Vision-based agricultural seed screening method and system

Abstract

The invention discloses an agricultural seed screening method and system based on vision, and relates to the technical field of agriculture, and the method comprises the steps: obtaining the three-dimensional image data of seeds through a 3D imaging technology; extracting morphology, surface features and internal features of the seeds through an image processing algorithm; screening according to the characteristics of the seeds; the multiple seeds are subjected to parallel screening through a multi-channel synchronous screening system; based on the environmental adaptability model, screening seeds in combination with specific environmental conditions; and a dynamic tracking and correction technology is adopted in the screening process to ensure the screening precision. According to the method, the three-dimensional shape, the surface characteristics and the internal structure information of the seeds are accurately obtained, so that the quality of the seeds can be comprehensively evaluated, high-quality seeds meeting the standard are screened out, and meanwhile secondary seeds not meeting the requirement are removed. According to the multi-channel synchronous screening system, the processing capacity of each screening channel is optimized through a dynamic load balancing algorithm, and the screening efficiency is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and in particular to a vision-based agricultural seed screening method and system. Background Art

[0002] As agricultural production places increasing demands on seed quality, traditional manual screening methods and mechanical screening equipment still have many problems in the seed screening process, affecting the efficiency and accuracy of seed screening.

[0003] First, the traditional manual screening method is not only time-consuming, but also greatly affected by human factors, resulting in poor accuracy and consistency of the screening results. Especially when faced with seeds with complex shapes, uneven sizes or many surface defects, the effect of manual screening is more unstable, and it is easy to miss or make the wrong selection.

[0004] Secondly, although mechanical screening methods are more efficient than manual screening, their screening accuracy is limited by the accuracy of the equipment itself, especially when dealing with seeds with irregular shapes, surface defects or complex internal structures, they often cannot provide sufficient accuracy. For example, traditional screening equipment mostly uses vibrating screens or airflow screens, relying on external characteristics such as seed volume and weight for screening, while ignoring factors such as seed morphological characteristics, surface defects and internal structure, resulting in some seeds of poor quality but normal appearance being mistakenly screened into qualified categories, affecting the quality of the final seeds.

[0005] In addition, although the existing vision-based seed screening technology can obtain and analyze seed image data through cameras, the current image processing technology mostly relies on two-dimensional image information and cannot effectively obtain the three-dimensional structure of seeds, resulting in weak recognition capabilities for seed surface defects, internal features, and irregular shapes, and cannot provide comprehensive and accurate seed quality assessment. In addition, the existing three-dimensional imaging technology still faces the problem of insufficient accuracy in the data acquisition process, especially when there are small cracks, spots and other defects on the seed surface, it is difficult for the existing technology to achieve efficient and accurate recognition.

[0006] Although the multi-channel synchronous screening system has improved the screening efficiency to a certain extent, in actual applications, the processing load of different channels is unbalanced, which can easily lead to overload of some channels, affecting the overall screening speed and system stability. At the same time, most of the current load balancing and task scheduling algorithms rely on static optimization and fail to fully consider the dynamic changes of channels in the actual working process, resulting in the overall processing efficiency of the system still being unsatisfactory in complex screening tasks.

[0007] In the dynamic screening process, existing tracking technologies are mainly based on simple motion estimation methods, which cannot accurately predict and correct the movement trajectory of seeds during the screening process in real time. Especially in the case of fast seed movement or frequent screen position adjustments, existing tracking methods may not be able to maintain high accuracy, resulting in seeds being incorrectly positioned outside the screening area, thereby reducing the accuracy of screening.

[0008] In addition, although machine learning technology has been applied to the environmental adaptability screening of seeds, most existing environmental adaptability models only consider a single factor, such as seed morphology or environmental conditions, and lack the ability to comprehensively consider the complex relationship between multiple environmental factors and seed characteristics. Therefore, the existing adaptability screening methods are still difficult to achieve accurate seed screening in the face of complex environmental changes, and cannot fully utilize the impact of environmental conditions on seed growth potential. Summary of the invention

[0009] In view of the above existing problems, the present invention is proposed.

[0010] Therefore, the present invention provides a vision-based agricultural seed screening method, which can solve the problem that existing methods mostly rely on surface features and lack effective identification of seed three-dimensional structure, internal features and tiny defects.

[0011] In order to solve the above technical problems, the present invention provides the following technical solutions, including: Using 3D imaging technology to obtain three-dimensional image data of seeds; Extract the morphology, surface features, and internal features of seeds through image processing algorithms; Screening based on seed characteristics; Screen multiple seeds in parallel through a multi-channel synchronous screening system; Based on the environmental adaptability model, seeds are screened in combination with specific environmental conditions; Dynamic tracking and correction technology is used during the screening process to ensure screening accuracy.

[0012] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, the three-dimensional image data of the seeds obtained by using 3D imaging technology includes obtaining the three-dimensional image data of the seeds by using structured light, laser scanning or multi-view stereo vision technology, and the three-dimensional image data is used to generate a three-dimensional point cloud model of the seeds; definition Point cloud obtained by structured light, For the point cloud obtained by laser scanning, an adaptive point cloud generation algorithm based on the fusion of structured light and laser scanning is constructed, which is expressed as: , in, and are the weighting coefficients of structured light and laser scanning, are the coordinates of the fused 3D points.

[0013] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, the image processing algorithm includes adjusting the reconstruction of each point by a weighted centroid method based on weighted local curvature information, which is expressed as: , in, is the new point after reconstruction, is the weighting coefficient, is the original point in the point cloud, and n is the number of points in the neighborhood; The optimized Monte Carlo method is used to calculate the three-dimensional volume of the seed. For the case of irregular seed surface, adaptive sampling and weighted sampling strategies are used, which can be expressed as: , in, is an estimate of the unit volume, is the i-th sampling point used to calculate the seed volume, is an indicator function, if Falling inside the seed, , otherwise is the number of sampling points; is the seed volume; The weighted average method is used to calculate the overall seed density using a combination of local volume and mass, expressed as: , in, is the density of each local volume cell, is the volume of the unit, is the total number of volume elements, is the density of the seeds.

