A method and system for efficient removal of weeds in the breeding process of rapeseed

Abstract

The application provides a method and system for efficiently removing weeds in the breeding process of oilseed rape, belonging to the technical field of agricultural seed production. The method comprises: obtaining the shape parameters and area parameters of the oilseed rape field, combining the preset optimal plant spacing and the size of the intelligent mobile robot, performing grid division, and sowing oilseed rape seeds; obtaining the visual basic features of each oilseed rape plant, using statistical methods to calculate and screen out the oilseed rape plants determined as "weeds"; and removing. The application aims at the technical pain points in the prior art, such as low efficiency of manual operation for removing weeds in the seedling stage of oilseed rape production, high deployment cost of existing intelligent recognition models, and difficulty in adapting to large-scale production. By combining the core technical means of grid division, statistical operation, machine vision basic feature extraction, and intelligent mobile robot operation, efficient, accurate, and low-cost removal of weeds is realized.

Description

Technical Field

[0001] This invention relates to the field of agricultural seed production technology, and in particular to a method and system for efficiently removing hybrid plants during rapeseed seed production and cultivation. Background Technology

[0002] In the process of rapeseed seed production, the seedling stage is a critical growth stage. Removing unwanted plants (including weeds and plants with significantly different phenotypes from the planted population) during this stage can optimize the breeding growth environment and effectively reduce the amount of field patrol and weed removal work during the subsequent bolting and flowering stages. Therefore, weed removal during the seedling stage is a key step in the hybrid rapeseed seed production process.

[0003] In existing technologies, the removal of weeds during the rapeseed seed production stage is mostly carried out by manually inspecting the field and visually comparing them to identify and remove the weeds. This method suffers from low efficiency and is difficult to adapt to the needs of large-scale rapeseed seed production.

[0004] The seed industry is gradually moving towards smart breeding and production that integrates biotechnology, artificial intelligence, and big data information technology. The application of intelligent equipment developed using AI technologies such as machine vision, deep learning, and data fusion in field agriculture provides a technical feasibility for intelligent identification and removal of weeds during the rapeseed seedling stage. However, the high deployment cost of mature intelligent weed identification models limits the applicability of such AI technologies in rapeseed seedling removal, hindering large-scale application. Summary of the Invention

[0005] This invention aims to provide a method and system for efficient removal of weeds during rapeseed seed production. Addressing the technical pain points of existing technologies, such as low efficiency of manual weed removal during the seedling stage, high deployment costs of existing intelligent recognition models, and difficulty in adapting to large-scale production, this invention combines core technologies such as grid partitioning, statistical calculations, machine vision feature extraction, and intelligent mobile robot operation to achieve efficient, accurate, and low-cost removal of weeds.

[0006] The technical solution adopted in this invention is:

[0007] A method for efficiently removing hybrid plants during rapeseed seed production includes the following steps:

[0008] Step S1: Obtain the shape and area parameters of the rapeseed field, and based on the shape and area parameters, combined with the preset optimal plant spacing and the size of the intelligent mobile robot, divide the field into grids. Each grid is a sowing block and is assigned a unique location number. The crisscrossing grid lines are the movement path of the intelligent mobile robot. Sow a preset number of rapeseed seeds in each sowing block.

[0009] Step S2: Once the rapeseed seeds have grown and developed to a preset stage, the intelligent mobile robot is driven to move along the movement path. The image recognition module on the intelligent mobile robot takes pictures of the rapeseed plants to obtain the basic visual features of each rapeseed plant. Then, statistical methods are used to calculate the basic visual features and screen out the rapeseed plants identified as "hybrids". The basic visual features include at least plant height, average crown width, and RGB value of the main leaf color. When the image recognition module takes pictures, it adopts a fixed posture and maintains a constant shooting angle through real-time tilt compensation. At the same time, it performs scale normalization and perspective correction on the captured images based on the physical size of the grid of the sowing area, and combines the relative elevation of the ground of the plot to unify the plant height measurement benchmark, so that the equivalent shooting distance and imaging scale between the lens and each plant are consistent.

[0010] Step S3: The intelligent mobile robot moves along the moving path to the vicinity of the sowing block corresponding to the location number of the rapeseed plant identified as a "hybrid", and activates the weed removal execution module to remove it.

[0011] Furthermore, before obtaining the basic visual features of each rapeseed plant in step S2, the rapeseed plant position is determined first. Specifically, the actual growth position of each rapeseed plant is detected by the distance sensor carried by the intelligent mobile robot, combined with the unique location number of the sowing block, and the offset between the actual growth position and the sowing position of the corresponding sowing block is calculated. If the offset is greater than a preset threshold, the rapeseed plant is directly determined to be a "hybrid plant", and there is no need to perform the subsequent basic visual feature extraction and statistical calculation.

[0012] Furthermore, the specific method for obtaining the visual basic features in step S2 is as follows:

[0013] Plant height: The rapeseed plant is photographed by the image recognition module to obtain an elevation view and a top view. The pixel coordinates of the highest point of the rapeseed plant are located by edge detection and contour extreme point extraction. The actual plant height of the rapeseed plant is obtained by combining the intrinsic and extrinsic parameters including camera focal length, pixel size, shooting distance and shooting posture correction coefficient.

[0014] Average crown width: Extract the pixel outline of the rapeseed plant crown in the top view, calculate the horizontal pixel span and vertical pixel span of the outline, take the average of the two and combine them with the pixel-actual size conversion ratio to obtain the actual crown width average.

[0015] Leaf primary color RGB value: Remove background pixels from the top view and keep only fresh leaf pixels. Extract the RGB channel values ​​of the fresh leaf pixels and calculate the average value as the leaf primary color RGB value.

[0016] Furthermore, the specific process of obtaining the basic visual features of each rapeseed plant in step S2, and then using statistical methods to calculate the basic visual features, includes:

[0017] Step S21A: Select all or part of the rapeseed plants as a feature statistical sample; if a portion of the rapeseed plants are selected as the feature statistical sample, the sample size shall not be less than 1% of the total number of rapeseed plants.

[0018] Step S22A: Perform statistical operations on each feature item of the visual basic features of each rapeseed plant in the feature statistical sample to calculate the mean μ1 and standard deviation σ1 of the feature item, and use μ1±σ1 or μ1±2σ1 as the first hybrid plant determination threshold corresponding to the feature item.

[0019] Step S23A: Verify the basic visual features of all rapeseed plants one by one. If any feature item in the basic visual features of a single rapeseed plant exceeds the corresponding first hybrid plant determination threshold, the current rapeseed plant is determined to be a "hybrid plant". For rapeseed plants that do not exceed the first hybrid plant determination threshold, they are determined to be "normal plants".

[0020] Furthermore, before the removal execution module is activated in step S3 to remove the rapeseed plants confirmed as "hybrids", the image recognition module switches the shooting angle and re-shoots the image, then extracts the basic visual features again and compares them with the first hybrid plant determination threshold. If any feature item in the basic visual features still exceeds the corresponding first hybrid plant determination threshold, it is finally confirmed as a "hybrid" and directly removed. If it does not exceed the first hybrid plant determination threshold, the removal is paused, and a warning or record is issued.

[0021] Furthermore, the specific process of obtaining the basic visual features of each rapeseed plant in step S2, and then using statistical methods to calculate the basic visual features, includes:

[0022] Step S21B: Based on the grid division results, the field microenvironment data of each sowing block, and the operation batch of the intelligent mobile robot, the rapeseed field is divided into several secondary comparison units. Each secondary comparison unit includes a microenvironment homogenization unit and an operation batch verification unit. The microenvironment homogenization unit includes rapeseed plants corresponding to adjacent sowing blocks with consistent soil fertility and light conditions within the grid. The operation batch verification unit is composed of several adjacent microenvironment homogenization units and is adapted to the coverage range of a single operation of the intelligent mobile robot.

[0023] Step S22B: Obtain the basic visual features of each rapeseed plant within each microenvironment homogenization unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ2 and standard deviation σ2 of each feature item. Use μ2±σ2 or μ2±2σ2 as the second hybrid plant determination threshold for the corresponding feature item. Verify the basic visual features of all rapeseed plants within the same microenvironment homogenization unit plant by plant. If any feature item of the basic visual features of a single rapeseed plant exceeds the corresponding second hybrid plant determination threshold, the current rapeseed plant is determined to be a "suspected hybrid plant". Rapeseed plants that do not exceed the second hybrid plant determination threshold are determined to be "normal plants".

