air knife assembly arranged parallel to the vertical direction and a pneumatic pressure regulating valve. The control system activates the air source, causing the linear air knife assembly to release a horizontal air curtain along the vertically downward direction of the suspended roots. The horizontal air curtain is applied to the root surface, using the kinetic energy of the gas to blow away free water droplets and residual loose matrix particles attached to the root hair tissue.
[0039] The control system limits the output pressure of the pneumatic pressure regulating valve to below the mechanical yield strength of the plant root tissue. In actual parameter settings, this pressure threshold can be calibrated through preliminary tensile tests, and its value is typically within a range sufficient to overcome the surface tension of adhering water droplets and below the critical stress for lateral root tearing. The aforementioned impurity removal process helps to eliminate morphological interferences in visual image acquisition to the greatest extent possible, providing a relatively pure target for subsequent calculation of the total effective root area.
[0040] See attached document Figure 3 For the visual feature extraction and area calculation of overhanging root systems, the specific implementation of step S20 above includes the following process.
[0041] S21, the seedlings processed by the impurity removal mechanism 3 enter the visual acquisition mechanism 4 along the transport track. The seedling roots are in a free-hanging state. When the photoelectric switch inside the visual acquisition mechanism 4 detects that the seedling has reached the central acquisition position, it sends a trigger level to the control system.
[0042] The control system synchronously controls the first and second cameras, which are arranged horizontally at a 90-degree angle, to perform short exposures. With the help of the backlight turned on on the opposite side, the system acquires the original X-axis two-dimensional projection image and the original Y-axis two-dimensional projection image of the hanging root system of the same plant.
[0043] A parallel backlight transmission imaging method is employed, utilizing the property that strong light penetrates the thinner root edges and forms high-contrast shadows in the main root area. This effectively suppresses interference from greenhouse background stray light on root contour extraction during the raw image stage. The control system records the pixel matrices of these two sets of raw images as follows: and ,in and These represent the horizontal and vertical coordinate components in the two-dimensional image plane, respectively. The matrix data mentioned above serves as the basic input source for subsequent second-order partial derivative operations.
[0044] S22, after receiving image data, the control system performs tubular feature enhancement filtering based on the second-order partial derivative matrix. Airflow impurity removal was implemented previously, but some small water stains or tightly bound mud clumps with high surface tension may still adhere to areas with dense root hairs.
[0045] Plant roots exhibit high aspect ratio tubular and linear geometric shapes in a two-dimensional projection, while water droplets and mud clumps appear as blocky or isolated points. Based on the principle of local surface analysis of images, regions with drastic grayscale changes correspond to the edge structures of entities. Based on this morphological difference, the control system calculates the second-order partial derivative for any pixel in the image, constructing the following second-order partial derivative matrix: ; In the formula, The second-order partial derivative matrix representing the image pixels; It represents the second-order partial derivative of the pixel grayscale value in the horizontal direction; This represents the second-order partial derivative of the pixel grayscale value in the vertical direction; It represents the mixed second-order partial derivative value of the pixel grayscale in the horizontal and then vertical directions; It represents the mixed second-order partial derivative value of the pixel grayscale in the vertical and then horizontal directions.
[0046] The control system solves for the two eigenvalues of the matrix, setting the absolute value of the first eigenvalue to be less than or equal to the absolute value of the second eigenvalue. The control system calculates the ratio of the first eigenvalue to the second eigenvalue as a morphological discrimination factor. When the absolute value of the morphological discrimination factor is less than a preset tubular threshold constant, and the absolute value of the second eigenvalue is greater than a background noise threshold constant, the control system determines that the current pixel belongs to tubular root tissue.
[0047] The tubular threshold constant is determined by the prior statistical distribution of the aspect ratio of the target plant roots, while the background noise threshold constant is calibrated based on the basic dark current noise of the environment in which the visual acquisition mechanism 4 is located. The control system calculates the weight mapping coefficients of the corresponding pixels using this decision logic, retaining pixel regions with tubular features and suppressing noise regions with blocky or dot-like features. After filtering, the maximum inter-class variance method is applied to perform global threshold segmentation on the enhanced image, generating an X-axis binary mask matrix. With the Y-axis binarized mask matrix .
[0048] In this mask matrix, the pixel values of the effective root region are marked as one, and the pixel values of the remaining regions are marked as zero. For the gray-level histogram segmentation process using the Otsu's method, those skilled in the art can implement it using conventional image processing algorithms, which are well-known techniques in the field and will not be elaborated upon here.