[0014] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, the screening according to the characteristics of the seeds includes building a multi-dimensional feature fusion model by integrating volume and density features to further screen out seeds that do not meet the requirements, which is expressed as: , in, is the seed volume, is the mean volume of all seeds, is the standard deviation of volume; is the density of seeds, is the mean density of all seeds, is the standard deviation of density; is the weight coefficient, which is used to adjust the contribution of volume and density; Represents the comprehensive score of volume and density characteristics; if If the set threshold is exceeded, the seed is judged as unqualified; The severity of the crack is comprehensively analyzed by detecting the aspect ratio and curvature characteristics of the crack at different scales. The crack detection model is as follows: Perform multi-scale processing on the image and use different standard deviations Edge detection is performed at multiple scales, and the edge strength is expressed as: , in, Indicates The edge strength value at the layer scale, Indicates the image pixels for edge detection, represents the indicator function, Indicates that the image is The gradient value of the position, It is Standard deviation at layer scale, gradient By calculating the image brightness function The rate of change at this point is obtained. Represents the total number of points used for edge detection; According to the edge information of the crack, the length of the crack is calculated by morphological analysis and area , and combine the aspect ratio and curvature to judge the severity of the crack , expressed as: , in, is the curvature of the crack length, reflecting the degree of bending of the crack; The detection of lesions and foreign objects uses a deep learning model, which uses convolution operations to extract the texture features of the image, and then outputs the defect score of each seed image as the lesion and foreign object score through a fully connected layer, expressed as: , if If it exceeds the set threshold, the seed is judged to be defective; in, is the score for surface defects of the seed; is the kth feature extracted; is the corresponding weight coefficient; To calculate the total number of features for lesion and foreign body scores; A model combining local texture analysis and wavelet transform is used to evaluate the surface irregularity of seeds through multi-scale texture feature extraction and feature fusion, which is expressed as: , in, is the total number of wavelet scales, is the number of sampling points at this scale, is the ripple texture feature at the kth scale, is the mean value of the texture at this scale; After all features are integrated, a multi-dimensional screening criterion is formed, which can be expressed as: , in, Score the crack severity, Score lesions and foreign matter. Score surface irregularities, is the weight of each feature; Indicates the comprehensive rating of the seed.

[0015] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, the screening according to the characteristics of the seeds includes: when the volume and density of the seeds are within the preset range, and there are no obvious cracks, spots, foreign matter or irregular shapes on the surface, the seeds are judged as qualified seeds; if the seeds meet the above standards, they can directly enter the storage, marking and subsequent processing procedures; if there are small defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, the quality is further verified through secondary testing; if the verification result is qualified, it is retained as a qualified seed, otherwise it is transferred to the defective category for processing; When the volume of the seed is smaller than the preset standard range, or the density is lower than the set value, and there are no serious cracks, disease spots or foreign matter, the seed is judged as a substandard seed; if the volume is smaller than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seed can enter the reprocessing stage and restore its quality through drying or other treatment methods; if the volume and density of the seed are restored to the qualified standard after treatment, it can be re-evaluated as a qualified seed; if the seed has surface defects such as cracks, disease spots, foreign matter or irregular morphology, it should be discarded; if the volume and density of the seed do not meet the standards and there are irreparable defects on the surface, it will be directly discarded; When there are cracks, spots, or foreign matter defects on the seed surface, the seed is judged as a defective seed; if the crack depth is lower than the preset range or the spot is smaller than the preset range, but does not affect the seed growth potential, further defect analysis, i.e. crack morphology analysis or spot treatment, is used to decide whether to repair or retain the seed; if the crack exceeds the preset range or the spot is larger than the preset range, the seed cannot be restored and should be directly removed; if there is foreign matter on the seed surface and the foreign matter cannot be removed, it should be removed; for seeds with lighter defects, repair treatment will be carried out, and the qualification will be re-evaluated after repair. If it still does not meet the standards, it will be removed.

[0016] As a preferred solution of the vision-based agricultural seed screening method described in the present invention, wherein: the parallel screening of multiple seeds by a multi-channel synchronous screening system includes using multiple independent channels to screen the seeds, each channel uses a high-speed camera and an image processing module to capture and process the seeds in real time, and further combines a collaborative optimization algorithm to perform task scheduling and load balancing; Minimize the total processing time of the system on all channels and the overall load of the system. The optimization goal is expressed as: , in, Indicates the number of channels; Indicates the number of seeds to be screened; It is Channel processing The time to seed is affected by channel load and task difficulty; It is The current load of each channel, which indicates the ratio between the processing capacity of the channel and the amount of tasks; represents the optimization objective, which is the weighted sum of the total processing time and load of the system, which needs to be minimized; Processing time per channel It is determined by multiple factors, including the working status of the channel, the difficulty of seed processing, and the load of the channel, expressed as: , in, The working status of the channel , Seed Priority and task difficulty The time at which the decision is based; It is The processing capacity weight of each channel indicates the working efficiency of the channel relative to other channels; It is The current load of a channel, which indicates the amount of tasks currently assigned to the channel; Channel working status It will affect the processing time of each task, and the load It is dynamically adjusted according to the number and difficulty of tasks; In order to achieve load balancing, a dynamically adjusted task allocation strategy is adopted to adjust the task allocation in real time according to the processing capacity and load of each channel. The load balancing formula is expressed as: , in, Indicates The seeds are assigned to The ratio of channels; It is The processing capacity weight of each channel; is a joint function of channel working status, seed priority and task difficulty, indicating the Channel processing The time required to seed In order to avoid system overload, it is necessary to set constraints on the load of each channel. The load constraint condition is expressed as: , in, It is the maximum total load that the system can bear, to prevent overload operation from causing performance degradation or system crash; Load per channel It is determined by the amount of tasks it is currently processing and the complexity of the tasks, and the total load cannot exceed the maximum capacity; Set the task feedback mechanism, expressed as: , in, is the feedback adjustment coefficient, which controls the dynamic adjustment speed of task allocation; and Respectively represent The load of each channel before and after task adjustment; represents the task allocation adjustment amount; Taking into account load balancing, task allocation, time constraints, and system processing capabilities, the final optimization goal is expressed as: , Constraints: The load constraint for each channel is expressed as: , The total load constraint is expressed as: , The task allocation ratio constraint for each channel is expressed as: .

