A vision-based agricultural seed screening method and system

Abstract

The present invention discloses a vision-based agricultural seed screening method and system, which relates to the field of agricultural technology. It includes obtaining three-dimensional image data of seeds using 3D imaging technology; extracting the morphological, surface features, and internal features of seeds through image processing algorithms; screening seeds according to their features; performing parallel screening of multiple seeds through a multi-channel synchronous screening system; screening seeds based on an environmental adaptability model in combination with specific environmental conditions; and adopting dynamic tracking and correction technology during the screening process to ensure the screening accuracy. Through accurately obtaining the three-dimensional morphology, surface features, and internal structure information of seeds, the present invention can comprehensively evaluate the quality of seeds, screen out high-quality seeds that meet the standards, and at the same time eliminate substandard seeds that do not meet the requirements. The multi-channel synchronous screening system optimizes the processing capabilities of each screening channel through a dynamic load balancing algorithm, significantly improving the screening efficiency.

Description

Technical Field

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

[0002] With the continuous improvement of the requirements for seed quality in agricultural production, there are still many problems in the traditional manual screening method and mechanical screening equipment during the seed screening process, which affect the screening efficiency and accuracy of seeds.

[0003] Firstly, the traditional manual screening method not only takes a long time, but is also greatly affected by human factors, resulting in poor accuracy and consistency of the screening results. Especially when facing 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 misselect.

[0004] Secondly, although the mechanical screening method is more efficient than manual screening, its 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, it often cannot provide sufficient accuracy. For example, traditional screening equipment mostly uses vibrating screens or air-flow screens, which rely on external characteristics such as seed volume and weight for screening, while ignoring factors such as seed morphological characteristics, surface defects and internal structures, resulting in some seeds with poor quality but normal appearance being mis-screened into the qualified category, affecting the quality of the final seeds.

[0005] In addition, although the existing vision-based seed screening technologies can obtain image data of seeds through cameras and perform analysis, the current image processing technologies mostly rely on two-dimensional image information and cannot effectively obtain the three-dimensional structure of seeds, resulting in weak recognition ability for seed surface defects, internal characteristics and irregular shapes, and unable to provide a comprehensive and accurate seed quality assessment. In addition, the existing three-dimensional imaging technologies still face the problem of insufficient accuracy during data acquisition. Especially when there are small cracks, disease spots and other defects on the seed surface, it is very difficult for the existing technologies to achieve efficient and accurate recognition.

[0006] Although the multi-channel synchronous screening system has improved the screening efficiency to a certain extent, in practical applications, the processing loads of different channels are unbalanced, which easily leads 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 do not fully consider the dynamic changes of channels during the actual working process, resulting in an unsatisfactory overall processing efficiency of the system in complex screening tasks.

[0007] During the dynamic screening process, existing tracking technologies mainly rely on simple motion estimation methods and are unable to predict and correct the motion trajectory of seeds in real time and accurately. Especially when the seeds move fast or the position of the sieve is adjusted frequently, existing tracking methods may not be able to maintain high accuracy, resulting in the seeds being mispositioned outside the screening area, thus reducing the screening accuracy.

[0008] In addition, although machine learning technologies have 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 relationships between multiple environmental factors and seed characteristics. Therefore, existing adaptability screening methods are still difficult to achieve precise seed screening in the face of complex environmental changes and cannot fully utilize the influence of environmental conditions on the growth potential of seeds. 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 the effective recognition of the three-dimensional structure, internal features and micro defects of seeds.

[0011] To solve the above technical problems, the present invention provides the following technical solutions, including:

[0012] Obtain three-dimensional image data of seeds by using 3D imaging technology;

[0013] Extract the morphology, surface features and internal features of seeds through image processing algorithms;

[0014] Screen according to the characteristics of seeds;

[0015] Parallelly screen multiple seeds through a multi-channel synchronous screening system;

[0016] Based on the environmental adaptability model, screen seeds in combination with specific environmental conditions;

[0017] Adopt dynamic tracking and correction technology during the screening process to ensure the screening accuracy.

[0018] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, wherein: the obtaining of three-dimensional image data of seeds by using 3D imaging technology includes obtaining three-dimensional image data of 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;

[0019] Define as the 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:

[0020] ,

[0021] where, and are the weighting coefficients of structured light and laser scanning respectively, is the three-dimensional point coordinate after fusion.

[0022] As a preferred solution of the vision-based agricultural seed screening method described in the present invention, wherein: the image processing algorithm includes adjusting the reconstruction of each point by the weighted centroid method based on local curvature information, which is expressed as:

[0023] ,

[0024] where, 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;

[0025] The three-dimensional volume of the seed is calculated by the optimized Monte Carlo method. For the case where the surface of the seed is irregular, an adaptive sampling and weighted sampling strategy is used, which is expressed as:

[0026] ,

[0027] where, is the estimated value of the unit volume, is the i-th sampling point used to calculate the volume of the seed, is the indicator function. If falls inside the seed, then , otherwise it is is the number of sampling points; is the volume of the seed;

[0028] The overall density of the seed is calculated by using the combination of local volume and mass and the weighted average method, which is expressed as:

[0029] ,

[0030] where, is the density of each local volume unit, is the volume of this unit, is the total number of volume units, is the density of the seed.