[0049] S23, the control system calculates the comprehensive effective total root area to eliminate single-view occlusion errors based on the binarized mask matrix. The spatial distribution of plant secondary roots easily leads to self-occlusion within the two-dimensional projection plane of a single viewpoint, causing the single camera's field of view to be unable to accurately quantify the true biological volume. Introducing 3D point cloud reconstruction would result in excessive computational load. The control system overcomes this deficiency by fusing the two-way projection area. The control system statistically analyzes the effective pixel area in the mask matrix: ; ; In the formula, The total area of the effective root system pixels of the two-dimensional projected image along the X-axis; The total area of the effective root system pixels of the two-dimensional projected image along the Y-axis; Represents the horizontal coordinate component within the two-dimensional image plane; Represents the vertical coordinate components within the two-dimensional image plane; Represents the binary mask matrix along the X-axis; This represents the binary mask matrix along the Y-axis.
[0050] Subsequently, the control system uses a root mean square model to fuse and calculate the area parameters obtained from the orthogonal directions: ; In the formula, This represents the total effective root area; the mathematical model compensates for the occlusion loss from a single viewpoint at a geometric level, approximating the true volume characteristics of plant roots with lower computational cost.
[0051] The control system performs area validity verification logic on the calculated total effective root area. When the total effective root area is greater than the minimum survival area threshold preset by the system, the minimum survival area threshold is determined based on the historical lower limit of the survival area of the corresponding plant variety in the current growth period. The calculated total effective root area data is temporarily stored in the internal cache of the control system as a quantitative benchmark for performing adaptive wound cutting and calculating feedforward compensation liquid volume during the inoculation treatment stage.
[0052] See attached document Figure 4 Regarding the standardized root trauma treatment and residual data mapping before inoculation, the specific implementation of step S30 above includes the following process.
[0053] S31, the control system calculates the target wound area based on the acquired total effective root area. Natural growth differences exist among individual plants; using uniform mechanical shearing can easily lead to seedlings with different root systems experiencing extremely uneven pathogen infection pressure. To ensure the biological rigor of disease resistance identification, the control system uses the following model to calculate the target wound area: ; In the formula, Represents the target wound area; This represents the total effective root area calculated above; This represents the preset equivalent inoculation pressure coefficient. The equivalent inoculation pressure coefficient is determined through pathogenicity calibration in preliminary experiments. In actual agricultural engineering applications, for most solanaceous or cucurbit vegetables, this coefficient is usually set between 0.15 and 0.35 to ensure that sufficient pathogen infection channels are created in the roots without causing the plant to wilt and die due to irreversible severe water loss.
[0054] After calculating the target wound area, the control system extracts the effective pixel width along the vertical axis of the image plane, compresses and reduces the dimensionality of the projection of the three-dimensional space onto the two-dimensional plane along the horizontal direction, and constructs a comprehensive vertical pixel distribution function: ; In the formula, Represents the overall effective pixel width at a specific vertical coordinate; Represents the horizontal coordinate component within the two-dimensional image plane; This represents the vertical coordinate component within the two-dimensional image plane. This mathematical operation extracts the rate of change of cross-sectional volume of the root system along the downward direction of gravity, providing the underlying data structure for subsequent one-dimensional spatial cutting optimization.
[0055] S32, the control system determines the dynamic cutting point coordinates based on the comprehensive longitudinal pixel distribution function and drives the mechanical actuator to complete the truncation. Since the seedling is in a suspended posture, the bottom of the root system naturally droops to the lowest point of the field of view. The control system uses the vertical coordinates corresponding to the bottom of the root system as the starting optimization position and performs discrete accumulation and summation operations from bottom to top along the vertical direction, that is, accumulating the comprehensive effective pixel width row by row.
[0056] When the accumulated area value equals or just exceeds the target wound area, the control system terminates the optimization process and records the current vertical coordinates as the dynamic cutting point's vertical coordinates. To avoid algorithmic deadlocks or over-cutting due to extremely sparse root systems in individual seedlings, the control system has a boundary out-of-bounds judgment rule: when the accumulated area reaches the set root neck safety boundary coordinates but still does not reach the target wound area, the control system triggers an abnormal interruption mechanism, marks the seedling as a non-standard sample, and stops the truncation action.