[0017] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, the environmental adaptability model includes predicting the adaptability of seeds in a specific environment through a machine learning model, setting the seed characteristics as ,in For the The feature vector of a seed contains morphological characteristics and environmental condition information; the target environment is set to ,in For the The characteristic vector of an environment; Set the environmental adaptability output to , indicating the The fitness score of each seed in a specific environment; If the random forest algorithm is used, that is, the random forest is trained through a collection of multiple decision trees, the establishment of the decision tree depends on the selection of features. Assume that There are decision trees, each tree outputs a prediction value, and the final adaptive prediction is the average of the prediction values ​​of each tree, expressed as: , in, Indicates Output of each tree, based on seed features and environmental characteristics ; is the total number of decision trees; If the support vector machine algorithm is used, that is, the support vector machine classifies or regresses the seeds by finding an optimal hyperplane, and uses SVM for regression, it can be expressed as: , in, is a weight vector representing the linear relationship between seed characteristics and environmental conditions; It is a nonlinear mapping function that maps the original features to a high-dimensional space; is the bias term, represents the transpose of the weight vector; If a neural network algorithm is used, that is, the neural network model is constructed through a multi-layer perceptron MLP, it can capture the nonlinear relationship between input features. Assuming that the neural network has layer, The output of the layer is , then the output of the neural network , expressed as: , in, is the input layer; Indicates The output of the layer, represents the activation function, Indicates The weight matrix of the layer, Indicates The bias term of the layer represents the adjustment value of the output of the neurons in this layer. Indicates The output of the layer is the activation value of the neurons in the previous layer. Represents the weight matrix of the output layer, connecting the last hidden layer and the output layer, represents the output of the output layer, i.e., the adaptability score of the seed in a specific environment. It is the bias term of the output layer, which adjusts the final prediction result; The dynamic tracking and correction technology includes estimating the movement of pixels in the image through an optical flow algorithm, inferring the movement direction and speed of the seeds based on the pixel changes between image frames, and assuming that the seeds in the image are located at positions , at time and Tracking between moments, the optical flow algorithm is expressed as: , in, and are the seeds along the image Axis and The velocity component of the axis represents the displacement speed of the seeds during the screening process; and Indicates the seed at time and The rate of displacement between moments; At each point in the image , the change of seed position is expressed by the following optical flow constraint equation: , in, It's time Midpoint of the image at time The brightness value of It's time Midpoint of the image at time The brightness value at represents the displacement of the point; The Kalman filter is used to estimate the state of the dynamic system, track the movement of the seed, and set the state variable of the seed to , which contains the current location of the seed and speed , the state update equation of the Kalman filter is: , in, Indicates The state vector at the moment, including the position and velocity of the seed; is the state transition matrix, describing the transition from time To time system dynamics; is the control input matrix, which represents the influence of external control on the state; is the control input; is the process noise, which represents the uncertainty in the system; The measurement update equation is expressed as: , in, is the measurement data, indicating the seed at time The measurement location; is the observation matrix, and the state vector Mapping to measurement space; is the measurement noise, representing the sensor error; Status Update: , in, is the predicted state estimate, which represents the current state predicted based on the estimate at the previous moment; is the Kalman gain; is the updated state estimate; The screening path is adjusted, expressed as: , in, is the position of the filter at time k, According to the movement state of the seeds Dynamically adjusted path offset, Indicates the filter position at the previous time step.

[0018] A vision-based agricultural seed screening system, comprising: An image acquisition module, used for acquiring three-dimensional image data of seeds using 3D imaging technology; An image processing module is used to extract the morphology, surface features and internal features of seeds through image processing algorithms; Feature screening module, used to screen seeds based on their features; A parallel screening module, used for screening multiple seeds in parallel through a multi-channel synchronous screening system; Environmental screening module, used to screen seeds based on environmental adaptability models combined with specific environmental conditions; The screening correction module is used to ensure the screening accuracy by using dynamic tracking and correction technology during the screening process.

[0019] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a vision-based agricultural seed screening method when executing the computer program.

[0020] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a vision-based agricultural seed screening method.

[0021] Beneficial effects of the present invention: The method of the present invention can comprehensively evaluate the quality of seeds by accurately acquiring the three-dimensional morphology, surface characteristics and internal structure information of seeds, screen out high-quality seeds that meet the standards, and eliminate inferior seeds that do not meet the requirements. The multi-channel synchronous screening system optimizes the processing capacity of each screening channel through a dynamic load balancing algorithm, significantly improving the screening efficiency. The screening method based on the environmental adaptability model conducts intelligent screening of seed adaptability in combination with specific environmental conditions, further improving the accuracy of predicting seed growth potential. Dynamic tracking and correction technology predicts the movement trajectory of seeds in real time to ensure that the seeds are always in the correct screening area during the screening process, thereby improving the overall screening accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic flow chart of a vision-based agricultural seed screening method provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0028] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0029] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a vision-based agricultural seed screening method, comprising: Step S1: using 3D imaging technology to obtain three-dimensional image data of seeds.

[0031] Step S2: Extracting the morphology, surface features and internal features of the seeds through image processing algorithms.

[0032] Step S3: Screening based on seed characteristics.

[0033] Step S4: Screen multiple seeds in parallel using a multi-channel synchronous screening system.

[0034] Step S5: Screening seeds based on the environmental adaptability model and in combination with specific environmental conditions.