[0031] As a preferred solution of the vision-based agricultural seed screening method described in the present invention, wherein: the screening according to the characteristics of the seeds includes constructing a multi-dimensional feature fusion model by comprehensively considering the volume and density characteristics, and further screening out the seeds that do not meet the requirements, expressed as:

[0032] ,

[0033] wherein, is the volume of the seed, is the average volume of all seeds, is the standard deviation of the volume; is the density of the seed, is the average density of all seeds, is the standard deviation of the density; is the weight coefficient, used to adjust the contribution degrees of the volume and density; represents the comprehensive score of the volume and density characteristics; if exceeds the set threshold, the seed is determined to be unqualified;

[0034] By detecting the aspect ratio and curvature characteristics of the crack at different scales and comprehensively analyzing the severity of the crack, the crack detection model is as follows:

[0035] Perform multi-scale processing on the image, and use different standard deviations to perform edge detection at multiple scales, and the edge intensity is expressed as:

[0036] ,

[0037] wherein, represents the edge intensity value at the th layer scale, represents the th pixel point for edge detection in the image, represents the indicator function, represents the gradient value of the image at the position, is the standard deviation at the th layer scale, and the gradient is obtained by calculating the change rate of the image brightness function at this point, represents the total number of points for edge detection;

[0038] According to the edge information of the crack, calculate the length and area of the crack through morphological analysis, and combine the aspect ratio and curvature to judge the severity of the crack, expressed as:

[0039] ,

[0040] Among them, is the curvature of the crack length, reflecting the degree of crack bending;

[0041] The detection of lesions and foreign objects uses a deep learning model. Texture features of the image are extracted using convolutional operations, and then the defect score of each seed image is output through a fully connected layer, representing the lesion and foreign object score, expressed as:

[0042] ,

[0043] If exceeds the set threshold, then the seed is determined to have a defect;

[0044] Among them, is the score of the seed surface defect; is the k-th extracted feature; is the corresponding weight coefficient; is the total number of features for calculating the lesion and foreign object score;

[0045] A model combining local texture analysis and wavelet transform is used to evaluate the surface irregularity of the seed through multi-scale texture feature extraction and feature fusion, expressed as:

[0046] ,

[0047] Among them, is the total number of wavelet scales, is the number of sampling points at this scale, is the ripple texture feature at the k-th scale, is the mean value of the texture at this scale;

[0048] After all features are fused, a multi-dimensional screening criterion is formed, expressed as:

[0049] ,

[0050] Among them, is the crack severity score, is the lesion and foreign object score, is the surface irregularity score, is the weight of each feature; represents the comprehensive score of the seed.

[0051] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, wherein: the screening based on the characteristics of the seeds includes that when the volume and density of the seeds are both within the preset range, and there are no obvious cracks, lesions, foreign objects or morphological irregularities on the surface, the seeds are determined to be qualified seeds; if the seeds meet the above criteria, they can directly enter the storage, marking and subsequent processing processes; if there are minor defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, secondary detection is continued to verify their quality; if the verification result is qualified, they are retained as qualified seeds, otherwise they are transferred to the defective category for processing;

[0052] When the volume of the seeds is less than the predetermined standard range, or the density is lower than the set value, and there are no serious cracks, lesions or foreign objects, the seeds are determined to be substandard seeds; if the volume is less than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seeds can enter the reprocessing stage, and their quality can be restored by drying or other processing methods; if the volume and density of the seeds are restored to the qualified standard after processing, they can be re-evaluated as qualified seeds; if the seeds have surface defects such as cracks, lesions, foreign objects or morphological irregularities, they should be removed; if the volume and density of the seeds do not meet the standard and there are irreparable defects on the surface, they are directly removed;

[0053] When there are cracks, lesions or foreign object defects on the surface of the seeds, the seeds are determined to be defective seeds; if the crack depth is lower than the preset range or the lesion is smaller than the preset range, but it does not affect the growth potential of the seeds, through further defect analysis, that is, crack morphology analysis or lesion treatment, it is decided whether to repair or retain; if the crack exceeds the preset range or the lesion is larger than the preset range interval and the seeds cannot be restored, they should be directly removed; if there are foreign objects on the surface of the seeds and the foreign objects cannot be removed, they should be removed; for seeds with minor defects, repair treatment is carried out, and after repair, it is re-evaluated whether they are qualified. If they still do not meet the standard, they are removed.