[0024] Step S23B: Obtain the basic visual features of each rapeseed plant within each batch verification unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ3 and standard deviation σ3 of each feature item. Use μ3±σ3 or μ3±2σ3 as the third hybrid plant determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants identified as "suspected hybrid plants" in step S22B with the third hybrid plant determination threshold. If any feature item in the basic visual features exceeds the corresponding third hybrid plant determination threshold, it is determined to be a "hybrid plant".

[0025] Furthermore, before the removal execution module is activated in step S3 to remove the rapeseed plants confirmed as "hybrids", the image recognition module switches the shooting angle and re-shoots the image, then extracts the basic visual features again and compares them with the second and / or the third hybrid determination threshold. If any feature item in the basic visual features still exceeds the corresponding second and / or the third hybrid determination threshold, it is finally confirmed as a "hybrid" and directly removed. If it does not exceed the second and / or the third hybrid determination threshold, the removal is paused, and a warning or record is issued.

[0026] Furthermore, the specific process of obtaining the basic visual features of each rapeseed plant in step S2, and then using statistical methods to calculate the basic visual features, includes:

[0027] Step 21C: Based on the grid division results, the rapeseed field is divided into basic units, combined units, and whole-field area units according to the total area size from small to large. The basic unit includes several rapeseed plants corresponding to several adjacent planting blocks. The combined unit is composed of several adjacent basic units, and the whole-field area unit is composed of several adjacent combined units, which is the complete rapeseed field.

[0028] Step S22C: Obtain the visual basic features of each rapeseed plant within each basic unit. Then, perform statistical operations on each feature item of the visual basic features of each rapeseed plant to calculate the mean μ4 and standard deviation σ4 of each feature item. Use μ4±σ4 or μ4±2σ4 as the fourth hybrid plant determination threshold for the corresponding feature item. Verify the visual basic features of all rapeseed plants within the same basic unit plant by plant. If any feature item of the visual basic features of a single rapeseed plant exceeds the corresponding fourth hybrid plant determination threshold, the current rapeseed plant is determined to be a "primary suspected hybrid plant". Rapeseed plants that do not exceed the fourth hybrid plant determination threshold are determined to be "normal plants".

[0029] Step S23C: Obtain the basic visual features of each rapeseed plant within each combined unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ5 and standard deviation σ5 of each feature item. Use μ5±σ5 or μ5±2σ5 as the fifth hybrid determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants determined as "primary suspected hybrids" in step S22C with the fifth hybrid determination threshold. If any feature item in the basic visual features exceeds the corresponding fifth hybrid determination threshold, it is determined as a "secondary suspected hybrid".

[0030] Step S24C: Obtain the basic visual features of each rapeseed plant within each whole-field area unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ6 and standard deviation σ6 of each feature item. Use μ6±σ6 or μ6±2σ6 as the sixth hybrid determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants identified as "secondary suspected hybrids" in step S23C with the sixth hybrid determination threshold. If any feature item in the basic visual features exceeds the corresponding sixth hybrid determination threshold, it is determined to be a "hybrid".

[0031] Further, before the removal execution module is activated in step S3 to remove the rapeseed plants confirmed as "hybrids", the image recognition module switches the shooting angle and re-shoots the image, then extracts the basic visual features again and compares them with the fourth, fifth, and / or sixth hybrid plant determination thresholds. If any feature item in the basic visual features still exceeds the corresponding fourth, fifth, and / or sixth hybrid plant determination threshold, it is finally confirmed as a "hybrid" and directly removed. If it does not exceed the fourth, fifth, and / or sixth hybrid plant determination thresholds, the removal is paused, and a warning or record is issued.

[0032] A highly efficient system for removing hybrid plants during rapeseed seed production is disclosed, which implements the aforementioned method for efficiently removing hybrid plants during rapeseed seed production. The system includes an intelligent mobile robot, and an image recognition module, a weed removal execution module, and a computational processing module mounted on and electrically connected to the intelligent mobile robot. The intelligent mobile robot receives instructions and moves along a path in the rapeseed field. The image recognition module photographs the rapeseed plants to obtain the basic visual features of each plant. The weed removal execution module removes the rapeseed plants identified as "hybrids." The computational processing module receives the basic visual features and then uses statistical methods to perform calculations on the visual features to filter out the rapeseed plants identified as "hybrids."

[0033] The beneficial effects of this invention are:

[0034] 1. Abandoning the existing method of manually inspecting and comparing fields to remove weeds, this technology uses intelligent mobile robots to automatically move along a preset grid path and photograph rapeseed plants in batches. This eliminates the need for manual identification and judgment of each plant, enabling continuous operation, significantly reducing the intensity of manual labor, and solving the core pain point of low efficiency in manual operations. It is suitable for large-scale production scenarios in rapeseed seed production fields of different sizes, while also reducing the amount of field inspection and weed removal work during subsequent bolting and flowering stages.

[0035] 2. A model-free statistical approach is used to analyze the basic visual features of rapeseed plants. This eliminates the need for deep learning recognition models, saving the costs associated with model training, data labeling, and high-end AI chips. The basic visual features selected are easily extracted and computationally inefficient indicators such as plant height, average crown width, and RGB values ​​of leaf primary color. These features can be processed locally using conventional computing modules, eliminating the need for cloud computing power. This overcomes the limitation of existing intelligent recognition models due to their high cost, facilitating the large-scale promotion of the technology.

[0036] 3. By dividing the sowing area into grids and assigning a unique location number, the plant, location number and captured image are bound one-to-one, which makes it easy for the intelligent mobile robot to accurately locate the target plant; combined with statistical screening of basic visual features, normal plants and hybrid plants are accurately distinguished, reducing misjudgment and avoiding the problem of wrongly removing normal rapeseed plants and missing hybrid plants, thus optimizing the breeding growth environment and ensuring the purity of rapeseed seed production.

[0037] 4. Based on grid division design, the sowing layout and robot movement path are designed, combined with the optimal spacing between rapeseed seedlings and robot size, to adapt to rapeseed seed production fields of different shapes and sizes; the visual feature extraction method is simple, without the need for complex image analysis algorithms, and the hardware only requires ordinary image recognition modules, intelligent mobile robots and conventional computing modules, without the need for customized high-end equipment, adapting to the actual application conditions of field agriculture scenarios, and easily realizing industrialization.

[0038] 5. In line with the development direction of intelligent breeding and seed production in the seed industry, machine vision, intelligent mobile and statistical analysis technologies are deeply integrated into the process of removing weeds during the seedling stage of rapeseed seed production. This replaces traditional manual operation and realizes intelligentization of the entire process of weed identification, positioning and removal. It promotes the transformation of rapeseed seed production from manual and extensive to intelligent and refined, and provides a feasible technical path for the intelligent development of rapeseed seed production. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating an efficient method for removing hybrid plants during rapeseed seed production in Example 1. Detailed Implementation

[0041] The embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0042] Example 1

[0043] Please see Figure 1 This embodiment provides a method for efficient removal of hybrid plants during rapeseed seed production and cultivation. The method includes the following steps:

[0044] Step S1: Obtain the shape and area parameters of the rapeseed field, and based on the shape and area parameters, combined with the preset optimal plant spacing and the size of the intelligent mobile robot, divide the field into grids. Each grid is a sowing block and is assigned a unique location number. The crisscrossing grid lines are the movement path of the intelligent mobile robot. Sow a preset number of rapeseed seeds in each sowing block and record the sowing location coordinates simultaneously. In this embodiment, one rapeseed seed is placed in the center of each sowing block as an example.

[0045] Among them, "shape parameters" refer to the boundary contour information of rapeseed fields, such as rectangles, trapezoids, and irregular quadrilaterals, for example, the coordinates of polygon vertices. "Area parameters" refer to the actual cultivated area of ​​the rapeseed field. "Preset optimal plant spacing" is an empirical value set based on the biological characteristics of rapeseed varieties, the ventilation and light requirements during the seedling stage, and the target for population density control, used to ensure normal development space for individual rapeseed plants. "Intelligent mobile robot dimensions" refer to the maximum projected size of its physical shape, including the overall length and width, wheel width, etc., used to avoid structural interference in path planning. Specifically, the structure of the intelligent mobile robot used in this embodiment is existing technology, such as an adaptive addition based on the functions required in this embodiment of "An Automatic Rapeseed Transplanter and Its Automatic Transplanting Method" (CN118104444B), which can be obtained by those skilled in the art without creative effort. "Grid division" divides the rapeseed field into several rectangular cells of the same or approximately the same size, with no overlap or gaps between cells, and covering the entire cultivated area. The "unique location number" is a two-dimensional coordinate code (such as "R3C5" indicating the 3rd row and 5th column), a linear sequence number (such as "001" to "N"), or a geospatial code (such as a hash code based on WGS84 coordinates), used to establish a one-to-one mapping relationship between seeding blocks and physical spatial locations.