[0035] Step S6: Dynamic tracking and correction technology is used during the screening process to ensure screening accuracy.

[0036] The three-dimensional image data of the seeds obtained by using 3D imaging technology includes obtaining the three-dimensional image data of the seeds by using structured light, laser scanning or multi-view stereo vision technology, and the three-dimensional image data is used to generate a three-dimensional point cloud model of the seeds; definition Point cloud obtained by structured light, For the point cloud obtained by laser scanning, an adaptive point cloud generation algorithm based on the fusion of structured light and laser scanning is constructed, which is expressed as: , in, and are the weighting coefficients of structured light and laser scanning, are the coordinates of the fused 3D points.

[0037] The image processing algorithm includes adjusting the reconstruction of each point by a weighted centroid method weighted based on local curvature information, expressed as: , in, is the new point after reconstruction, is the weighting coefficient, is the original point in the point cloud, and n is the number of points in the neighborhood; The optimized Monte Carlo method is used to calculate the three-dimensional volume of the seed. For the case of irregular seed surface, adaptive sampling and weighted sampling strategies are used, which can be expressed as: , in, is an estimate of the unit volume, is the i-th sampling point used to calculate the seed volume, is an indicator function, if Falling inside the seed, , otherwise is the number of sampling points; is the seed volume; The weighted average method is used to calculate the overall seed density using a combination of local volume and mass, expressed as: , in, is the density of each local volume cell, is the volume of the unit, is the total number of volume elements, is the density of the seeds.

[0038] The screening according to the seed characteristics includes building a multi-dimensional feature fusion model by integrating volume and density features, and further screening out seeds that do not meet the requirements, which is expressed as: , in, is the seed volume, is the mean volume of all seeds, is the standard deviation of volume; is the density of seeds, is the mean density of all seeds, is the standard deviation of density; is the weight coefficient, which is used to adjust the contribution of volume and density; Represents the comprehensive score of volume and density characteristics; if If the set threshold is exceeded, the seed is judged as unqualified; The severity of the crack is comprehensively analyzed by detecting the aspect ratio and curvature characteristics of the crack at different scales. The crack detection model is as follows: Perform multi-scale processing on the image and use different standard deviations Edge detection is performed at multiple scales, and the edge strength is expressed as: , in, Indicates The edge strength value at the layer scale, Indicates the image pixels for edge detection, represents the indicator function, Indicates that the image is The gradient value of the position, It is Standard deviation at layer scale, gradient By calculating the image brightness function The rate of change at this point is obtained. Represents the total number of points used for edge detection; According to the edge information of the crack, the length of the crack is calculated by morphological analysis and area , and combine the aspect ratio and curvature to judge the severity of the crack , expressed as: , in, is the curvature of the crack length, reflecting the degree of bending of the crack; The detection of lesions and foreign objects uses a deep learning model, which uses convolution operations to extract the texture features of the image, and then outputs the defect score of each seed image as the lesion and foreign object score through a fully connected layer, expressed as: , if If it exceeds the set threshold, the seed is judged to be defective; in, is the score for surface defects of the seed; is the kth feature extracted; is the corresponding weight coefficient; To calculate the total number of features for lesion and foreign body scores; A model combining local texture analysis and wavelet transform is used to evaluate the surface irregularity of seeds through multi-scale texture feature extraction and feature fusion, which is expressed as: , in, is the total number of wavelet scales, is the number of sampling points at this scale, is the ripple texture feature at the kth scale, is the mean value of the texture at this scale; After all features are integrated, a multi-dimensional screening criterion is formed, which can be expressed as: , in, Score the crack severity, Score lesions and foreign matter. Score surface irregularities, is the weight of each feature, Indicates the comprehensive rating of the seed.

[0039] The screening according to the characteristics of the seeds includes: when the volume and density of the seeds are within the preset range, and there are no obvious cracks, spots, foreign matter or irregular shapes on the surface, the seeds are judged as qualified seeds; if the seeds meet the above standards, they can directly enter the storage, marking and subsequent processing procedures; if there are small defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, the quality is further verified through secondary testing; if the verification results are qualified, they are retained as qualified seeds, otherwise they are transferred to the defective category for processing; When the volume of the seed is smaller than the preset standard range, or the density is lower than the set value, and there are no serious cracks, disease spots or foreign matter, the seed is judged as a substandard seed; if the volume is smaller than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seed can enter the reprocessing stage and restore its quality through drying or other treatment methods; if the volume and density of the seed are restored to the qualified standard after treatment, it can be re-evaluated as a qualified seed; if the seed has surface defects such as cracks, disease spots, foreign matter or irregular morphology, it should be discarded; if the volume and density of the seed do not meet the standards and there are irreparable defects on the surface, it will be directly discarded; When there are cracks, spots, or foreign matter defects on the seed surface, the seed is judged as a defective seed; if the crack depth is lower than the preset range or the spot is smaller than the preset range, but does not affect the seed growth potential, further defect analysis, i.e. crack morphology analysis or spot treatment, is used to decide whether to repair or retain the seed; if the crack exceeds the preset range or the spot is larger than the preset range, the seed cannot be restored and should be directly removed; if there is foreign matter on the seed surface and the foreign matter cannot be removed, it should be removed; for seeds with lighter defects, repair treatment will be carried out, and the qualification will be re-evaluated after repair. If it still does not meet the standards, it will be removed.