[0054] As a preferred embodiment of the vision-based agricultural seed screening method of the present invention, wherein: the parallel screening of multiple seeds by the multi-channel synchronous screening system includes using multiple independent channels to screen the seeds. Each channel captures and processes the seeds in real time through a high-speed camera and an image processing module, and further combines a cooperative optimization algorithm for task scheduling and load balancing;

[0055] Minimize the total processing time of the system on all channels and the overall load of the system. The optimization objective is expressed as:

[0056] ,

[0057] wherein, represents the number of channels; represents the number of seeds to be screened; is the time for the th channel to process the th seed, which is affected by the channel load and task difficulty; is the current load of the th channel, representing the ratio between the processing capacity and the task volume of the channel;

[0058] The processing time of each channel is jointly determined by multiple factors, including the working state of the channel, the processing difficulty of the seed, and the load condition of the channel, expressed as:

[0059] ,

[0060] where is the base time determined by the working state of the channel, the seed priority and the task difficulty; is the processing capacity weight of the th channel, representing the working efficiency of this channel relative to other channels; is the current load of the th channel, representing the amount of tasks currently assigned to this channel;

[0061] The working state of the channel will affect the processing time of each task, while the load is dynamically adjusted according to the number and difficulty of tasks;

[0062] To achieve load balancing, a dynamically adjusted task allocation strategy is adopted. According to the processing capacity and load conditions of each channel, the task allocation is adjusted in real time. The load balancing formula is expressed as:

[0063] ,

[0064] where represents the ratio of the th seed allocated to the th channel; is the processing capacity weight of the th channel; is the joint function of the channel working state, seed priority and task difficulty, representing the time required for the th channel to process the th seed;

[0065] To avoid system overload, constraints need to be set on the load of each channel. The load constraint condition is expressed as:

[0066] ,

[0067] Among them, is the maximum total load that the system can bear, preventing performance degradation or system crashes caused by overloading operations;

[0068] The load of each channel 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 bearing capacity;

[0069] Set up a task feedback mechanism, expressed as:

[0070] ,

[0071] Among them, is the feedback adjustment coefficient, controlling the dynamic adjustment speed of task allocation; and respectively represent the loads of the th channel before and after task adjustment; represents the task allocation adjustment amount;

[0072] Taking into account load balancing, task allocation, time constraints, and the processing capacity of the system, the final optimization goal is expressed as:

[0073] ,

[0074] Constraint conditions:

[0075] The load constraint of each channel is expressed as:

[0076] ,

[0077] The total load constraint is expressed as:

[0078] ,

[0079] The task allocation ratio constraint of each channel is expressed as:

[0080] .

[0081] As a preferred solution of the vision-based agricultural seed screening method described in the present invention, wherein: the environmental adaptability model includes predicting the adaptability of seeds in a specific environment through a machine learning model, setting the seed characteristics as , among which is the feature vector of the th seed, including morphological characteristics and environmental condition information; setting the target environment as , among which is the The feature vector of an environment;

[0082] Set the environmental adaptability output as , indicating the adaptability score of the th seed in a specific environment;

[0083] If the random forest algorithm is used, that is, the random forest is trained through a collection of multiple decision trees. The establishment of a decision tree depends on the selection of features. Suppose there are decision trees, and each tree outputs a prediction value. The final adaptability prediction is the average of the prediction values of each tree, expressed as:

[0084] ,

[0085] where, represents the output of the th tree, based on the seed feature and the environmental feature ; is the total number of decision trees;

[0086] If the support vector machine algorithm is used, that is, the support vector machine classifies or regresses the seeds by finding an optimal hyperplane. Using SVM for regression, it is expressed as:

[0087] ,

[0088] where, is the weight vector, representing the linear relationship between the seed feature and the environmental conditions; is the non - linear mapping function, which maps the original features to a high - dimensional space; is the bias term, represents the transpose of the weight vector;

[0089] If the neural network algorithm is used, that is, the neural network model is constructed by a multi - layer perceptron MLP, which can capture the non - linear relationship between input features. Suppose the neural network has layers, the output of the th layer is , then the output of the neural network

[0090] ,

[0091] where, is the input layer; represents the output of the th layer, represents the activation function, represents the th layer's weight matrix, represents the The bias term of the layer, representing the adjustment value of the output of the neurons in this layer, indicating the output of the layer, which is the activation value of the neurons in the previous layer, indicating the weight matrix of the output layer, connecting the last hidden layer and the output layer, representing the output of the output layer, i.e., the adaptability score of the seed in a specific environment;

[0092] The dynamic tracking and correction technology includes estimating the motion of pixels in the image through the optical flow algorithm, inferring the motion direction and speed of the seed based on the pixel changes between image frames, assuming that the seed in the image is located at position , at time and tracking between moments, and the optical flow algorithm is expressed as:

[0093] ,

[0094] where and are the velocity components of the seed along the axis and axis in the image respectively, representing the displacement speed of the seed during the screening process; and represent the displacement rate of the seed between time and moments;

[0095] At each point in the image, the change in the seed position is represented by the following optical flow constraint equation:

[0096] ,

[0097] where is the brightness value of the point in the image at time ; is the brightness value of the point in the image at time , representing the displacement amount of this point;

[0098] The Kalman filter is used for state estimation of the dynamic system to track the motion of the seed. The state variable of the seed is set as , which includes the current position and velocity of the seed. The state update equation of the Kalman filter is:

[0099] ,

[0100] Among them, represents the state vector at the th moment, including the position and velocity of the seed; is the state transition matrix, describing the system dynamics from the moment to the moment ; is the control input matrix, representing the influence of external control on the state; is the control input; is the process noise, representing the uncertainty in the system;

[0101] The measurement update equation is expressed as:

[0102] ,

[0103] Among them, is the measurement data, representing the measured position of the seed at the moment ; is the observation matrix, mapping the state vector to the measurement space; is the measurement noise, representing the sensor error;

[0104] State update:

[0105] ,

[0106] Among them, is the predicted state estimate, representing the current state predicted based on the estimate of the previous moment; is the Kalman gain; is the updated state estimate;

[0107] The screening path adjustment is expressed as:

[0108] ,

[0109] Among them, is the position of the filter at time k, is the path offset dynamically adjusted according to the state vector of the seed, represents the filter position of the previous time step.

[0110] A vision-based agricultural seed screening system, including:

[0111] An image acquisition module for acquiring three-dimensional image data of seeds using 3D imaging technology;

[0112] An image processing module for extracting the morphological, surface and internal features of seeds through image processing algorithms;

[0113] A feature screening module for screening according to the features of seeds;

[0114] A parallel screening module for parallel screening of multiple seeds through a multi-channel synchronous screening system;

[0115] An environmental screening module for screening seeds based on an environmental adaptability model in combination with specific environmental conditions;

[0116] A screening correction module for ensuring the screening accuracy by adopting dynamic tracking and correction techniques during the screening process.

[0117] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the vision-based agricultural seed screening method are implemented.

[0118] A computer-readable storage medium has a computer program stored thereon. When the computer program is executed by a processor, the steps of the vision-based agricultural seed screening method are implemented.

[0119] Advantages of the present invention: The method of the present invention can accurately obtain the three-dimensional shape, surface features and internal structure information of seeds. The present invention can comprehensively evaluate the quality of seeds, screen out high-quality seeds that meet the standards, and at the same time eliminate substandard seeds that do not meet the requirements. The multi-channel synchronous screening system optimizes the processing capabilities of each screening channel through a dynamic load balancing algorithm, significantly improving the screening efficiency. The screening method based on the environmental adaptability model intelligently screens the adaptability of seeds in combination with specific environmental conditions, further improving the accuracy of predicting the growth potential of seeds. The dynamic tracking and correction technology ensures that the seeds are always in the correct screening area during the screening process by real-time predicting the movement trajectory of the seeds, thereby improving the overall screening accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0121] Figure 1 It is a schematic flowchart of a vision-based agricultural seed screening method provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0122] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0123] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0124] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is separate or selectively mutually exclusive with other embodiments.

[0125] The present invention is described in detail in conjunction with schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views showing the device structure will be enlarged locally out of the general proportion, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0126] At the same time, in the description of the present invention, it should be noted that the orientation or positional relationships indicated by terms such as "upper, lower, inner, and outer" are based on the orientation or positional relationships shown in the drawings. It is 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 orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0127] Unless otherwise clearly defined and limited in the present invention, the terms "mounted, connected, and coupled" 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 can be indirectly connected through an intermediate medium, or can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0128] Example 1. Refer to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a vision-based agricultural seed screening method, including:

[0129] Step S1: Obtain the three-dimensional image data of the seeds using 3D imaging technology.

[0130] Step S2: Extract the morphological, surface, and internal features of the seeds through image processing algorithms.

[0131] Step S3: Screen according to the characteristics of the seeds.

[0132] Step S4: Parallelly screen multiple seeds through a multi-channel synchronous screening system.

[0133] Step S5: Screen the seeds based on the environmental adaptability model in combination with specific environmental conditions.

[0134] Step S6: Adopt dynamic tracking and correction technology during the screening process to ensure the screening accuracy.

[0135] The obtaining of the three-dimensional image data of the seeds using 3D imaging technology includes obtaining the three-dimensional image data of the seeds 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;

[0136] Define as the point cloud obtained by structured light, as the point cloud obtained by laser scanning, and construct an adaptive point cloud generation algorithm based on the fusion of structured light and laser scanning, expressed as:

[0137] ,

[0138] where and are the weighting coefficients of structured light and laser scanning respectively, is the three-dimensional point coordinate after fusion.