[0046] Step S2: Once the rapeseed plants have grown to the preset stage, the intelligent mobile robot is driven to move along the movement path. The image recognition module on the intelligent mobile robot takes pictures of all the rapeseed plants to obtain the basic visual features of each rapeseed plant. Then, statistical methods are used to calculate the basic visual features and screen out the rapeseed plants identified as "hybrids". The basic visual features include at least plant height, average crown width, and RGB value of the main leaf color. The image recognition module adopts a fixed posture (fixed shooting angle and position) when taking pictures and maintains a constant shooting angle through real-time tilt compensation. At the same time, the scale normalization and perspective correction of the captured images are performed based on the physical size of the grid of the sowing area, and the plant height measurement benchmark is unified by combining the relative elevation of the ground of the plot, so that the equivalent shooting distance and imaging scale between the lens and each plant are consistent.

[0047] The "preset stage" refers to the physiological stage of rapeseed seedlings after emergence and uniform emergence, from the true leaf stage to the 5-leaf stage. During this stage, the plant morphology is stable, phenotypic differences are significant, and the rapid jointing stage has not yet begun, which is conducive to clear image acquisition and accurate feature extraction. The "image recognition module" is a common industrial-grade vision component integrating a visible light CMOS sensor, a multi-degree-of-freedom robotic arm, a fixed-focus lens, and an embedded image processing unit. It has automatic white balance, exposure compensation, and distortion correction capabilities. At the same time, in order to reduce interference from factors such as the distance between the lens and the plant, the flatness of the ground, and the shooting angle, various sensors carried by the intelligent mobile robot need to perform real-time detection and dynamic adjustment of the "image recognition module." During the shooting process, a unified measurement benchmark is used to eliminate the interference of changes in shooting angle, differences in ground flatness, and differences in shooting distance between the lens and each plant on the measurement of the basic visual features of the plant and subsequent statistical calculation results. "Plant height" in "basic visual characteristics" reflects the plant's longitudinal growth potential and can characterize the consistency of genetic background or differences in response to environmental stress; "average crown width" reflects the degree of lateral expansion of the plant and can characterize the nutrient allocation pattern and the state of competition among the population; "RGB value of leaf dominant color" reflects chlorophyll content, senescence degree and early disease manifestations and can characterize physiological metabolic state and genotype expression stability. Together, these three constitute a multidimensional quantitative description of phenotypic variation in rapeseed seedlings, which has strong biological significance and low computational coupling.

[0048] The process involves using statistical methods to calculate visual basic features and then screening out rapeseed plants identified as "hybrids". The visual basic features of all or part of the rapeseed plants can be used as a sample set. The distribution center and dispersion of the plant height, average crown width, and RGB value of the main leaf color can be calculated for each dimension. Then, a judgment threshold range is set, and individuals that deviate from the range are identified as "hybrids". This process does not rely on training sample labels and neural network structure. It only requires conventional arithmetic operations and comparison logic and can be completed locally in the embedded computing module without the need for external model loading or remote service calls.

[0049] The "plant height" is determined by capturing images of the rapeseed plant using the image recognition module, obtaining elevation views (left, right, front, and rear views) and a top view. Edge detection and contour extreme point extraction are used to locate the pixel coordinates of the highest point of the rapeseed plant. Combined with intrinsic and extrinsic parameters including camera focal length, pixel size, shooting distance, and shooting posture correction coefficients, the actual plant height is obtained. The "average crown width" is obtained by extracting the pixel contour of the rapeseed plant crown in the top view, calculating the horizontal and vertical pixel spans of the contour, averaging these two values, and then combining them with a pixel-to-actual-size conversion ratio. The "leaf primary color RGB value" is obtained by removing background pixels from the top view, retaining only fresh leaf pixels, extracting the RGB channel values ​​of the fresh leaf pixels, and calculating the average value as the leaf primary color RGB value.

[0050] Step S3: The intelligent mobile robot moves along the moving path to the vicinity of the planting block corresponding to the location number of the rapeseed plant identified as a "hybrid" and starts the weed removal execution module to remove it;

[0051] Among them, the "removal execution module" is a miniature gripper, rotary cutting blade or pneumatic suction device assembled at the end of a multi-degree-of-freedom robotic arm. Its movement is controlled by the removal command output by the computing module and works in coordination with the real-time pose information fed back by the image recognition module to ensure that it only acts on the target plant without damaging the adjacent normal plants.

[0052] This embodiment structures rapeseed fields into uniquely numbered planting blocks, ensuring traceability of the initial location of each seed. Intelligent mobile robots automatically inspect along a pre-defined grid path, achieving comprehensive and thorough image acquisition. An image recognition module extracts three easily obtainable, highly representative, and loosely coupled visual features: plant height, average crown width, and RGB values ​​of the leaf's primary color, avoiding complex image semantic understanding. A model-free statistical distribution analysis method is used to set dynamic thresholds to identify individuals significantly deviating from the population's phenotypic distribution. Finally, the robot precisely locates and physically removes these individuals. The entire process does not introduce deep learning models, relies on manually labeled data, or require additional high-end hardware investment. It only requires conventional image acquisition and local computing power, significantly reducing system deployment and maintenance costs while ensuring recognition accuracy. This effectively solves the dual bottlenecks of low manual efficiency and difficulty in promoting intelligent models, providing a replicable, scalable, and implementable intelligent technology path for removing unwanted plants during the rapeseed seedling stage.

[0053] For situations where two or more rapeseed seeds (e.g., three seeds) are simultaneously sown in a single planting hole within each sowing block to propagate multiple rapeseed plants, or where a single rapeseed seed is simultaneously sown in a planting hole at different locations within each sowing block to propagate multiple rapeseed plants (e.g., three planting holes, one seed per hole), or where two or more rapeseed seeds are simultaneously sown in a planting hole at different locations within each sowing block to propagate multiple rapeseed plants (e.g., three planting holes, two seeds per hole), the process for each rapeseed plant should be performed according to the aforementioned procedure. After the image recognition module acquires the image of the rapeseed plants within the corresponding sowing block, it is necessary to first segment the image containing multiple rapeseed plants to obtain the image of each individual rapeseed plant before performing subsequent processing. It should be noted that the relative positions of each planting hole within each sowing block should remain consistent.

[0054] Example 2

[0055] This embodiment provides an efficient method for removing hybrid plants during rapeseed seed production. Compared with Embodiment 1, it further provides that before obtaining the basic visual features of each rapeseed plant in step S2, the rapeseed plant position is determined. Specifically, the actual growth position of each rapeseed plant is detected by the distance sensor carried by the intelligent mobile robot, combined with the unique location number of the sowing block, and the offset between the actual growth position and the sowing position of the corresponding sowing block is calculated. If the offset is greater than a preset threshold, the rapeseed plant is directly determined to be a "hybrid plant" without the need for subsequent basic visual feature extraction and statistical calculation.

[0056] The "distance sensor" refers to a sensing device used for non-contact measurement of the spatial distance between a target object and the sensor. Its technical principles include, but are not limited to, ultrasonic ranging, laser time-of-flight (ToF) ranging, or infrared triangulation ranging. This type of sensor has a mature application foundation in the field of agricultural robots, featuring fast response, strong environmental adaptability, and flexible installation. In this embodiment, the "distance sensor" can be fixedly installed on the front side of the intelligent mobile robot chassis or on the image recognition module bracket, ensuring effective detection range covering a single sowing block. The detection starting point coordinates of the distance sensor are jointly determined by the current pose of the intelligent mobile robot (including X / Y coordinates and heading angle) and the unique position number of the sowing block—that is, by looking up the table based on the grid division results to obtain the theoretical coordinates (x0, y0) of the sowing position of the corresponding sowing block in the global field coordinate system, and then combining the spatial transformation relationship between the intelligent mobile robot's body coordinate system and the global coordinate system, the real-time spatial origin of the sensor probe in the global coordinate system is calculated. "Actual growth position" refers to the two-dimensional spatial coordinates of the visible stem base of the rapeseed plant in the field plane coordinate system, which can be obtained by inferring the distance value measured by a distance sensor; "offset" refers to the Euclidean distance between the actual growth position (x, y) and the corresponding sowing position (x0, y0) of the sowing block, i.e. This value characterizes the degree of spatial deviation of the plant from the ideal sowing point and can directly reflect whether it is at the expected growth site.