[0057] After determining the vertical coordinates of the dynamic cutting point, the control system converts the pixel's vertical coordinate into the actual Z-axis execution height based on a pre-calibrated pixel-spatial resolution mapping model. This mapping model eliminates nonlinear errors caused by camera mounting offset and lens distortion. Once the height is obtained, the control system outputs digital pulse control commands, which are sent to the linear servo drive module of the inoculation mechanism 5.
[0058] The linear servo drive module precisely moves the shearing cutter to the aforementioned height, then triggers the cutter to perform a horizontal closing motion, completing the quantitative truncation of the overhanging root system. The process of obtaining the calibration coefficients for the conversion of camera pixel coordinates to spatial coordinates can be achieved by those skilled in the art using the conventional Zhang Zhengyou camera calibration method, which is well-known in the field and will not be elaborated upon here.
[0059] S33, the control system calculates the residual area parameters and performs radio frequency identity feedforward correlation. After the truncation action is completed, the seedling retains only the healthy root tissue above the wound. The control system calculates the residual effective root area parameters of the seedling by subtracting the actual excised pixel area obtained by the above summation from the total effective root area.
[0060] Compared to direct secondary image capture and calculation, the indirect solution method using numerical subtraction avoids the computational burden of repeated image processing, significantly improving pipeline cycle efficiency.
[0061] After the parameters are calculated, the shifting mechanism 1 moves the culture unit 2 carrying the seedling to the RFID reader / writer station of the inoculation process. The control system uses the internal industrial bus to activate the RFID reader / writer component located at the inoculation mechanism 5 to read the unique identification code stored in the RFID tag on the outer wall of the corresponding culture unit 2.
[0062] The control system binds and maps the calculated residual effective root area parameters with the identification code and stores them in the relational database of the internal storage unit.
[0063] This data association operation establishes a feedforward data record of the residual root volume of individual seedlings, enabling the core physiological parameters of the plant to be accurately accessed in subsequent micro-irrigation and quantitative water replenishment processes after the plant is removed from the visual monitoring station.
[0064] See attached document Figure 4 For quantitative inoculation of pathogens after wound treatment and refined moisture management during the recovery period, the specific implementation of step S40 above includes the following process.
[0065] S41, the control system performs volume compensation calculation for the inoculation liquid level based on the residual effective root area parameter. The inoculation mechanism 5 is equipped with a root-soaking container holding a pathogen suspension. When the plant's residual roots are immersed in the root-soaking container, the displacement of an equal volume of liquid causes an overall rise in the actual liquid level. To ensure that all seedling root wounds are submerged to a uniform effective infection depth and to avoid inconsistencies in infection area due to differences in individual root volume, the control system uses a three-dimensional volume mapping formula to calculate the residual root volume: ; In the formula, Represents the volume of the residual root system; The residual effective root area parameter represents the area calculated and bound by the aforementioned process; Three-dimensional mapping coefficients representing plant root morphology; Represents the allometric growth index. Stored internally by the control system. and The constant values were obtained by power function regression analysis of the volume of the three-dimensional scan point cloud and the two-dimensional projected area of healthy root samples of the same variety.
[0066] After calculating the residual root volume, the control system calculates the liquid level rise height based on the principle of equivalent displacement of liquid: ; In the formula, Represents the height of the liquid surface bulge; Represents the volume of the residual root system; This represents the effective internal bottom area of the root-immersion container.
[0067] S42, the control system generates a volume compensation command based on the height of the liquid level rise, driving the inoculation mechanism 5 to perform a dynamic root-immersion action. Setting the vertical downward direction along gravity as the positive Z-axis, the control system calculates the target coordinates of the gripping robot in the vertical downward direction: ; In the formula, This represents the target coordinates for the downward movement of the gripper robot. The reference coordinates for the liquid level in the root-immersion container when it is in a static state; Represents the height of the liquid surface bulge; This represents the system's preset wound infection immersion depth. This wound infection immersion depth is determined based on the minimum length of plant physiological tissue required for effective infection by the target pathogen.
[0068] The control system sends the calculated coordinate commands to the servo control driver of the inoculation mechanism 5. The gripping robot, carrying the severed seedling roots, descends vertically to the target coordinate position and maintains the set immersion time.