[0040] The parallel screening of multiple seeds by the multi-channel synchronous screening system includes using multiple independent channels to screen the seeds, each channel uses a high-speed camera and an image processing module to capture and process the seeds in real time, and further combines the collaborative optimization algorithm to perform task scheduling and load balancing; Minimize the total processing time of the system on all channels and the overall load of the system. The optimization goal is expressed as: , in, Indicates the number of channels; Indicates the number of seeds to be screened; It is Channel processing The time to seed is affected by channel load and task difficulty; It is The current load of each channel, which indicates the ratio between the processing capacity of the channel and the amount of tasks; represents the optimization objective, which is the weighted sum of the total processing time and load of the system, which needs to be minimized; Processing time per channel It is determined by multiple factors, including the working status of the channel, the difficulty of seed processing, and the load of the channel, expressed as: , in, The working status of the channel , Seed Priority and task difficulty The time at which the decision is based; It is The processing capacity weight of each channel indicates the working efficiency of the channel relative to other channels; It is The current load of a channel, which indicates the amount of tasks currently assigned to the channel; Channel working status It will affect the processing time of each task, and the load It is dynamically adjusted according to the number and difficulty of tasks; In order to achieve load balancing, a dynamically adjusted task allocation strategy is adopted to adjust the task allocation in real time according to the processing capacity and load of each channel. The load balancing formula is expressed as: , in, Indicates The seeds are assigned to The ratio of channels; It is The processing capacity weight of each channel; is a joint function of channel working status, seed priority and task difficulty, indicating the Channel processing The time required to seed In order to avoid system overload, it is necessary to set constraints on the load of each channel. The load constraint condition is expressed as: , in, It is the maximum total load that the system can bear, to prevent overload operation from causing performance degradation or system crash; Load per channel It is determined by the amount of tasks it is currently processing and the complexity of the tasks, and the total load cannot exceed the maximum capacity; Set the task feedback mechanism, expressed as: , in, is the feedback adjustment coefficient, which controls the dynamic adjustment speed of task allocation; and Respectively represent The load of each channel before and after task adjustment; represents the task allocation adjustment amount; Taking into account load balancing, task allocation, time constraints, and system processing capabilities, the final optimization goal is expressed as: , Constraints: The load constraint for each channel is expressed as: , The total load constraint is expressed as: , The task allocation ratio constraint for each channel is expressed as: .

[0041] The environmental adaptability model includes predicting the adaptability of seeds in a specific environment through a machine learning model, setting the seed characteristics as ,in For the The feature vector of a seed contains morphological characteristics and environmental condition information; the target environment is set to ,in For the The characteristic vector of an environment; Set the environmental adaptability output to , indicating the The adaptability score of each seed in a specific environment; If the random forest algorithm is used, that is, the random forest is trained through a collection of multiple decision trees, the establishment of the decision tree depends on the selection of features. Assume that There are decision trees, each tree outputs a prediction value, and the final adaptive prediction is the average of the prediction values ​​of each tree, expressed as: , in, Indicates Output of each tree, based on seed features and environmental characteristics ; is the total number of decision trees; If the support vector machine algorithm is used, that is, the support vector machine classifies or regresses the seeds by finding an optimal hyperplane, and uses SVM for regression, it can be expressed as: , in, is a weight vector representing the linear relationship between seed characteristics and environmental conditions; It is a nonlinear mapping function that maps the original features to a high-dimensional space; is the bias term, represents the transpose of the weight vector; If a neural network algorithm is used, that is, the neural network model is constructed through a multi-layer perceptron MLP, it can capture the nonlinear relationship between input features. Assuming that the neural network has layer, The output of the layer is , then the output of the neural network , expressed as: , in, is the input layer; Indicates The output of the layer, represents the activation function, Indicates The weight matrix of the layer, Indicates The bias term of the layer represents the adjustment value of the output of the neurons in this layer. Indicates The output of the layer is the activation value of the neurons in the previous layer. Represents the weight matrix of the output layer, connecting the last hidden layer and the output layer, represents the output of the output layer, i.e., the adaptability score of the seed in a specific environment. It is the bias term of the output layer, which adjusts the final prediction result; The dynamic tracking and correction technology includes estimating the movement of pixels in the image through an optical flow algorithm, inferring the movement direction and speed of the seeds based on the pixel changes between image frames, and assuming that the seeds in the image are located at positions , at time and Tracking between moments, the optical flow algorithm is expressed as: , in, and are the seeds along the image Axis and The velocity component of the axis represents the displacement speed of the seeds during the screening process; and Indicates the seed at time and The rate of displacement between moments; At each point in the image , the change of seed position is expressed by the following optical flow constraint equation: , in, It's time Midpoint of the image at time The brightness value of It's time Midpoint of the image at time The brightness value at represents the displacement of the point; The Kalman filter is used to estimate the state of the dynamic system, track the movement of the seed, and set the state variable of the seed to , which contains the current location of the seed and speed , the state update equation of the Kalman filter is: , in, Indicates The state vector at the moment, including the position and velocity of the seed; is the state transition matrix, describing the transition from time To time system dynamics; is the control input matrix, which represents the influence of external control on the state; is the control input; is the process noise, which represents the uncertainty in the system; The measurement update equation is expressed as: , in, is the measurement data, indicating the seed at time The measurement location; is the observation matrix, and the state vector Mapping to measurement space; is the measurement noise, representing the sensor error; Status Update: , in, is the predicted state estimate, which represents the current state predicted based on the estimate at the previous moment; is the Kalman gain; is the updated state estimate, is the difference between the measured and predicted values; After the seed motion state is estimated in real time, the filter needs to dynamically adjust its path to ensure that the seed is always in the correct screening area. Assume that the current position of the filter is , and at each moment, according to the state estimate of the Kalman filter , the filter's path adjustment amount is .

[0042] The screening path is adjusted, expressed as: , in, is the position of the filter at time k, According to the movement state of the seeds Dynamically adjusted path offset, Indicates the filter position at the previous time step.

[0043] It should be noted that the optical flow method first estimates the velocity of the seed , providing initial information for subsequent dynamic estimation. The Kalman filter then uses this velocity information, combined with measurement noise and prediction values, to correct the motion state of the seed through state update and measurement update steps.

[0044] The Kalman filter provides the precise dynamic state of the seed (including position and velocity). The filter path is adjusted based on these dynamic estimates by calculating the offset , ensuring that the screener can adapt to the movement of seeds in real time and ensure screening accuracy.