[0139] The image processing algorithm includes adjusting the reconstruction of each point through a weighted centroid method based on local curvature information, expressed as:

[0140] ,

[0141] where 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;

[0142] Use the optimized Monte Carlo method to calculate the three-dimensional volume of the seeds. For the case of irregular seed surfaces, use an adaptive sampling and weighted sampling strategy, expressed as:

[0143] ,

[0144] Among them, is the estimated value per unit volume, is the i-th sampling point for calculating the seed volume, is the indicator function. If falls inside the seed, then , otherwise it is is the number of sampling points; is the seed volume;

[0145] Using the combination of local volume and mass, the overall density of the seeds is calculated by the weighted average method, expressed as:

[0146] ,

[0147] Among them, is the density of each local volume unit, is the volume of this unit, is the total number of volume units, is the density of the seeds.

[0148] The screening according to the characteristics of the seeds includes constructing a multi-dimensional feature fusion model by comprehensively considering the volume and density characteristics, and further screening out the seeds that do not meet the requirements, expressed as:

[0149] ,

[0150] Among them, is the seed volume, is the average volume of all seeds, is the standard deviation of the volume; is the density of the seeds, is the average density of all seeds, is the standard deviation of the density; is the weight coefficient, used to adjust the contribution degrees of the volume and density; represents the comprehensive score of the volume and density characteristics; if exceeds the set threshold, then the seed is determined to be unqualified;

[0151] By detecting the aspect ratio and curvature characteristics of the crack at different scales and comprehensively analyzing the severity of the crack, the crack detection model is as follows:

[0152] The image is processed at multiple scales, and using different standard deviations edge detection is performed at multiple scales, and the edge intensity is expressed as:

[0153] ,

[0154] Among them, represents the Edge intensity value at the layer scale, indicating the th pixel point for edge detection in the image, indicating the indicator function, indicating the gradient value of the image at the position, being the standard deviation at the layer scale, and the gradient is obtained by calculating the rate of change of the image brightness function at this point, indicating the total number of points for edge detection;

[0155] According to the edge information of the crack, the length of the crack is calculated through morphological analysis and the area , and the severity of the crack is judged by combining the aspect ratio and curvature, expressed as: That is:

[0156] ,

[0157] where is the curvature of the crack length, reflecting the degree of bending of the crack;

[0158] For the detection of lesions and foreign objects, a deep learning model is used to extract the texture features of the image through convolution operations, and then the defect score of each seed image is output through a fully connected layer as the lesion and foreign object score, expressed as:

[0159] ,

[0160] If exceeds the set threshold, the seed is determined to have a defect;

[0161] where is the score of the seed surface defect; is the kth extracted feature; is the corresponding weight coefficient; is the total number of features for calculating the lesion and foreign object score;

[0162] Using a model that combines local texture analysis and wavelet transform, through multi-scale texture feature extraction and feature fusion, the surface irregularity of the seed is evaluated, expressed as:

[0163] ,

[0164] where 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;

[0165] After all features are fused, a multi-dimensional screening criterion is formed, expressed as:

[0166] ,

[0167] where, is the crack severity score, is the score for lesions and foreign objects, is the surface irregularity score, are the weights of each feature; represents the comprehensive score of the seeds.

[0168] The screening according to the characteristics of the seeds includes that when the volume and density of the seeds are both within the preset range, and there are no obvious cracks, lesions, foreign objects or morphological irregularities on the surface, the seeds are judged as qualified seeds; if the seeds meet the above criteria, they can directly enter the storage, marking and subsequent processing procedures; if there are minor defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, the quality of the seeds is further verified through secondary detection; if the verification result is qualified, they are retained as qualified seeds, otherwise they are transferred to the defective category for processing;

[0169] When the volume of the seeds is less than the predetermined standard range, or the density is lower than the set value, and there are no serious cracks, lesions or foreign objects, the seeds are judged as substandard seeds; if the volume is less than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seeds can enter the reprocessing stage, and their quality is restored through drying or other processing methods; if the volume and density of the seeds are restored to the qualified standard after processing, they can be re-evaluated as qualified seeds; if the seeds have surface defects such as cracks, lesions, foreign objects or morphological irregularities, they should be removed; if the volume and density of the seeds do not meet the standards and there are irreparable defects on the surface, they are directly removed;

[0170] When there are crack, lesion and foreign object defects on the surface of the seeds, the seeds are judged as defective seeds; if the crack depth is lower than the preset range or the lesion is smaller than the preset range, but it does not affect the growth potential of the seeds, through further defect analysis, that is, crack morphology analysis or lesion treatment, it is decided whether to repair or retain; if the crack exceeds the preset range or the lesion is larger than the preset range interval, and the seeds cannot be restored, they should be directly removed; if there are foreign objects on the surface of the seeds and the foreign objects cannot be removed, they should be removed; for seeds with minor defects, repair treatment is carried out, and after repair, it is re-evaluated whether they are qualified, and if they still do not meet the standards, they are removed.