[0069] This dynamic compensation control logic counteracts the interference of liquid level fluctuations in the root soaking container caused by differences in root volume, ensuring that the cut surface of each seedling receives the same water pressure and contact area for pathogen infection. After immersion inoculation, the gripping robot transplants the seedlings into the recovery culture unit 2 filled with fresh sterile substrate, and they then enter the subsequent environmentally controlled cultivation stage along with the transfer mechanism 1. For the multi-axis linkage and closed-loop position control of the gripping robot's servo motor, those skilled in the art can use conventional proportional-integral-derivative control algorithms; the underlying kinematic calculations are well-known in the field and will not be elaborated upon here.
[0070] S43, the seedlings enter the environmental recovery zone where the micro-irrigation mechanism 6 is located via the transport track. The control system executes feedforward micro-irrigation logic based on the residual area. After root pruning, the water absorption cross-section of the seedlings is significantly reduced. If the standard irrigation strategy is maintained, excessive water accumulation can easily lead to rhizosphere microenvironment hypoxia and secondary infections by non-target pathogens such as Pythium. The control system retrieves the residual effective root area parameter uniquely bound to the current seedling identifier from the relational database via the industrial bus and substitutes it into the following feedforward water requirement calculation formula to calculate the target micro-irrigation amount: ; In the formula, Represents the target micro-irrigation volume; This represents the unit root water absorption coefficient under the current environmental saturated vapor pressure difference. The parameter representing the residual effective root area; This represents the substrate evaporation compensation under current temperature and humidity conditions. Among them, The data is calculated based on the current temperature and relative humidity collected by the greenhouse environmental control system. The water loss rate of the plantless control culture unit 2 was calibrated under the same temperature and humidity conditions.
[0071] To prevent distortion in water requirement calculations due to residual noise in previous images or extremely diseased plants, the control system performs upper and lower limit threshold checks on the target micro-irrigation volume. When the calculated target micro-irrigation volume is lower than the set minimum water volume for survival, the control system triggers a forced assignment process, forcibly assigning it to the minimum water volume for survival; when the target micro-irrigation volume is higher than the maximum water holding capacity of the cultivation substrate, it is limited to the maximum water holding capacity.
[0072] The minimum water requirement for survival is determined by the critical water content at which plants wilt, while the maximum water holding capacity is determined by the porosity and volumetric properties of the cultivation substrate. After data verification, the control system converts the final target micro-irrigation volume into the duty cycle and operating cycle parameters of a pulse width modulation signal, driving the fluid pump of the micro-irrigation mechanism 6 to perform quantitative water injection.
[0073] This feedforward control architecture directly intervenes in the subsequent physiological water supply by using the biomorphological parameters obtained from the preceding destruction station, avoiding the water management imbalance caused by the lag response of traditional soil moisture sensors, and providing a consistent and standardized survival environment for the disease progression of disease resistance identification.
[0074] See attached document Figure 5 For the extraction of disease phenotype and quantification of resistance after the end of the disease cycle, the specific implementation of the above step S50 includes the following process.
[0075] S51, after inoculation, the seedlings complete the predetermined pathogen incubation and disease development cycle in the environmentally controlled cultivation area. This disease development cycle is pre-set based on the infection kinetics of the target pathogen and the duration of environmental temperature and humidity. After the cultivation period ends, the transfer mechanism 1 moves the cultivation unit 2 to the disease monitoring mechanism located at the end of the transmission trajectory. The control system triggers the color industrial camera mounted on top of the disease monitoring mechanism via the industrial bus to acquire images, obtaining a digital image matrix of the seedling canopy.
[0076] The control system performs region-of-interest (ROI) cropping and normalization on the digital image matrix to generate an input image matrix. The normalization process specifically includes: scaling the cropped image to a fixed resolution of 512 x 512 pixels using bilinear interpolation, and mapping pixel grayscale values to a floating-point range of zero to one to eliminate data distribution drift caused by minor fluctuations in ambient lighting. The spatial dimensions of the input image matrix are fixed at the system-preset pixel width and height, and the channel dimensions are three color channels: red, green, and blue.
[0077] S52, the control system calls the internally deployed disease semantic segmentation neural network model to process the input image matrix. This disease semantic segmentation neural network model adopts a fully convolutional network architecture, and its internal hierarchical structure consists of an encoder module, a feature concatenation module, and a decoder module cascaded in sequence.