[0057] The "preset threshold" is set based on a comprehensive consideration of the optimal plant spacing at the rapeseed seedling development stage, soil looseness, repeatability of the sowing machinery, and distance sensor measurement error. For example, it can be 15% to 35% of the optimal plant spacing, with typical values ​​ranging from 2.5 cm to 8.0 cm. In this embodiment, the "preset threshold" is pre-written into the configuration parameter table of the computational processing module and can be remotely updated via a host computer. When the offset of a rapeseed plant exceeds this threshold, the system immediately generates a structured judgment record locally, which includes at least: the sowing block location number, the judgment timestamp, the measured offset value, and the judgment conclusion of "hybrid plant". This record simultaneously triggers a skip instruction, blocking the execution of subsequent image acquisition, feature extraction, and statistical analysis processes for that rapeseed plant.

[0058] This embodiment introduces a distance sensor for pre-detection of plant spatial position and uses the offset between the actual growth position and the theoretical sowing position as the first-level criterion. This enables rapid identification of spatially abnormal individuals caused by non-genetic factors such as sowing deviation, wind disturbance, animal trampling, or invasion of foreign weeds. Using this mechanism, preliminary screening of some weeds can be completed without utilizing image recognition modules and statistical computing resources, significantly reducing the overall system computational load and image acquisition frequency. Furthermore, the visual basic feature extraction process is initiated only for plants whose offset does not exceed the limit, allowing limited perception and computing resources to focus on objects that truly require refined judgment. This improves the throughput of weed identification per unit time and the real-time performance of the system, enhancing its adaptability to large-scale rapeseed planting scenarios.

[0059] Example 3

[0060] This embodiment provides a method for efficient removal of hybrid plants during rapeseed seed production. Compared with Embodiment 1, it further provides a specific process for obtaining the basic visual characteristics of each rapeseed plant in step S2, and then using statistical methods to calculate the basic visual characteristics, including:

[0061] Step S21A: Select all or part of the rapeseed plants as the feature statistical sample; if part of the rapeseed plants are selected as the feature statistical sample, the sample size shall not be less than 1% of the total number of rapeseed plants. For example, when the rapeseed field area is large and the total number of plants exceeds 100,000, the 30% sample size shall not be less than 30,000 plants.

[0062] Step S22A: Perform statistical operations on each feature item of the visual basic features of each rapeseed plant in the feature statistical sample, calculate the mean μ1 and standard deviation σ1 of the feature item, and use μ1±σ1 or μ1±2σ1 as the first hybrid plant judgment threshold for the corresponding feature item.

[0063] Among them, "mean μ1" represents the central tendency of the feature item in the statistical sample, and "standard deviation σ1" represents its dispersion; "μ1±σ1" corresponds to the normal fluctuation range at the confidence level (e.g., 68%), and "μ1±2σ1" corresponds to the normal fluctuation range at the high confidence level (e.g., 95%); both are open threshold forms, which can be selected and enabled according to the actual identification accuracy requirements in the field, without the need to pre-set fixed values; this threshold form does not rely on any prior knowledge or historical models, and is dynamically generated entirely from the field data collected in the current batch of operations.

[0064] Step S23A: Verify the basic visual features of all rapeseed plants one by one. If any feature item in the basic visual features of a single rapeseed plant exceeds the corresponding first hybrid plant determination threshold, the current rapeseed plant is determined to be a "hybrid". Rapeseed plants that do not exceed the first hybrid plant determination threshold are determined to be "normal plants".

[0065] The "plant-by-plant verification" refers to comparing the extracted plant height, average crown width, and RGB value of the main leaf color of each rapeseed plant uniquely bound by grid number in this application with their respective first hybrid plant judgment threshold ranges. This comparison operation is a simple numerical relationship judgment (>, <) and does not involve complex pattern matching or classification decisions. "Any feature item exceeds" indicates that the judgment logic is an "OR" relationship. As long as one item falls outside the threshold range, the hybrid plant marking process is initiated to ensure high sensitivity to individuals that significantly deviate from the population phenotype. This mechanism does not require multiple features to be abnormal at the same time, adapting to the reality that hybrid plants may only appear in a single dimension (such as leaf color mutation but normal plant height).

[0066] This embodiment selects representative feature statistical samples and dynamically constructs statistical distribution models for each item based on visual basic feature data collected on-site. An adaptive first hybrid plant determination threshold is generated by combining the mean and standard deviation. On this basis, by comparing each item with all plants one by one, a hybrid plant identification mechanism is realized that does not require pre-trained models, does not rely on manual annotation, and is not limited by prior knowledge of varieties and plots. This mechanism, with the inherent robustness and generalization ability of statistics, suppresses false alarms caused by environmental disturbances (such as local lighting differences and lens distortion) while ensuring identification sensitivity. This makes the hybrid plant determination results more consistent with the actual field phenotypic distribution, thereby supporting the stable achievement of the "efficient, accurate, and low-cost removal" technical effect in this application.

[0067] Meanwhile, this embodiment also provides that before the removal execution module starts removing the rapeseed plants confirmed as "hybrids" in step S3, the image recognition module switches the shooting angle and re-shoots, then extracts the basic visual features again, and compares them with the first hybrid plant determination threshold again; if any feature item in the basic visual features still exceeds the corresponding first hybrid plant determination threshold, then it is finally confirmed as a "hybrid" and directly removed; if it does not exceed the first hybrid plant determination threshold, then the removal is paused, and a warning or record is given.

[0068] "Switching the shooting angle" refers to the image recognition module changing its spatial observation orientation relative to the rapeseed plant, including but not limited to: switching from a vertical overhead shot to a 45° side shot, switching from a single-sided level view to a diagonal shot, or switching from a fixed height shot to a multi-height combination shot after adjustment. This action does not rely on 3D reconstruction or multi-view stereo matching, but only requires adjusting the gimbal posture of the image recognition module or the pose of the intelligent mobile robot. Its technical purpose is to avoid the deviation in the extraction of basic visual features caused by morphological distortion, partial occlusion, or uneven lighting due to a single imaging perspective.

[0069] "Re-extraction" refers to independently executing the same visual basic feature acquisition process as in step S2 on the new perspective image: performing background separation, target region localization, pixel span measurement, and channel mean calculation on the new image; this process does not reuse the intermediate results of the previous image, nor does it introduce cross-perspective feature fusion logic; its technical purpose is to obtain independent feature samples of the same plant under different observation conditions, forming an orthogonal verification of the initial screening results.

[0070] This embodiment embeds a two-stage verification mechanism of "initial screening - verification" before the elimination process. By leveraging the changes in imaging conditions caused by switching shooting angles, the basic visual features of the same plant can be represented by multiple independent characteristics. On this basis, the first hybrid plant judgment threshold established in step S22A is reused for consistency comparison. This avoids the computational overhead caused by model iteration updates and effectively suppresses the influence of inherent errors in single-view imaging on the final decision. Ultimately, without increasing the system hardware complexity and algorithm deployment cost, the robustness and engineering reliability of hybrid plant identification results are significantly improved, effectively ensuring the purity of rapeseed seed production.

[0071] Example 4

[0072] This embodiment provides a method for efficient removal of hybrid plants during rapeseed seed production. Compared with Embodiment 1, it further provides a specific process in step S2 of obtaining the basic visual characteristics of each rapeseed plant and then using statistical methods to calculate the basic visual characteristics, including:

[0073] Step S21B: Based on the grid division results, field microenvironment data of each sowing block, and the operation batch of the intelligent mobile robot, the rapeseed field is divided into several secondary comparison units. The secondary comparison units include microenvironment homogeneity units and operation batch verification units. The microenvironment homogeneity units include rapeseed plants corresponding to adjacent sowing blocks with consistent soil fertility and light conditions within the grid. The operation batch verification unit consists of several adjacent microenvironment homogeneity units and is adapted to the coverage range of a single operation of the intelligent mobile robot.

[0074] The “microenvironment homogenization unit” refers to a set of rapeseed plants corresponding to a group of sowing blocks that are spatially adjacent and whose field microenvironment parameters (including but not limited to soil fertility and light conditions within the grid) differ below a preset homogenization threshold. This reflects the relative stability of the rapeseed plant growth background within the same small area. Its technical function is to provide a comparable local reference system for visual basic features, making the statistical distribution of features such as plant height, average crown width, and leaf RGB value closer to the actual biological variation law. In this embodiment, this unit serves as the basic input range for the first-level statistical analysis, ensuring that the mean μ2 and standard deviation σ2 obtained in subsequent calculations can truly reflect the phenotypic fluctuation boundary of normal plants under homogenous conditions.

[0075] "Work batch verification unit" refers to a work area formed by the continuous spatial aggregation of multiple homogeneous microenvironment units, which matches the single full-charge endurance or single task payload capacity of the intelligent mobile robot. Its technical function is to construct a higher-level verification scale to verify whether "suspected hybrids" still exhibit abnormal consistency over a larger area. In this embodiment, this unit serves as the input range for the second-level statistical analysis. Its coverage area is larger than that of the homogeneous microenvironment units but smaller than that of the entire field area, thus balancing environmental representativeness and computational efficiency, and supporting cross-unit robustness verification of suspected individuals.