[0045] The optical flow algorithm provides a preliminary estimate of the seed velocity, and the Kalman filter continuously refines the dynamic estimate of the seed through state estimation and measurement updates. The adjustment of the screening path depends on the real-time state update of the Kalman filter to ensure that the filter tracks and keeps the seed in the correct screening area.

[0046] Example 2 is an embodiment of the present invention, which provides a vision-based agricultural seed screening method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0047] The experiment selected 10,000 seeds, including normal seeds (70%), inferior seeds (20%) and seeds with foreign species (10%). Each type of seeds was randomly mixed to simulate the seed screening situation in actual agricultural production. The experimental environment was standardized light conditions with no obvious external interference. Two different seed screening methods were used for comparative testing during the experiment.

[0048] In the experiment of the traditional method, all seeds are first processed by a combination of manual and mechanized screening. The manual part is mainly responsible for classifying seeds through visual and manual screening. Although manual screening can quickly determine more obvious defects based on experience, its accuracy is greatly affected by the operator's ability and fatigue. Mechanized screening uses an automated vibrating screen system, which performs preliminary screening based on seed weight, and the camera cooperates with the automatic image processing algorithm to make a rough judgment on the morphology of the seeds. The vibrating screen screens seeds by adjusting the grid aperture and vibration intensity. The accuracy of the mechanical screening equipment depends on the physical parameter settings, and it is difficult to accurately identify problems such as subtle morphology and irregular cracks. Therefore, the accuracy of the screening results is low, especially when dealing with small seed defects (such as microcracks, lesions, and foreign matter), the misscreening phenomenon is serious. In addition, due to the coordination of equipment and manual operation in the entire screening process, the screening efficiency is low.

[0049] The method of the present invention performs seed screening through a multi-channel synchronous screening system based on 3D imaging. First, the three-dimensional image data of the seeds is obtained through a high-speed camera, and the morphology, surface features and internal features of the seeds are extracted in combination with an optimized image processing algorithm. By using structured light, laser scanning and multi-view stereo vision technology, a three-dimensional point cloud model of the seeds is generated in real time, and the weighted centroid method is used to accurately adjust and correct the seed surface. During the screening process, the optical flow algorithm and the Kalman filter are combined to track the seeds in dynamic motion in real time, predict their motion trajectory, and dynamically adjust the screening path to ensure that the seeds are always in the correct position and improve the accuracy of screening. The experimental results are shown in Table 1.

[0050] Table 1 Comparison of experimental results method Screening accuracy (%) Screening efficiency (particles / minute) Dynamic tracking deviation(mm) Load balancing deviation (%) Adaptive screening accuracy (%) Traditional methods 84.5 160 not applicable not applicable 67.2 Method of the present invention 97.8 470 0.6 4.1 91.3 As shown in Table 1, the present invention significantly improves the deficiencies of traditional methods in screening accuracy, screening efficiency, dynamic tracking, load balancing and environmental adaptability screening. Traditional screening methods rely on simple mechanical principles or manual observation, and are unable to identify defects such as tiny cracks and lesions, resulting in insufficient screening accuracy. The present invention uses 3D imaging technology to accurately capture the surface and internal features of seeds, and combines the weighted centroid method and optimized image processing algorithm to effectively improve the ability to recognize complex seed morphology, thereby greatly improving screening accuracy.

[0051] In terms of screening efficiency, traditional single-channel equipment is limited by mechanical structure and manual participation, and it is difficult to meet the requirements of efficient processing. The present invention introduces a multi-channel synchronous screening system, dynamically allocates tasks through a collaborative optimization algorithm, realizes parallel processing, and significantly improves the seed screening speed. In addition, the combination of optical flow algorithm and Kalman filter enables dynamic tracking technology to predict the movement trajectory of seeds in real time and accurately adjust the screening path to ensure accurate positioning of seeds during the screening process, avoiding the problem of misscreening caused by path deviation.

[0052] In terms of load balancing, the dynamic task allocation technology of the present invention ensures that each channel of the multi-channel screening system is evenly loaded, significantly improving equipment utilization and screening stability. The screening method combined with the environmental adaptability model uses machine learning technology to comprehensively analyze the matching degree between seed characteristics and environmental conditions, so that the screened seeds are more in line with the needs of the target environment. In summary, the present invention effectively solves the defects in traditional methods through a number of innovative technical solutions, and comprehensively improves the efficiency and accuracy of the screening process.

[0053] Embodiment 3 is an embodiment of the present invention, which provides a vision-based agricultural seed screening method and system, including: An image acquisition module, used for acquiring three-dimensional image data of seeds using 3D imaging technology; An image processing module is used to extract the morphology, surface features and internal features of seeds through image processing algorithms; Feature screening module, used to screen seeds based on their features; A parallel screening module, used for screening multiple seeds in parallel through a multi-channel synchronous screening system; Environmental screening module, used to screen seeds based on environmental adaptability models combined with specific environmental conditions; The screening correction module is used to ensure the screening accuracy by using dynamic tracking and correction technology during the screening process.

[0054] Embodiment 4: The fourth embodiment of the present invention is different from the first three embodiments in that: If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0055] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0056] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0057] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0058] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0059] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A vision-based agricultural seed screening method, characterized in that: include, Using 3D imaging technology to obtain three-dimensional image data of seeds; Extract the morphology, surface features, and internal features of seeds through image processing algorithms; Screening based on seed characteristics; Screen multiple seeds in parallel through a multi-channel synchronous screening system; Based on the environmental adaptability model, seeds are screened in combination with specific environmental conditions; Dynamic tracking and correction technology is used during the screening process to ensure screening accuracy.