[0078] Input data enters the encoder module, which uses four consecutive sets of cascaded convolutional layers and max pooling layers to gradually extract local features of color and texture of the canopy leaves and downsample them to obtain global morphological semantic information. The decoder module upsamples the feature maps through four sets of deconvolutional layers to restore the spatial resolution of the data. The feature stitching module directly connects the feature maps of the corresponding dimensions in the encoder module across layers to the decoder module to supplement the edge spatial details lost during downsampling.
[0079] The model's data flow at the end is reduced in dimensionality through a 1x1 convolution kernel in the output layer, and combined with a normalized exponential function, outputs the probability of each pixel belonging to a different category.
[0080] In this application scenario, the model input is the normalized three-channel tensor data described above; the output is a two-dimensional class label matrix with the same spatial resolution as the input. Each discrete pixel in this two-dimensional class label matrix corresponds to a real biological state, specifically including the background region corresponding to a value of zero, the healthy canopy region corresponding to a value of one, and the diseased, yellowed, and withered region corresponding to a value of two.
[0081] The construction and training process of the aforementioned disease semantic segmentation neural network model was performed in advance before model deployment. Under the same greenhouse light environment as actual disease resistance identification, real images of plant canopies at various growth stages with target lesions were collected as training sample sets. Professionals used polygon annotation tools to manually mask the sample sets at the pixel level, generating a label set as supervision signals.
[0082] The training phase aims to minimize the difference between the predicted probability distribution and the true label distribution, employing a multi-class cross-entropy function as the loss function for backpropagation and gradient calculation. The parameter setting and weight update process of the adaptive moment estimator optimizer used in model training can be implemented using conventional deep learning frameworks by those skilled in the art, and are well-known techniques in the field, so they will not be elaborated upon here.
[0083] S53, the control system calculates the canopy disease severity parameters based on the output two-dimensional category label matrix. The control system iterates through and counts the number of pixels of different categories in the matrix: ; ; In the formula, Represents the pixel area of a healthy canopy; Represents the area of the coronal pixels of the lesion; Represents the row coordinate variable; Represents column coordinate variables; Represents an indicator function; This represents the category value of a two-dimensional category label matrix at a specific row and column coordinate.
[0084] To prevent the failure to detect effective plant pixels due to complete plant death and lodging or mechanical obstruction of the camera's field of view, which could lead to a zero-division error in subsequent division operations, the control system is equipped with a survival area verification logic.
[0085] When the sum of the healthy canopy pixel area and the diseased canopy pixel area is less than the set minimum canopy area lower limit, the control system triggers a forced assignment process, directly determining the seedling as completely lethal and forcibly assigning a disease index of 100. This minimum canopy area lower limit is determined based on the field of view of the visual camera and the remaining cross-sectional area after the plant naturally wilts. When the total area meets the lower limit requirement, the control system calculates the disease index using the following formula: ; In the formula, Represents the disease index; Represents the pixel area of a healthy canopy; Represents the area of pixels in the coronal layer of the lesion.
[0086] S54, the control system integrates feedforward root data to generate a comprehensive resistance evaluation index. Relying on a single canopy incidence rate cannot objectively reflect the plant's true disease resistance potential; plants with lower initial biomass are more likely to develop disease phenotypes at a faster rate than the baseline. The control system retrieves the residual effective root area parameters bound to the RFID tag of culture unit 2 from the relational database and performs a comprehensive resistance evaluation calculation: ; In the formula, Represents comprehensive resistance evaluation indicators; Represents the disease index; The root system volume compensation weight coefficient; The parameter representing the residual effective root area; The statistical average value of the residual effective root area parameter of the same batch of identified samples.
[0087] The root mass compensation weighting coefficient was determined through pathological benchmark correction experiments, with a value range of 0.1 to 0.3 to limit excessive fluctuations in the compensation term. This evaluation calculation model mathematically balances the interference of initial biomass differences in resistance identification.
[0088] After the calculation is completed, the control system stores the comprehensive resistance evaluation index into the database, generates a complete disease resistance phenotype feedforward data record including the initial root morphology, target wound area, dynamic micro-irrigation amount and the final disease evolution trajectory, and outputs it to the human-computer interaction interface to provide multi-dimensional data support for the screening of resistant varieties.
[0089] Specific application examples: In this embodiment, taking the automated inoculation and resistance evaluation of a single tomato seedling as an example, we demonstrate how the control system performs adaptive regulation based on the model constructed above.