[0171] The parallel screening of multiple seeds by the multi-channel synchronous screening system includes screening seeds using multiple independent channels. Each channel captures and processes seeds in real-time through a high-speed camera and an image processing module, and further combines a cooperative optimization algorithm for task scheduling and load balancing;

[0172] Minimize the total processing time of the system across all channels and the overall load of the system. The optimization objective is expressed as:

[0173] ,

[0174] where, represents the number of channels; represents the number of seeds to be screened; is the time for the th channel to process the th seed, affected by the channel load and task difficulty; is the current load of the th channel, representing the ratio between the processing capacity and the amount of tasks of the channel; represents the optimization objective, which is the weighted sum of the total processing time and the load of the system and needs to be minimized;

[0175] The processing time of each channel is jointly determined by multiple factors, including the working state of the channel, the processing difficulty of the seeds, and the load situation of the channel, and is expressed as:

[0176] ,

[0177] where, is the base time determined by the working state of the channel, the seed priority and the task difficulty; is the processing capacity weight of the th channel, representing the working efficiency of this channel relative to other channels; is the current load of the th channel, representing the amount of tasks currently assigned to this channel;

[0178] The working state of the channel will affect the processing time of each task, while the load

[0179] is dynamically adjusted according to the number and difficulty of tasks;

[0180] ,

[0181] Among them, represents the ratio of the th seed allocated to the th channel; is the processing capacity weight of the th channel; is a joint function of the channel working state, seed priority, and task difficulty, representing the time required for the th channel to process the th seed;

[0182] To avoid system overload, it is necessary to set constraints on the load of each channel. The load constraint condition is expressed as:

[0183] ,

[0184] Among them, is the maximum total load that the system can bear, preventing performance degradation or system crash caused by overloading operations;

[0185] The load of each channel 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 bearing capacity;

[0186] Set the task feedback mechanism, expressed as:

[0187] ,

[0188] Among them, is the feedback adjustment coefficient, controlling the dynamic adjustment speed of task allocation; and respectively represent the loads of the th channel before and after task adjustment; represents the task allocation adjustment amount;

[0189] Taking into account load balancing, task allocation, time constraints, and the processing capacity of the system, the final optimization goal is expressed as:

[0190] ,

[0191] Constraint conditions:

[0192] The load constraint of each channel is expressed as:

[0193] ,

[0194] The total load constraint is expressed as:

[0195] ,

[0196] The allocation ratio constraint for each channel task, expressed as:

[0197] .

[0198] 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 , where is the feature vector of the th seed, including morphological features and environmental condition information; setting the target environment as , where is the feature vector of the th environment;

[0199] Setting the environmental adaptability output as , representing the adaptability score of the th seed in a specific environment;

[0200] If the random forest algorithm is used, that is, the random forest is trained through a collection of multiple decision trees, and the establishment of the decision tree depends on the selection of features. Suppose there are decision trees, and each tree outputs a prediction value. The final adaptability prediction is the average of the prediction values of each tree, expressed as:

[0201] ,

[0202] where, represents the output of the th tree, based on the seed features and the environmental features ; is the total number of decision trees;

[0203] If the support vector machine algorithm is used, that is, the support vector machine classifies or regresses seeds by finding an optimal hyperplane. Using SVM for regression, it is expressed as:

[0204] ,

[0205] where, is the weight vector, representing the linear relationship between the seed features and the environmental conditions; is the non - linear mapping function, mapping the original features to a high - dimensional space; is the bias term, represents the transpose of the weight vector;

[0206] If the neural network algorithm is used, that is, the neural network model is constructed through a multi - layer perceptron MLP, which can capture the non - linear relationship between input features. Suppose the neural network has layer The output of the layer is , then the output of the neural network , is expressed as:

[0207] ,

[0208] Among them, is the input layer; represents the output of the -th layer, represents the activation function, represents the weight matrix of the -th layer, represents the bias term of the -th layer, indicating the adjustment value of the output of the neurons in this layer, represents the output of the -th layer, which 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, that is, the adaptability score of the seed in a specific environment, is the bias term of the output layer, adjusting the final prediction result;

[0209] The dynamic tracking and correction technology includes estimating the motion of pixels in the image through the optical flow algorithm, inferring the motion direction and speed of the seed based on the pixel changes between image frames. Assuming that the seed in the image is located at position , at time and for tracking, the optical flow algorithm, is expressed as:

[0210] ,

[0211] Among them, and are the velocity components of the seed along the axis and axis in the image, representing the displacement speed of the seed during the screening process; and represent the displacement rate of the seed between time and ;

[0212] At each point in the image, the change in the seed position is represented by the following optical flow constraint equation:

[0213] ,

[0214] Among them, is time Brightness value of the midpoint of the image at a moment ; is the time Brightness value of the midpoint of the image at a moment at, representing the displacement of this point;

[0215] The Kalman filter is used for state estimation of the dynamic system to track the movement of the seed. The state variables of the seed are set as which includes the current position and velocity of the seed. The state update equation of the Kalman filter is:

[0216] ,

[0217] where represents the state vector at the th moment, including the position and velocity of the seed; is the state transition matrix, describing the system dynamics from the moment to the moment ; is the control input matrix, representing the influence of external control on the state; is the control input; is the process noise, representing the uncertainty in the system;

[0218] The measurement update equation is expressed as:

[0219] ,

[0220] where is the measurement data, representing the measured position of the seed at the moment ; is the observation matrix, mapping the state vector to the measurement space; is the measurement noise, representing the sensor error;

[0221] State update:

[0222] ,

[0223] where is the predicted state estimate, representing the current state predicted based on the estimate of the previous moment; is the Kalman gain; is the updated state estimate, is the difference between the measured value and the predicted value;

[0224] After the real-time estimation of the seed movement state, the filter needs to dynamically adjust its path to ensure that the seed is always in the correct screening area. Assume the current position of the filter is and at each moment, according to the state estimation of the Kalman filter , the path adjustment amount of the filter is .

[0225] Filter path adjustment, expressed as:

[0226] ,

[0227] where is the position of the filter at time k, is the path offset dynamically adjusted according to the state vector of the seed , represents the filter position of the previous time step.

[0228] It should be noted that the optical flow method first estimates the speed of the seed , providing initial information for subsequent dynamic estimation. The Kalman filter then uses this speed information, combines the measurement noise and the predicted value, and corrects the motion state of the seed through the state update and measurement update steps.

[0229] The Kalman filter provides the precise dynamic state of the seed (including position and speed). The adjustment of the filter path is based on these dynamic estimation results, by calculating the offset , ensuring that the filter can adapt to the motion of the seed in real time and guaranteeing the screening accuracy.

[0230] The optical flow algorithm provides a preliminary estimation of the seed speed, the Kalman filter continuously refines the dynamic estimation of the seed through state estimation and measurement update, and 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.

[0231] Embodiment 2: An embodiment of the present invention 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.

[0232] 10,000 seeds were selected for the experiment. The seed types were normal seeds (70%), sub-variety seeds (20%) and seeds containing foreign objects (10%). Seeds of each category were randomly mixed to simulate the seed screening situation in actual agricultural production. The experimental environment was standardized lighting conditions without obvious external interference. Two different seed screening methods were used for comparative testing during the experiment.

[0233] In the experiments of traditional methods, all seeds were first processed by combining manual and mechanized screening. The manual part was mainly responsible for classifying seeds through visual and manual screening. Although manual screening could quickly determine obvious defects based on experience, its accuracy was greatly affected by the operator's ability and fatigue. Mechanized screening used an automated vibrating screen system, which initially screened seeds according to their weight, and at the same time, a camera was combined with an automatic image processing algorithm to roughly judge the morphology of the seeds. The vibrating screen screened seeds by adjusting the mesh aperture and vibration intensity. The accuracy of the mechanical screening equipment depended on the setting of physical parameters and it was difficult to achieve accurate identification of problems such as fine morphology and irregular cracks. Therefore, the accuracy of the screening results was relatively low, especially when dealing with small seed defects (such as micro-cracks, disease spots, foreign objects), the mis-screening phenomenon was serious. In addition, due to the cooperation of equipment and manual operations in the entire screening process, the screening efficiency was low.

[0234] The method of the present invention screens seeds through a multi-channel synchronous screening system based on 3D imaging. First, three-dimensional image data of the seeds is obtained by a high-speed camera, and the morphology, surface features and internal features of the seeds are extracted by combining an optimized image processing algorithm. By using structured light, laser scanning and multi-view stereo vision techniques, a three-dimensional point cloud model of the seeds is generated in real time, and the surface of the seeds is accurately adjusted and corrected by using the weighted centroid method. During the screening process, by combining the optical flow algorithm and the Kalman filter, the seeds in dynamic motion are tracked in real time, their motion trajectories are predicted, and the screening path is dynamically adjusted to ensure that the seeds are always in the correct position, improving the accuracy of screening. The experimental results are shown in Table 1.

[0235] Table 1 Comparison table of experimental results

[0236] Method Screening accuracy (%) Screening efficiency (pieces / minute) Dynamic tracking deviation (mm) Load balancing deviation (%) Adaptive screening correct rate (%) Traditional method 84.5 160 Not applicable Not applicable 67.2 Method of the present invention 97.8 470 0.6 4.1 91.3

[0237] As can be seen from Table 1, the present invention significantly improves the deficiencies of the traditional method in screening accuracy, screening efficiency, dynamic tracking, load balancing and environmental adaptability screening. The traditional screening method relies on simple mechanical principles or manual observation and cannot identify defects such as micro-cracks and disease spots, resulting in insufficient screening accuracy. The present invention accurately captures the surface and internal features of the seeds through 3D imaging technology, and combines the weighted centroid method and the optimized image processing algorithm to effectively improve the ability to identify complex seed morphologies, thereby greatly improving the screening accuracy.