[0076] This embodiment explicitly models the heterogeneity of the field microenvironment as a structured unit that can be divided, mapped, and statistically hierarchically. This makes the discrimination criterion of visual basic features no longer dependent on a uniform threshold across the entire field. Instead, it dynamically generates dual local and secondary local criteria based on environmental similarity. This effectively suppresses systematic phenotypic shifts caused by regional fertility gradients, shade belts, or uneven irrigation, and avoids misclassifying normal variations driven by the environment as genetic hybrids.

[0077] Step S22B: Obtain the basic visual features of each rapeseed plant within each homogenized microenvironment unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ2 and standard deviation σ2 of each feature item. Use μ2±σ2 or μ2±2σ2 as the second hybrid plant determination threshold for the corresponding feature item. Verify the basic visual features of all rapeseed plants within the same homogenized microenvironment unit plant by plant. If any feature item of the basic visual features of a single rapeseed plant exceeds the corresponding second hybrid plant determination threshold, the current rapeseed plant is determined to be a "suspected hybrid plant". Rapeseed plants that do not exceed the second hybrid plant determination threshold are determined to be "normal plants".

[0078] The "second hybrid plant determination threshold" is a discrete boundary calculated independently for each of the three feature components—plant height, average crown width, and RGB value of leaf primary color—of all rapeseed plants within a homogeneous microenvironment unit. This threshold is not set as a fixed value but is dynamically generated based on actual observation data within the unit, reflecting the central trend and natural fluctuation range of the phenotypic distribution within the unit. Its technical function is to establish the first screening defense line, identifying individuals that significantly deviate from the phenotypic center of the population under the same microenvironment. In this embodiment, the threshold is directly derived from the raw visual basic feature data output by the image recognition module, performing only basic descriptive statistics with low computational load, and can be completed in real time in the embedded processor of the computing module. When μ2±σ2 is used, it covers approximately 68% of the normally distributed samples, suitable for scenarios with highly homogeneous microenvironments and low noise. When μ2±2σ2 is used, it covers approximately 95% of the normally distributed samples, suitable for scenarios with mild field disturbance (such as local pests or slight trampling). The two forms can be automatically switched based on historical field data or manually configured by agronomists.

[0079] "Suspected hybrid plants" are individuals that exhibit significant outliers in at least one visual basic feature category relative to other plants within a homogeneous microenvironment unit. This status is not a final determination, but an intermediate marker used to trigger secondary verification within subsequent batch verification units. Its technical role is to reduce the number of high-confidence verification objects, avoid redundant multi-level statistical operations on all plants in the field, and improve overall process efficiency.

[0080] Step S23B: Obtain the basic visual features of each rapeseed plant in each work batch verification unit, and then perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ3 and standard deviation σ3 of each feature item. Use μ3±σ3 or μ3±2σ3 as the third hybrid plant determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants determined as "suspected hybrid plants" in step S22B with the third hybrid plant determination threshold. If any feature item in the basic visual features exceeds the corresponding third hybrid plant determination threshold, it is determined as a "hybrid plant".

[0081] The "third hybrid plant determination threshold" is a statistical boundary recalculated at the scale of the work batch verification unit. Its coverage is larger than the homogeneous microenvironment unit, but it still maintains spatial proximity and operational logic coherence. The technical function of this threshold is to provide a collaborative verification mechanism across microenvironment units to check whether the anomaly of "suspected hybrid plants" has cross-unit consistency. If a plant is abnormal within its own microenvironment unit, but is still at the edge of the overall distribution in the larger-scale work batch unit, its anomaly is more likely to originate from genetic essence rather than local environmental disturbance. In this embodiment, the threshold only applies to a finite subset that has been marked as "suspected hybrid plants" and does not participate in the full plant calculation, significantly reducing the consumption of computing resources. μ3±σ3 is suitable for scenarios where the microenvironment difference within the work batch is small and the robot positioning accuracy is high. μ3±2σ3 is suitable for scenarios where there are cross-unit microenvironmental gradients (such as slope illumination gradients) or systematic deviations in image acquisition angles. The two forms can be dynamically selected based on the GPS / RTK positioning accuracy and IMU attitude data quality of the intelligent mobile robot.

[0082] This embodiment constructs a two-level comparison architecture of "microenvironment homogenization unit → operation batch verification unit" and sets a second and a third heterogeneous plant judgment threshold, respectively, to achieve environmental adaptability of visual basic feature discrimination: first, preliminary outlier detection is completed in the smallest comparable unit, and then anomaly robustness verification is completed in a larger unit with operational logical significance; the thresholds at both levels are generated by local measured data, without relying on prior models or external training samples, which avoids the deployment cost of deep learning models and improves the generalization ability to complex field scenarios; finally, only plants that simultaneously meet the dual conditions of "outlier within the unit" and "robust anomaly within the batch" are identified as "heterogeneous plants", which greatly reduces the false positive rate caused by environmental fluctuations and ensures seed purity and impurity removal reliability.

[0083] Meanwhile, this embodiment also provides that before the removal execution module starts removing the rapeseed plants confirmed as "hybrids" in step S3, the image recognition module switches the shooting angle and re-shoots the plant, then extracts the basic visual features again, and compares them with the second and / or the third hybrid plant determination threshold. If any feature item in the basic visual features still exceeds the corresponding second and / or the third hybrid plant determination threshold, it is finally confirmed as a "hybrid" and directly removed. If it does not exceed the second and / or the third hybrid plant determination threshold, the removal is paused, and a warning or record is given.

[0084] This embodiment, based on the two-level statistical judgment framework established in this application, introduces a shooting perspective switching and dual-threshold cross-verification mechanism, so that the judgment of "hybrid plants" no longer depends on a single static image and a single-scale statistical result. By using the reproducibility verification of visual basic features under different spatial observation conditions, it effectively suppresses false positive identification caused by individual posture changes, local shadow interference, or sensor transient drift. At the same time, the second and third hybrid plant judgment thresholds are anchored to two reference systems of different granularities: the microenvironment homogenization unit and the work batch verification unit, respectively, so that the verification process has both local sensitivity and batch robustness. Finally, under the premise of ensuring that the removal action only applies to plants with phenotypic abnormalities that are consistent across perspectives and units, it significantly improves the accuracy of hybrid plant identification and the fault tolerance safety of agricultural production, and effectively implements the seed production quality management principle of "cautious removal, rather than wrongful removal than missed judgment".

[0085] Example 5

[0086] This embodiment provides a method for efficient removal of hybrid plants during rapeseed seed production. Compared with Embodiment 1, it further provides a specific process for obtaining the basic visual characteristics of each rapeseed plant in step S2, and then using statistical methods to calculate the basic visual characteristics, including:

[0087] Step 21C: Based on the grid division results, the rapeseed field is divided into basic units, combined units, and whole-field area units according to the total area size from small to large. The basic unit includes rapeseed plants corresponding to several adjacent planting blocks; the combined unit is composed of several adjacent basic units; and the whole-field area unit is composed of several adjacent combined units, which is the complete rapeseed field.

[0088] The basic unit is the smallest spatial unit for statistical analysis, divided based on the similarity of adjacent sown plots in the field microenvironment (including soil fertility gradient, local light distribution, and irrigation uniformity). It is used to characterize the range of natural phenotypic variation of rapeseed plants under similar growth conditions. The combined unit serves as an intermediate-scale analysis level, aggregating basic units with slight environmental differences but consistent overall trends to enhance the representativeness of statistical samples and suppress local random biases. The whole-field region unit is the largest-scale analysis level, covering all sown plots, used to establish the baseline phenotypic distribution of the rapeseed population across the entire field. These three units constitute a nested spatial structure from local to global and from micro to macro, enabling subsequent multi-level threshold determinations to have spatial comparability and logical progression. For example, in this embodiment, the basic unit is designed to contain 3×3 adjacent sown plots, the combined unit consists of 4 adjacent basic units, and the whole-field region unit is the entire rapeseed field.

[0089] Step S22C: Obtain the basic visual features of each rapeseed plant in each basic unit, and then perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ4 and standard deviation σ4 of each feature item. Use μ4±σ4 or μ4±2σ4 as the fourth hybrid plant determination threshold for the corresponding feature item. Verify the basic visual features of all rapeseed plants in the same basic unit one by one. If any feature item in the basic visual features of a single rapeseed plant exceeds the corresponding fourth hybrid plant determination threshold, the current rapeseed plant is determined to be a "primary suspected hybrid plant". Rapeseed plants that do not exceed the fourth hybrid plant determination threshold are determined to be "normal plants".