[0090] 1. Visual feature extraction and target trauma calculation: After impurities are removed, the seedlings enter the vision acquisition mechanism 4. The control system extracts the mask matrix and calculates the total effective root pixel area of the X-axis two-dimensional projection image. The total effective root pixel area of the 2D projected image along the Y-axis is 4000. The value is 3000. The root mean square model is used for fusion calculation to obtain the total effective root area that eliminates single-view occlusion errors: .
[0091] The system's preset equivalent inoculation pressure coefficient If the value is 0.2, the target wound area can be calculated as follows: .
[0092] Reference Figure 6 The horizontal axis represents the overall effective pixel width. The vertical axis represents the vertical coordinate components in the two-dimensional image plane. This image shows the discrete variation trend of the pixel width of the cross section of the overhanging root system along the direction of gravity with depth.
[0093] The control system extracts the overall effective pixel width along the vertical axis of the image. The algorithm constructs a function curve. Starting from the vertical coordinates corresponding to the lowest point of the root system, it performs discrete summation operations vertically from bottom to top. When the actual excised pixel area obtained by the summation precisely reaches the target wound area of 1000, the optimization stops and the current vertical coordinates are locked. The shearing tool moves to this height to perform truncation, at which point the residual effective root area parameter of the seedling is determined. Updated to: 5000 1000 = 4000. This parameter is written to the RFID tag for feedforward binding.
[0094] 2. Root soaking compensation and micro-irrigation feedforward control: Seedlings are conveyed into the inoculation facility 5. The system calibrates the three-dimensional mapping coefficients of plant root morphology. The allometric growth index is 0.05. The value is 1.0. The calculated residual root volume is: .
[0095] The effective internal base area of the root-immersion container is known. The value is 100. According to the equivalence principle, the height of the liquid surface rise caused by the displaced liquid is: .
[0096] Let the reference coordinates of the liquid level in the root-immersing container be in a static state. The system presets the wound immersion depth to 150. The value is 10. The control system calculates the target coordinates for the downward movement of the gripping robot: .
[0097] The robotic arm precisely descended to coordinate 158, offsetting the liquid level fluctuations caused by volume expansion. After the transplant recovery, the control system retrieved... Perform quantitative fluid replenishment. Assume a unit root water absorption coefficient under the current ambient saturated vapor pressure difference. The evaporation compensation of the substrate under the current temperature and humidity conditions is 0.005. The value is 10. Calculate and output the target micro-irrigation amount: .
[0098] 3. Quantitative assessment of disease progression and comprehensive evaluation of resistance: After the onset of the disease cycle, the control system outputs a two-dimensional category label matrix through a neural network, and iterates through and statistically calculates the pixel area of the healthy canopy. The number is 60,000, representing the area of the coronal pixels of the lesion. The value is 40,000. Calculate the disease index: .
[0099] The system reads the statistical average value of the residual effective root area parameter of the same batch of identification samples. Set the root system volume compensation weight coefficient to 5000. The value is 0.2. The control system performs fusion calculations to eliminate initial biomass differences: .
[0100] Ultimately, the comprehensive resistance evaluation index output by this tomato plant was determined. Revised to 57.6.
[0101] Experimental verification and effect comparison: To verify the reliability of this invention, a control experiment was set up. Two groups of 100 tomato seedlings each were selected from the same batch of sowing and at the same physiological stage. The seedlings were cut to a fixed length of 2 cm and irrigated with a constant amount of 50 mL using the traditional fixed parameter method. The seedlings were then compared with the feedforward adaptive method of this invention to identify resistance to Fusarium wilt.
[0102] Reference Figure 7 The horizontal axis represents the discrete sample numbers 1 to 100 of 100 independently tested seedlings, and the vertical axis represents the comprehensive resistance evaluation index finally calculated by the formula. The specific numerical values are shown by the two independent broken lines in the figure, which intuitively illustrate the fluctuation of the final resistance scores of the two groups of samples as individuals switch.
[0103] Traditional method group corresponds to the broken line: Since no adaptive adjustment is made for the differences in root volume of individual seedlings, seedlings with smaller root systems have too high a proportion of their root system removed when cut to a fixed length, and they are subjected to extremely high equivalent infection pressure. Furthermore, subsequent uniform irrigation is very likely to cause local microenvironment hypoxia.
[0104] As shown in the figure, the broken line is severely affected by the dispersion of the initial biomass, with extremely drastic fluctuations and a very large range. The overall coefficient of variation is as high as 28.4%, indicating that the traditional method generates a large amount of false disease noise data.