2. A vision-based agricultural seed screening method as claimed in claim 1, characterized in that: The three-dimensional image data of the seeds obtained by using 3D imaging technology includes obtaining the three-dimensional image data of the seeds by using structured light, laser scanning or multi-view stereo vision technology, and the three-dimensional image data is used to generate a three-dimensional point cloud model of the seeds; definition Point cloud obtained by structured light, For the point cloud obtained by laser scanning, an adaptive point cloud generation algorithm based on the fusion of structured light and laser scanning is constructed, which is expressed as: ， in, and are the weighting coefficients of structured light and laser scanning, are the coordinates of the fused 3D points.

3. A vision-based agricultural seed screening method as claimed in claim 2, characterized in that: The image processing algorithm includes adjusting the reconstruction of each point by a weighted centroid method weighted based on local curvature information, expressed as: ， in, is the new point after reconstruction, is the weighting coefficient, is the original point in the point cloud, and n is the number of points in the neighborhood; The optimized Monte Carlo method is used to calculate the three-dimensional volume of the seed. For the case of irregular seed surface, adaptive sampling and weighted sampling strategies are used, which can be expressed as: ， in, is an estimate of the unit volume, is the i-th sampling point used to calculate the seed volume, is an indicator function, if Falling inside the seed, , otherwise is the number of sampling points; is the seed volume; The weighted average method is used to calculate the overall seed density using a combination of local volume and mass, expressed as: ， in, is the density of each local volume cell, is the volume of the unit, is the total number of volume elements, is the density of the seeds.

4. A vision-based agricultural seed screening method as claimed in claim 3, characterized in that: The screening according to the characteristics of the seeds comprises: By integrating volume and density features, a multi-dimensional feature fusion model is constructed to further filter out seeds that do not meet the requirements, which is expressed as: ， in, is the seed volume, is the mean volume of all seeds, is the standard deviation of volume; is the density of seeds, is the mean density of all seeds, is the standard deviation of density; is the weight coefficient, which is used to adjust the contribution of volume and density; if If the set threshold is exceeded, the seed is judged as unqualified; It represents the comprehensive score of volume and density characteristics; The severity of the crack is comprehensively analyzed by detecting the aspect ratio and curvature characteristics of the crack at different scales. The crack detection model is as follows: Perform multi-scale processing on the image and use different standard deviations Edge detection is performed at multiple scales, and the edge strength is expressed as: ， in, Indicates The edge strength value at the layer scale, Indicates the image pixels for edge detection, represents the indicator function, Indicates that the image is The gradient value of the position, It is Standard deviation at layer scale, gradient By calculating the image brightness function The rate of change at this point is obtained. Represents the total number of points used for edge detection; According to the edge information of the crack, the length of the crack is calculated by morphological analysis and area , and combine the aspect ratio and curvature to judge the severity of the crack , expressed as: ， in, is the curvature of the crack length, reflecting the degree of bending of the crack; The detection of lesions and foreign objects uses a deep learning model, which uses convolution operations to extract the texture features of the image, and then outputs the defect score of each seed image as the lesion and foreign object score through a fully connected layer, expressed as: ， if If it exceeds the set threshold, the seed is judged to be defective; in, is a score for surface defects in seeds; is the kth feature extracted; is the corresponding weight coefficient; To calculate the total number of features for lesion and foreign body scores; A model combining local texture analysis and wavelet transform is used to evaluate the surface irregularity of seeds through multi-scale texture feature extraction and feature fusion, which is expressed as: ， in, is the total number of wavelet scales, is the number of sampling points at this scale, is the ripple texture feature at the kth scale, is the mean value of the texture at this scale; After all features are integrated, a multi-dimensional screening criterion is formed, which can be expressed as: ， in, Score the severity of the crack, Score lesions and foreign matter. Score surface irregularities, is the weight of each feature; Indicates the comprehensive rating of the seed.

5. A vision-based agricultural seed screening method as claimed in claim 4, characterized in that: The screening according to the characteristics of the seeds comprises: When the volume and density of the seeds are within the preset range, and there are no obvious cracks, spots, foreign matter or irregular shapes on the surface, the seeds are judged as qualified seeds; if the seeds meet the above standards, they can directly enter the storage, marking and subsequent processing procedures; if there are small defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, they will continue to be verified through secondary testing; if the verification results are qualified, they will be retained as qualified seeds, otherwise they will be transferred to the defective category for processing; When the volume of the seed is smaller than the preset standard range, or the density is lower than the set value, and there are no serious cracks, disease spots or foreign matter, the seed is judged as a substandard seed; if the volume is smaller than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seed can enter the reprocessing stage and restore its quality through drying or other treatment methods; if the volume and density of the seed are restored to the qualified standard after treatment, it can be re-evaluated as a qualified seed; if the seed has surface defects such as cracks, disease spots, foreign matter or irregular morphology, it should be discarded; if the volume and density of the seed do not meet the standards and there are irreparable defects on the surface, it will be directly discarded; When there are cracks, spots, or foreign matter defects on the seed surface, the seed is judged as a defective seed; if the crack depth is lower than the preset range or the spot is smaller than the preset range, but does not affect the seed growth potential, further defect analysis, i.e. crack morphology analysis or spot treatment, is used to decide whether to repair or retain the seed; if the crack exceeds the preset range or the spot is larger than the preset range, the seed cannot be restored and should be directly removed; if there is foreign matter on the seed surface and the foreign matter cannot be removed, it should be removed; for seeds with lighter defects, repair treatment will be carried out, and the qualification will be re-evaluated after repair. If it still does not meet the standards, it will be removed.