[0090] Here, "each feature item of the basic visual features" refers to the data sequence of three independent dimensions: plant height, average crown width, and RGB value of the leaf's primary color. The fourth heterophyte detection threshold is a preliminary anomaly identification boundary set for homogeneous groups within the basic unit. Its function is to quickly capture individuals that significantly deviate from the phenotype of similar plants in a local microenvironment. This threshold does not rely on prior labels or training data; it is automatically constructed based solely on the degree of dispersion of the measured basic visual features within the current basic unit. For example, in this embodiment, the statistical operation uses an unbiased estimation method to calculate σ4, and μ4±σ4 is used as the default threshold.

[0091] Step S23C: Obtain the basic visual features of each rapeseed plant in each combined unit, and then perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ5 and standard deviation σ5 of each feature item. Use μ5±σ5 or μ5±2σ5 as the fifth hybrid determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants that were determined to be "primary suspected hybrids" in step S22C with the fifth hybrid determination threshold. If any feature item in the basic visual features exceeds the corresponding fifth hybrid determination threshold, it is determined to be "secondary suspected hybrid".

[0092] The fifth hybrid determination threshold is a statistical boundary reconstructed at a higher spatial scale. Its statistical sample covers multiple basic units, thus exhibiting greater robustness and generalization ability. This step is essentially a secondary verification of "primary suspected hybrids," aiming to eliminate single-point anomalies caused by local occlusion, shooting angle deviation, or instantaneous light disturbance. In one optional implementation, the statistical operation within the combined unit only applies to all plants within the basic unit where the plant marked as "primary suspected hybrid" is located and its two adjacent basic units. In another optional implementation, the fifth hybrid determination threshold is dynamically updated using a sliding window method, with the window covering three continuously scrolling basic units. Furthermore, the "comparison" operation in this step can be performed by independently judging each feature separately, or by uniformly judging after weighted fusion of multiple features. The weights can be set according to the empirical values ​​of the discriminative power of each feature in seedling hybrid identification.

[0093] Step S24C: Obtain the basic visual features of each rapeseed plant in each whole-field area unit, and then perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ6 and standard deviation σ6 of each feature item. Use μ6±σ6 or μ6±2σ6 as the sixth hybrid plant determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants that were determined to be "secondary suspected hybrid plants" in step S23C with the sixth hybrid plant determination threshold. If any feature item in the basic visual features exceeds the corresponding sixth hybrid plant determination threshold, it is determined to be a "hybrid plant".

[0094] Among them, the "whole field regional unit" corresponds to the complete rapeseed planting field, and its statistical results reflect the overall phenotypic central trend and natural fluctuation range of the rapeseed population in the whole field; the sixth hybrid plant judgment threshold is the final decision threshold, representing strong anomalous individuals that continue to deviate from the normal distribution of the population at the global scale; this step realizes the semantic dimensionality upgrade from "spatial local anomaly" to "global continuous anomaly", ensuring that plants judged as "hybrid plants" are not only abnormal in a small range, but also lack phenotypic rationality in the entire planting system.

[0095] This embodiment constructs a three-level spatial partitioning structure of "basic unit → combined unit → whole-field region unit," and performs statistical modeling and threshold determination of feature items at each level, forming a fine-to-coarse, progressively converging path for identifying hybrid plants. It captures local sensitive anomalies using basic units, suppresses accidental disturbances using combined units, and finally anchors global stability criteria using whole-field region units. This ensures that a single rapeseed plant must pass three consecutive statistical tests at different scales to be finally identified as a "hybrid." This avoids the oversensitivity or undersensitivity problems caused by single-scale thresholds and achieves high-precision, low-false-judgment hybrid plant screening without the need for deep learning models. This mechanism, together with the gridded sowing, positional offset initial screening, and visual basic feature extraction in this application, constitutes a logically consistent, technically decoupled, and lightweight seedling-stage hybrid plant removal technology system.

[0096] Meanwhile, in this embodiment, before the removal execution module is activated in step S3 to remove the rapeseed plants confirmed as "hybrids", the image recognition module switches the shooting angle and re-shoots the image, then extracts the basic visual features again and compares them with the fourth, fifth, and / or sixth hybrid plant determination thresholds. If any feature item in the basic visual features still exceeds the corresponding fourth, fifth, and / or sixth hybrid plant determination thresholds, it is finally confirmed as a "hybrid" and directly removed. If it does not exceed the fourth, fifth, and / or sixth hybrid plant determination thresholds, the removal is paused, and a warning or record is issued.

[0097] This embodiment adds a multi-angle review step after the three-level progressive judgment. By combining independently collected visual basic features with multi-level judgment thresholds, the final confirmation of the "hybrid plant" identity is achieved. Based on this, the removal, warning, or recording actions are dynamically selected according to the review results. This not only effectively avoids the risk of false removal caused by abnormal single image acquisition, but also leaves a response window for human-machine collaborative decision-making. Ultimately, without increasing the dependence on model training and cloud computing power, the reliability, interpretability, and traceability of the hybrid plant removal action are significantly improved, supporting the stable implementation of high-purity rapeseed seed production operations.

[0098] Example 6

[0099] This embodiment provides a highly efficient system for removing hybrid plants during rapeseed seed production and cultivation. The system includes:

[0100] The system includes an intelligent mobile robot, and an image recognition module, a weed removal execution module, and a computational processing module mounted on and electrically connected to the intelligent mobile robot. After receiving instructions, the intelligent mobile robot moves along a preset path in the rapeseed field. The image recognition module is used to photograph the rapeseed plants and obtain the basic visual features of each plant. The weed removal execution module is used to remove rapeseed plants identified as "hybrids". The computational processing module receives the basic visual features and then uses statistical methods to perform calculations on the basic visual features to filter out the rapeseed plants identified as "hybrids".

[0101] The various modules in the efficient hybrid plant removal system during the rapeseed seed production process described above can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the intelligent mobile robot in hardware form or independent of it, or stored in the memory of the intelligent mobile robot in software form, so that the processor can call and execute the corresponding operations of each module.

[0102] The intelligent mobile robot can be a tracked, wheeled, or legged mobile platform. Its chassis size is adapted to the size parameters of the intelligent mobile robot on which the grid division in this application is based. It can move stably along the crisscrossing grid lines and has positioning functions (such as integrating GPS+RTK differential positioning and inertial navigation). The positioning accuracy is better than ±5 cm to ensure that it can accurately reach the location number area corresponding to each sowing block. The intelligent mobile robot is also equipped with a power management unit, which supports continuous operation time of no less than 4 hours, adapting to the needs of continuous daytime operation in the field environment. Its structural strength and protection level meet IP67, and it can withstand field mud, dust and short-term rain conditions.

[0103] The image recognition module includes a visible light camera, a supplementary light source, and an embedded image processing unit. The visible light camera is installed at the front end of the intelligent mobile robot or on a liftable gimbal, supporting fixed-height vertical shooting. The installation height can be adjusted according to the pre-set plant height based on the development stage of the rapeseed seedlings, for example, 30 cm to 50 cm above the ground during the 2-4 leaf stage. The supplementary light source is a ring LED array with a color temperature of 5500 K and an illuminance uniformity of ≥90%, used to eliminate shadow interference and ensure the accuracy of leaf RGB value extraction. The image acquisition control unit is communicatively connected to the processing module, responding to trigger commands to start single-frame or multi-frame image acquisition, and transmitting the raw image data to the processing module in real time.

[0104] The weed removal module includes a robotic arm, an end effector, and a drive control unit. The robotic arm is a lightweight, multi-degree-of-freedom structure with a working radius covering a range of ±15 cm from the center of a single sowing block. The end effector can be a pneumatic gripper, a rotary cutting tool, or a negative pressure suction head. The specific selection can be determined based on the morphology of the rapeseed seedlings and soil conditions. For example, in soft loam, a negative pressure suction head can be used to remove the seedlings by the roots, while in compacted soil, a miniature rotary tool can be used to cut the stems near the ground. The drive control unit receives the removal instructions and target position coordinates output by the processing module, drives the robotic arm to move precisely, and performs the removal action with a cycle of ≤3 s / plant.