[0105] The method of this invention corresponds to the broken line: relying on visual area calculation and feedforward closed-loop control, it realizes adaptive equivalent trauma control and accurate micro-irrigation water compensation for each individual seedling, eliminating interference from non-pathological factors caused by inconsistencies in the initial morphology of organisms. In the figure, this broken line is extremely flat and convergent, with the individual evaluation index values highly concentrated near the true resistance level, abnormal fluctuations are smoothed out, and the coefficient of variation is reduced to 6.1%.
[0106] Experimental conclusion: The data calculation and feedforward regulation scheme provided by this invention significantly improves the stability and consistency of phenotypic identification data for targeted soil-borne diseases without increasing human intervention, highlighting the true genotypic disease resistance potential of the target plants.
Claims
1. A method for identifying resistance to soil-borne diseases in vegetable seedlings, characterized in that, Includes the following steps: Obtain orthogonal two-dimensional projection images of the hanging root system of the target seedling, extract the effective outline of the root system, and calculate the total area of the comprehensive effective root system using the root mean square model; Based on the total effective root area, the target wound area and the vertical coordinate of the dynamic cutting point are calculated. The shearing tool is controlled to move to the execution height corresponding to the vertical coordinate of the dynamic cutting point to perform root truncation. The residual effective root area parameters are calculated and associated with the radio frequency tag identification code of the culture unit (2) carrying the target seedling. The descent target coordinates of the gripping robot are adjusted according to the residual effective root area parameters to perform immersion inoculation; The residual effective root area parameter associated with the radio frequency tag identification code is read and the target micro-irrigation amount is calculated to perform quantitative liquid replenishment. The disease index is extracted after the end of the disease cycle, and a comprehensive resistance evaluation index is calculated and output by combining the residual effective root area parameter with the disease index.
2. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The extraction of the effective root contour includes: Calculate the second-order partial derivatives of the pixels of the orthogonal two-dimensional projection image to construct a second-order partial derivative matrix, and solve for the first eigenvalue and the second eigenvalue; The ratio of the first feature value to the second feature value is calculated as a morphological discrimination factor. When the absolute value of the morphological discrimination factor is less than the tubular threshold constant and the absolute value of the second feature value is greater than the background noise threshold constant, the current pixel is determined to belong to the tubular root system tissue, and the pixel domain of the tubular feature is retained. The tubular threshold constant is determined by the prior statistical distribution of the length-to-width ratio of the target plant root system, and the background noise threshold constant is calibrated based on the basic dark current noise of the environment in which the visual acquisition mechanism (4) that acquires the orthogonal two-dimensional projection image is located. The pixel domain of the tubular feature is segmented globally using the Otsu's method to generate a binary mask matrix as the effective contour of the root system.
3. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 2, characterized in that, The calculation of the total effective root area using the root mean square model includes: The effective pixel area of the binarized mask matrix in the orthogonal directions is calculated respectively. The total area of the comprehensive effective root system is obtained by taking the square root of the sum of the squares of the areas of each effective pixel. When the total effective root area is greater than the minimum survival area threshold, it is temporarily stored in the internal cache as a quantitative benchmark. The minimum survival area threshold is determined based on the historical lower limit of the survival area of the corresponding plant variety during the current growth period.
4. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The calculation of the target wound area and the vertical coordinates of the dynamic cutting point based on the total effective root area includes: The target wound area is obtained by multiplying the total effective root area by the equivalent inoculation pressure coefficient. Extract the effective pixel width from the effective contour of the root system to construct a comprehensive longitudinal pixel distribution function. Take the vertical coordinate corresponding to the bottom of the hanging root system as the starting optimization position and perform discrete accumulation and summation operation from bottom to top along the vertical direction. That is, accumulate the effective pixel width row by row. When the accumulated area value is equal to or just exceeds the target wound area, record the current vertical coordinate as the vertical coordinate of the dynamic cutting point.
5. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The step of adjusting the downward target coordinates of the gripping robot to perform immersion inoculation based on the residual effective root area parameter includes: The residual effective root area parameter is combined with the allometric growth index and the three-dimensional mapping coefficient to perform a power function operation to solve for the residual root volume. Divide the volume of the residual root system by the effective internal bottom area of the root soaking container to obtain the height of the liquid level rise. The target coordinates for the downward movement of the gripping manipulator are obtained by subtracting the height of the liquid surface bulge from the reference coordinates of the liquid surface and adding the immersion depth of the wound infection.
6. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The step of reading the RFID tag identification code and calling back the associated residual effective root area parameter to calculate the target micro-irrigation amount and perform quantitative liquid replenishment includes: The target micro-irrigation amount is calculated by multiplying the residual effective root area parameter by the unit root water absorption coefficient and adding the substrate base evaporation compensation amount. The target micro-irrigation amount is checked for upper and lower limits. When the calculated target micro-irrigation amount is lower than the set minimum water volume for maintaining survival, a forced assignment process is triggered, and the target micro-irrigation amount is forcibly assigned to the minimum water volume for maintaining survival. When the target micro-irrigation volume is higher than the maximum water holding capacity of the cultivation substrate, the target micro-irrigation volume is limited to the maximum water holding capacity. The minimum water requirement for survival is determined by the critical water content at which the plant wilts, and the maximum water holding capacity is determined by the porosity and volume characteristics of the cultivation substrate. The final target micro-irrigation volume is converted into a pulse width modulation signal to control the irrigation valve and the precision fluid pump to perform quantitative replenishment to the culture unit (2).
7. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The extracted disease index after the end of the disease cycle includes: The region of interest (ROI) is cropped and normalized on the canopy digital image matrix acquired after the onset of the disease cycle to generate the input image matrix; The disease semantic segmentation neural network model is invoked to process the input image matrix and output a two-dimensional category label matrix; The disease index is forcibly assigned a value of one hundred when the sum of the diseased and healthy canopy pixel areas in the two-dimensional category label matrix is calculated. When the sum of the area of the diseased canopy pixels and the area of the healthy canopy pixels is greater than or equal to the minimum lower limit of the canopy area, the ratio of the area of the diseased canopy pixels to the sum of the two areas is calculated and multiplied by one hundred to obtain the disease index. The minimum canopy area lower limit is determined based on the field of view of the visual camera that acquires the canopy digital image matrix and the remaining cross-sectional area of the plant after natural wilting.
8. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 7, characterized in that, The disease semantic segmentation neural network model adopts a fully convolutional network architecture, including: The encoder module, feature splicing module, and decoder module are cascaded together in sequence. Each discrete pixel in the two-dimensional category label matrix corresponds to a background region with a value of zero, a healthy canopy region with a value of one, and a diseased, yellowed, and withered region with a value of two.
9. The method for identifying resistance to soil-borne diseases in vegetable seedlings according to claim 1, characterized in that, The comprehensive resistance evaluation index, calculated by combining the residual effective root area parameter and the disease index, includes: Calculate the difference between the residual effective root area parameter and the statistical average of the residual effective root area parameters of the same batch of identified samples, divide the difference by the statistical average, multiply the result by the root volume compensation weight coefficient, and then add it to the value. The result of subtracting the disease index from the value of 100 is multiplied by the sum of the above results to obtain the comprehensive resistance evaluation index.
10. A device for identifying resistance to soil-borne diseases in vegetable seedlings, used to implement the method for identifying resistance to soil-borne diseases in vegetable seedlings as described in any one of claims 1 to 9, characterized in that, include: The shifting mechanism (1) is equipped with a culture unit (2), and the outer wall of the culture unit (2) is fitted with an RFID tag containing a unique identification code; It is equipped with a linear air knife assembly and a pneumatic pressure regulating valve arranged in the vertical direction, which is used to blow off the attached objects and remove impurities (3). The visual acquisition mechanism (4) for acquiring orthogonal two-dimensional projection images is equipped with a first camera and a second camera arranged horizontally at a 90-degree angle, and is equipped with a photoelectric switch and a backlight. The inoculation mechanism (5) used to perform root cutting and immersion inoculation is equipped with a shearing cutter with a linear servo drive module, a clamping manipulator and an immersion container with a liquid level limit structure. The irrigation mechanism (6) for performing quantitative replenishment is equipped with a reader, a precision fluid pump and an irrigation valve; A disease monitoring facility equipped with color industrial cameras for acquiring digital image matrices of the canopy; A control system that establishes a two-way data communication connection with the displacement mechanism (1), the impurity removal mechanism (3), the visual acquisition mechanism (4), the inoculation mechanism (5), the irrigation mechanism (6), and the disease monitoring mechanism via an industrial bus and has a built-in storage unit.