6. A vision-based agricultural seed screening method as claimed in claim 5, characterized in that: The parallel screening of multiple seeds by the multi-channel synchronous screening system comprises: Use multiple independent channels to screen seeds. Each channel uses a high-speed camera and image processing module to capture and process seeds in real time, and further combines collaborative optimization algorithms for task scheduling and load balancing. Minimize the total processing time of the system on all channels and the overall load of the system. The optimization goal is expressed as: ， in, Indicates the number of channels; Indicates the number of seeds to be screened; It is Channel processing The time to seed is affected by channel load and task difficulty; It is The current load of each channel, which indicates the ratio between the processing capacity of the channel and the amount of tasks; represents the optimization objective, which is the weighted sum of the total processing time and load of the system, which needs to be minimized; Processing time per channel It is determined by multiple factors, including the working status of the channel, the difficulty of seed processing, and the load of the channel, expressed as: ， in, The working status of the channel , Seed Priority and task difficulty The time at which the decision is based; It is The processing capacity weight of each channel indicates the working efficiency of the channel relative to other channels; It is The current load of a channel, which indicates the amount of tasks currently assigned to the channel; Channel working status It will affect the processing time of each task, and the load It is dynamically adjusted according to the number and difficulty of tasks; In order to achieve load balancing, a dynamically adjusted task allocation strategy is adopted to adjust the task allocation in real time according to the processing capacity and load of each channel. The load balancing formula is expressed as: ， in, Indicates The seeds are assigned to The ratio of channels; It is The processing capacity weight of each channel; is a joint function of channel working status, seed priority and task difficulty, indicating the Channel processing The time required to seed In order to avoid system overload, it is necessary to set constraints on the load of each channel. The load constraint condition is expressed as: ， in, It is the maximum total load that the system can bear, to prevent overload operation from causing performance degradation or system crash; Load per channel It is determined by the amount of tasks it is currently processing and the complexity of the tasks, and the total load cannot exceed the maximum capacity; Set the task feedback mechanism, expressed as: ， in, is the feedback adjustment coefficient, which controls the dynamic adjustment speed of task allocation; and Respectively represent The load of each channel before and after task adjustment; represents the task allocation adjustment amount; Taking into account load balancing, task allocation, time constraints, and system processing capabilities, the final optimization goal is expressed as: ， Constraints: The load constraint for each channel is expressed as: ， The total load constraint is expressed as: ， The task allocation ratio constraint for each channel is expressed as: 。 7. A vision-based agricultural seed screening method as claimed in claim 6, characterized in that: The environmental adaptability model includes: The machine learning model is used to predict the adaptability of seeds in a specific environment, and the seed characteristics are set as ,in For the The feature vector of a seed contains morphological characteristics and environmental condition information; the target environment is set to ,in For the The characteristic vector of an environment; Set the environmental adaptability output to , indicating the The fitness score of each seed in a specific environment; If the random forest algorithm is used, that is, the random forest is trained through a collection of multiple decision trees, the establishment of the decision tree depends on the selection of features. Assume that There are decision trees, each tree outputs a prediction value, and the final adaptive prediction is the average of the prediction values ​​of each tree, expressed as: ， in, Indicates Output of each tree, based on seed features and environmental characteristics ; is the total number of decision trees; If the support vector machine algorithm is used, that is, the support vector machine classifies or regresses the seeds by finding an optimal hyperplane, and uses SVM for regression, it can be expressed as: ， in, is a weight vector representing the linear relationship between seed characteristics and environmental conditions; It is a nonlinear mapping function that maps the original features to a high-dimensional space; is the bias term, represents the transpose of the weight vector; If a neural network algorithm is used, that is, the neural network model is constructed through a multi-layer perceptron MLP, it can capture the nonlinear relationship between input features. Assuming that the neural network has layer, The output of the layer is , then the output of the neural network , expressed as: ， in, is the input layer; Indicates The output of the layer, represents the activation function, Indicates The weight matrix of the layer, Indicates The bias term of the layer represents the adjustment value of the output of the neurons in this layer. Indicates The output of the layer is the activation value of the neurons in the previous layer. Represents the weight matrix of the output layer, connecting the last hidden layer and the output layer, represents the output of the output layer, i.e., the adaptability score of the seed in a specific environment. It is the bias term of the output layer, which adjusts the final prediction result; The dynamic tracking and correction technology includes estimating the movement of pixels in the image through an optical flow algorithm, inferring the movement direction and speed of the seeds based on the pixel changes between image frames, and assuming that the seeds in the image are located at positions , at time and Tracking between moments, the optical flow algorithm is expressed as: ， in, and are the seeds along the image Axis and The velocity component of the axis represents the displacement speed of the seeds during the screening process; and Indicates the seed at time and The rate of displacement between moments; At each point in the image , the change of seed position is expressed by the following optical flow constraint equation: ， in, It's time Midpoint of the image at time The brightness value of It's time Midpoint of the image at time The brightness value at represents the displacement of the point; The Kalman filter is used to estimate the state of the dynamic system, track the movement of the seed, and set the state variable of the seed to , which contains the current location of the seed and speed , the state update equation of the Kalman filter is: ， in, Indicates The state vector at the moment, including the position and velocity of the seed; is the state transition matrix, describing the transition from time To time system dynamics; is the control input matrix, which represents the influence of external control on the state; is the control input; is the process noise, which represents the uncertainty in the system; The measurement update equation is expressed as: ， in, is the measurement data, indicating the seed at time The measurement location; is the observation matrix, and the state vector Mapping to measurement space; is the measurement noise, representing the sensor error; Status Update: ， in, is the predicted state estimate, which represents the current state predicted based on the estimate at the previous moment; is the Kalman gain; is the updated state estimate; The screening path is adjusted, expressed as: ， in, is the position of the filter at time k, According to the movement state of the seeds Dynamically adjusted path offset, Indicates the filter position at the previous time step.

8. A system using a vision-based agricultural seed screening method as claimed in any one of claims 1 to 7, characterized in that: include, An image acquisition module, used for acquiring three-dimensional image data of seeds using 3D imaging technology; An image processing module is used to extract the morphology, surface features and internal features of seeds through image processing algorithms; Feature screening module, used to screen seeds based on their features; A parallel screening module, used for screening multiple seeds in parallel through a multi-channel synchronous screening system; Environmental screening module, used to screen seeds based on environmental adaptability models combined with specific environmental conditions; The screening correction module is used to ensure the screening accuracy by using dynamic tracking and correction technology during the screening process.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images