[0105] The computational processing module includes a microcontroller, an embedded AI acceleration chip, a storage unit, and communication interfaces. The microcontroller is the main control unit, such as an STM32H743 or NXP i.MX 8M Mini, with a main frequency of ≥800 MHz and supporting floating-point operations. The embedded AI acceleration chip is an optional configuration, only activated when a lightweight auxiliary model (such as background segmentation preprocessing) is required, and is not essential. The storage unit includes eMMC flash memory (≥8 GB) and LPDDR4 memory (≥512 MB), used to cache image data, store statistical threshold parameters, and historical judgment records. The communication interfaces include a CAN bus (connecting to the robot chassis controller), MIPI CSI-2 (connecting to the image recognition module), and GPIO / PWM (connecting to the noise removal execution module driver). The computational processing module runs an embedded Linux or FreeRTOS operating system, loads locally deployed statistical analysis programs, and does not rely on external networks or cloud services. All extraction of basic visual features and statistical calculations are completed locally.

[0106] The above technical solution uses an intelligent mobile robot as the physical carrier and motion execution center. The image recognition module completes the perception input, the computing and processing module completes the localized, model-free statistical decision-making, and the impurity removal execution module completes the terminal action output. The four are electrically connected to form a closed-loop control system. The modules work together to enable the system to autonomously complete the entire process from image acquisition, feature extraction, impurity identification to physical removal without network dependence, high-end AI chips, or human intervention. This effectively supports the implementation of various impurity removal methods defined in the above embodiments, significantly improves the efficiency and accuracy of impurity removal operations during the rapeseed seed production seed stage, reduces equipment deployment and maintenance costs, and has good field applicability and industrialization promotion value.

Claims

1. A method for efficient removal of volunteer plants in the breeding process of oilseed rape, characterized in that, Includes the following steps: Step S1: Obtain the shape and area parameters of the rapeseed field, and based on the shape and area parameters, combined with the preset optimal plant spacing and the size of the intelligent mobile robot, divide the field into grids. Each grid is a sowing block and is assigned a unique location number. The crisscrossing grid lines are the movement path of the intelligent mobile robot. Sow a preset number of rapeseed seeds in each sowing block. Step S2: Once the rapeseed seeds have grown and developed to a preset stage, the intelligent mobile robot is driven to move along the movement path. The image recognition module on the intelligent mobile robot takes pictures of the rapeseed plants to obtain the basic visual features of each rapeseed plant. Then, statistical methods are used to calculate the basic visual features and screen out the rapeseed plants identified as "hybrids". The basic visual features include at least plant height, average crown width, and RGB value of the main leaf color. When the image recognition module takes pictures, it adopts a fixed posture and maintains a constant shooting angle through real-time tilt compensation. At the same time, it performs scale normalization and perspective correction on the captured images based on the physical size of the grid of the sowing area, and combines the relative elevation of the ground of the plot to unify the plant height measurement benchmark, so that the equivalent shooting distance and imaging scale between the lens and each plant are consistent. Step S3: The intelligent mobile robot moves along the moving path to the vicinity of the sowing block corresponding to the position number of the rapeseed plant identified as a "hybrid", and activates the weed removal execution module carried by the intelligent mobile robot to remove it. The specific process of step S2, which involves acquiring the basic visual features of each rapeseed plant and then performing statistical calculations on these features, includes: Step S21A: Select all or part of the rapeseed plants as a feature statistical sample; if a portion of the rapeseed plants are selected as the feature statistical sample, the sample size shall not be less than 1% of the total number of rapeseed plants. Step S22A: Perform statistical operations on each feature item of the visual basic features of each rapeseed plant in the feature statistical sample to calculate the mean μ1 and standard deviation σ1 of the feature item, and use μ1±σ1 or μ1±2σ1 as the first hybrid plant determination threshold corresponding to the feature item. Step S23A: Verify the basic visual features of all rapeseed plants one by one. If any feature item in the basic visual features of a single rapeseed plant exceeds the corresponding first hybrid plant determination threshold, the current rapeseed plant is determined to be a "hybrid". For rapeseed plants that do not exceed the first hybrid plant determination threshold, they are determined to be "normal plants".

2. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 1, characterized in that, Before obtaining the basic visual features of each rapeseed plant in step S2, the rapeseed plant position is determined. Specifically, the actual growth position of each rapeseed plant is detected by the distance sensor carried by the intelligent mobile robot, combined with the unique location number of the sowing block, and the offset between the actual growth position and the sowing position of the corresponding sowing block is calculated. If the offset is greater than a preset threshold, the rapeseed plant is directly determined to be a "hybrid plant", and there is no need to perform the subsequent basic visual feature extraction and statistical calculation.

3. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 1, characterized in that, The specific method for obtaining the basic visual features in step S2 is as follows: Plant height: The rapeseed plant is photographed by the image recognition module to obtain an elevation view and a top view. The pixel coordinates of the highest point of the rapeseed plant are located by edge detection and contour extreme point extraction. The actual plant height of the rapeseed plant is obtained by combining the intrinsic and extrinsic parameters including camera focal length, pixel size, shooting distance and shooting posture correction coefficient. Average crown width: Extract the pixel outline of the rapeseed plant crown in the top view, calculate the horizontal pixel span and vertical pixel span of the outline, take the average of the two and combine them with the pixel-actual size conversion ratio to obtain the actual crown width average. Leaf primary color RGB value: Remove background pixels from the top view and keep only fresh leaf pixels. Extract the RGB channel values ​​of the fresh leaf pixels and calculate the average value as the leaf primary color RGB value.

4. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 1, characterized in that, Before the removal module is activated in step S3 to remove rapeseed plants identified as "hybrids", the image recognition module switches the shooting angle and re-shoots the plant, extracting the basic visual features again and comparing them with the first hybrid plant determination threshold. If any feature item in the basic visual features still exceeds the corresponding first hybrid plant determination threshold, it is finally identified as a "hybrid" and directly removed. If it does not exceed the first hybrid plant determination threshold, the removal is paused, and a warning or record is issued.

5. A method for efficiently removing hybrid plants during rapeseed seed production and cultivation, characterized in that, Includes the following steps: Step S1: Obtain the shape and area parameters of the rapeseed field, and based on the shape and area parameters, combined with the preset optimal plant spacing and the size of the intelligent mobile robot, divide the field into grids. Each grid is a sowing block and is assigned a unique location number. The crisscrossing grid lines are the movement path of the intelligent mobile robot. Sow a preset number of rapeseed seeds in each sowing block. Step S2: Once the rapeseed seeds have grown and developed to a preset stage, the intelligent mobile robot is driven to move along the movement path. The image recognition module on the intelligent mobile robot takes pictures of the rapeseed plants to obtain the basic visual features of each rapeseed plant. Then, statistical methods are used to calculate the basic visual features and screen out the rapeseed plants identified as "hybrids". The basic visual features include at least plant height, average crown width, and RGB value of the main leaf color. When the image recognition module takes pictures, it adopts a fixed posture and maintains a constant shooting angle through real-time tilt compensation. At the same time, it performs scale normalization and perspective correction on the captured images based on the physical size of the grid of the sowing area, and combines the relative elevation of the ground of the plot to unify the plant height measurement benchmark, so that the equivalent shooting distance and imaging scale between the lens and each plant are consistent. Step S3: The intelligent mobile robot moves along the moving path to the vicinity of the sowing block corresponding to the position number of the rapeseed plant identified as a "hybrid", and activates the weed removal execution module carried by the intelligent mobile robot to remove it. The specific process of step S2, which involves acquiring the basic visual features of each rapeseed plant and then performing statistical calculations on these features, includes: Step S21B: Based on the grid division results, the field microenvironment data of each sowing block, and the operation batch of the intelligent mobile robot, the rapeseed field is divided into several secondary comparison units. Each secondary comparison unit includes a microenvironment homogenization unit and an operation batch verification unit. The microenvironment homogenization unit includes rapeseed plants corresponding to adjacent sowing blocks with consistent soil fertility and light conditions within the grid. The operation batch verification unit is composed of several adjacent microenvironment homogenization units and is adapted to the coverage range of a single operation of the intelligent mobile robot. Step S22B: Obtain the basic visual features of all rapeseed plants within each microenvironment homogenization unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ2 and standard deviation σ2 of each feature item. Use μ2±σ2 or μ2±2σ2 as the second hybrid plant determination threshold for the corresponding feature item. Verify the basic visual features of all rapeseed plants within the same microenvironment homogenization unit plant by plant. If any feature item of the basic visual features of a single rapeseed plant exceeds the corresponding second hybrid plant determination threshold, the current rapeseed plant is determined to be a "suspected hybrid plant". Rapeseed plants that do not exceed the second hybrid plant determination threshold are determined to be "normal plants". Step S23B: Obtain the visual basic features of all rapeseed plants within each batch verification unit. Then, perform statistical operations on each feature item of the visual basic features of each rapeseed plant to calculate the mean μ3 and standard deviation σ3 of each feature item. Use μ3±σ3 or μ3±2σ3 as the third hybrid plant determination threshold for the corresponding feature item. Compare the corresponding visual basic features of the rapeseed plants determined as "suspected hybrid plants" in step S22B with the third hybrid plant determination threshold. If any feature item in the visual basic features exceeds the corresponding third hybrid plant determination threshold, it is determined as a "hybrid plant".

6. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 5, characterized in that, Before obtaining the basic visual features of each rapeseed plant in step S2, the rapeseed plant position is determined. Specifically, the actual growth position of each rapeseed plant is detected by the distance sensor carried by the intelligent mobile robot, combined with the unique location number of the sowing block, and the offset between the actual growth position and the sowing position of the corresponding sowing block is calculated. If the offset is greater than a preset threshold, the rapeseed plant is directly determined to be a "hybrid plant", and there is no need to perform the subsequent basic visual feature extraction and statistical calculation.

7. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 5, characterized in that, The specific method for obtaining the basic visual features in step S2 is as follows: Plant height: The rapeseed plant is photographed by the image recognition module to obtain an elevation view and a top view. The pixel coordinates of the highest point of the rapeseed plant are located by edge detection and contour extreme point extraction. The actual plant height of the rapeseed plant is obtained by combining the intrinsic and extrinsic parameters including camera focal length, pixel size, shooting distance and shooting posture correction coefficient. Average crown width: Extract the pixel outline of the rapeseed plant crown in the top view, calculate the horizontal pixel span and vertical pixel span of the outline, take the average of the two and combine them with the pixel-actual size conversion ratio to obtain the actual crown width average. Leaf primary color RGB value: Remove background pixels from the top view and keep only fresh leaf pixels. Extract the RGB channel values ​​of the fresh leaf pixels and calculate the average value as the leaf primary color RGB value.

8. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 5, characterized in that, Before the removal module is activated in step S3 to remove rapeseed plants identified as "hybrids", the image recognition module switches the shooting angle and re-shoots the plant, then extracts the basic visual features again and compares them with the second and / or the third hybrid determination threshold. If any feature item in the basic visual features still exceeds the corresponding second and / or third hybrid determination threshold, the plant is ultimately identified as a "hybrid" and removed directly. If the feature does not exceed the second and / or the third hybrid determination threshold, the removal process is paused, and a warning or record is issued.

9. A method for efficiently removing hybrid plants during rapeseed seed production and cultivation, characterized in that, Includes the following steps: Step S1: Obtain the shape and area parameters of the rapeseed field, and based on the shape and area parameters, combined with the preset optimal plant spacing and the size of the intelligent mobile robot, divide the field into grids. Each grid is a sowing block and is assigned a unique location number. The crisscrossing grid lines are the movement path of the intelligent mobile robot. Sow a preset number of rapeseed seeds in each sowing block. Step S2: Once the rapeseed seeds have grown and developed to a preset stage, the intelligent mobile robot is driven to move along the movement path. The image recognition module on the intelligent mobile robot takes pictures of the rapeseed plants to obtain the basic visual features of each rapeseed plant. Then, statistical methods are used to calculate the basic visual features and screen out the rapeseed plants identified as "hybrids". The basic visual features include at least plant height, average crown width, and RGB value of the main leaf color. When the image recognition module takes pictures, it adopts a fixed posture and maintains a constant shooting angle through real-time tilt compensation. At the same time, it performs scale normalization and perspective correction on the captured images based on the physical size of the grid of the sowing area, and combines the relative elevation of the ground of the plot to unify the plant height measurement benchmark, so that the equivalent shooting distance and imaging scale between the lens and each plant are consistent. Step S3: The intelligent mobile robot moves along the moving path to the vicinity of the sowing block corresponding to the position number of the rapeseed plant identified as a "hybrid", and activates the weed removal execution module carried by the intelligent mobile robot to remove it. The specific process of step S2, which involves acquiring the basic visual features of each rapeseed plant and then performing statistical calculations on these features, includes: Step 21C: Based on the grid division results, the rapeseed field is divided into basic units, combined units, and whole-field area units according to the total area size from small to large. The basic unit includes several rapeseed plants corresponding to several adjacent planting blocks. The combined unit is composed of several adjacent basic units, and the whole-field area unit is composed of several adjacent combined units, which is the complete rapeseed field. Step S22C: Obtain the basic visual features of each rapeseed plant within each basic unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ4 and standard deviation σ4 of each feature item. Use μ4±σ4 or μ4±2σ4 as the fourth hybrid plant determination threshold for the corresponding feature item. Verify the basic visual features of all rapeseed plants within the same basic unit plant by plant. If any feature item of the basic visual features of a single rapeseed plant exceeds the corresponding fourth hybrid plant determination threshold, the current rapeseed plant is determined to be a "primary suspected hybrid plant". Rapeseed plants that do not exceed the fourth hybrid plant determination threshold are determined to be "normal plants". Step S23C: Obtain the basic visual features of each rapeseed plant in each of the combined units. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ5 and standard deviation σ5 of each feature item. Use μ5±σ5 or μ5±2σ5 as the fifth hybrid determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants determined as "primary suspected hybrids" in step S22C with the fifth hybrid determination threshold. If any feature item in the basic visual features exceeds the corresponding fifth hybrid determination threshold, it is determined as a "secondary suspected hybrid". Step S24C: Obtain the basic visual features of each rapeseed plant within each whole-field area unit. Then, perform statistical operations on each feature item of the basic visual features of each rapeseed plant to calculate the mean μ6 and standard deviation σ6 of each feature item. Use μ6±σ6 or μ6±2σ6 as the sixth hybrid determination threshold for the corresponding feature item. Compare the corresponding basic visual features of the rapeseed plants determined as "secondary suspected hybrids" in step S23C with the sixth hybrid determination threshold. If any feature item in the basic visual features exceeds the corresponding sixth hybrid determination threshold, it is determined as a "hybrid".

10. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 9, characterized in that, Before obtaining the basic visual features of each rapeseed plant in step S2, the rapeseed plant position is determined. Specifically, the actual growth position of each rapeseed plant is detected by the distance sensor carried by the intelligent mobile robot, combined with the unique location number of the sowing block, and the offset between the actual growth position and the sowing position of the corresponding sowing block is calculated. If the offset is greater than a preset threshold, the rapeseed plant is directly determined to be a "hybrid plant", and there is no need to perform the subsequent basic visual feature extraction and statistical calculation.

11. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 9, characterized in that, The specific method for obtaining the basic visual features in step S2 is as follows: Plant height: The rapeseed plant is photographed by the image recognition module to obtain an elevation view and a top view. The pixel coordinates of the highest point of the rapeseed plant are located by edge detection and contour extreme point extraction. The actual plant height of the rapeseed plant is obtained by combining the intrinsic and extrinsic parameters including camera focal length, pixel size, shooting distance and shooting posture correction coefficient. Average crown width: Extract the pixel outline of the rapeseed plant crown in the top view, calculate the horizontal pixel span and vertical pixel span of the outline, take the average of the two and combine them with the pixel-actual size conversion ratio to obtain the actual crown width average. Leaf primary color RGB value: Remove background pixels from the top view and keep only fresh leaf pixels. Extract the RGB channel values ​​of the fresh leaf pixels and calculate the average value as the leaf primary color RGB value.

12. The method for efficient removal of hybrid plants during rapeseed seed production and cultivation according to claim 9, characterized in that, Before the removal module is activated in step S3 to remove rapeseed plants identified as "hybrids", the image recognition module switches the shooting angle and re-shoots the plant, then extracts the basic visual features again and compares them with the fourth, fifth, and / or sixth hybrid plant determination thresholds. If any feature item in the basic visual features still exceeds the corresponding fourth, fifth, and / or sixth hybrid plant determination threshold, it is finally identified as a "hybrid" and directly removed. If it does not exceed the fourth, fifth, and / or sixth hybrid plant determination thresholds, the removal is paused, and a warning or record is issued.

13. A highly efficient system for removing hybrid plants during rapeseed seed production and cultivation, used to implement the highly efficient method for removing hybrid plants during rapeseed seed production and cultivation as described in any one of claims 1 to 12, characterized in that, The system includes an intelligent mobile robot, and an image recognition module, a weed removal execution module, and a computational processing module mounted on and electrically connected to the intelligent mobile robot. The intelligent mobile robot moves along a path in the rapeseed field after receiving instructions. The image recognition module is used to photograph the rapeseed plants and obtain the basic visual features of each plant. The weed removal execution module is used to remove rapeseed plants identified as "hybrids." The computational processing module receives the basic visual features and then uses statistical methods to perform calculations on the basic visual features to filter out the rapeseed plants identified as "hybrids."

Images