[0238] In terms of screening efficiency, traditional single-channel devices are limited by mechanical structures and manual participation, making it difficult to meet the requirements of high-efficiency processing. The present invention introduces a multi-channel synchronous screening system, which dynamically allocates tasks through a collaborative optimization algorithm, realizes parallel processing, and significantly improves the seed screening speed. In addition, the combination of the optical flow algorithm and the Kalman filter enables the dynamic tracking technology to predict the seed movement trajectory in real time and precisely adjust the screening path, ensuring accurate positioning of the seeds during the screening process and avoiding mis-screening problems caused by path deviation.

[0239] In terms of load balancing, the dynamic task allocation technology of the present invention ensures uniform load on each channel of the multi-channel screening system, significantly improving the equipment utilization rate and screening stability. The screening method combined with the environmental adaptability model comprehensively analyzes the matching degree between seed characteristics and environmental conditions through machine learning technology, making the selected seeds more in line with the requirements of the target environment. In summary, the present invention effectively solves the defects in traditional methods through a number of innovative technical solutions, comprehensively improving the efficiency and accuracy of the screening process.

[0240] Embodiment 3: This is an embodiment of the present invention, which provides a vision-based agricultural seed screening method and system, including:

[0241] An image acquisition module, which is used to obtain three-dimensional image data of seeds by using 3D imaging technology;

[0242] An image processing module, which is used to extract the morphological, surface and internal characteristics of seeds through image processing algorithms;

[0243] A feature screening module, which is used to screen according to the characteristics of seeds;

[0244] A parallel screening module, which is used to perform parallel screening on multiple seeds through a multi-channel synchronous screening system;

[0245] An environment screening module, which is used to screen seeds based on the environmental adaptability model in combination with specific environmental conditions;

[0246] A screening correction module, which is used to ensure the screening accuracy by adopting dynamic tracking and correction technology during the screening process.

[0247] Embodiment 4: The fourth embodiment of the present invention is different from the first three embodiments in that:

[0248] If the above-mentioned 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, in essence, or the part that contributes to the prior art, or a 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 and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0249] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0250] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0251] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0252] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0253] 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 equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A vision-based agricultural seed screening method, characterized in that: including, acquiring three-dimensional image data of seeds using 3D imaging technology; extracting the morphological, surface, and internal features of seeds through image processing algorithms; screening seeds based on their features, selecting high-quality seeds that meet the standards while rejecting substandard seeds that do not meet the requirements; parallelly screening multiple seeds through a multi-channel synchronous screening system; screening seeds based on an environmental adaptability model in combination with specific environmental conditions; adopting dynamic tracking and correction technology during the screening process to ensure screening accuracy; The parallel screening of multiple seeds through the multi-channel synchronous screening system includes screening seeds using multiple independent channels. Each channel captures and processes seeds in real-time through a high-speed camera and an image processing module, and further combines a cooperative optimization algorithm for task scheduling and load balancing; minimizing the total processing time of the system across all channels and the overall load of the system. The optimization objective is expressed as: Among them, P represents the number of channels; Q represents the number of seeds to be screened; T i,j is the time for the i-th channel to process the j-th seed, which is affected by the channel load and task difficulty; L i is the current load of the i-th channel, representing the ratio between the processing capacity of the channel and the task volume; Z represents the optimization objective, which is the weighted sum of the total processing time and load of the system and needs to be minimized; The processing time T of each channel i,j is jointly determined by multiple factors, including the working state of the channel, the processing difficulty of the seeds, and the load situation of the channel, and is expressed as: Among them, f(C i , P j , D j ) is the base time determined by the working state C i of the channel, the seed priority P j and the task difficulty D j ; W i is the processing capacity weight of the i-th channel, indicating the working efficiency of this channel relative to other channels; L i is the current load of the i-th channel, indicating the amount of tasks already assigned to this channel; Working status C of the channel i will affect the processing time of each task, while the load L i is dynamically adjusted according to the number and difficulty of tasks; To achieve load balancing, a dynamically adjusted task allocation strategy is adopted. According to the processing capabilities and load conditions of each channel, the task allocation is adjusted in real-time. The load balancing formula is expressed as: Among them, A i,j represents the proportion of the j-th seed assigned to the i-th channel; W i is the processing capacity weight of the i-th channel; f(C i , P j , D j ) is a joint function of the channel working state, seed priority, and task difficulty, representing the time required for the i-th channel to process the j-th seed; To avoid system overload, constraints need to be set on the load of each channel. The load constraint condition is expressed as: Among them, L max is the maximum total load that the system can withstand, preventing performance degradation or system crashes caused by overloading operations; The load L of each channel i 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 bearing capacity; Set a task feedback mechanism, expressed as: Among them, η is the feedback adjustment coefficient, which controls the dynamic adjustment speed of task allocation; and respectively represent the loads of the i-th channel before and after task adjustment; ΔA i,j represents the task allocation adjustment amount; Taking into account load balancing, task allocation, time constraints, and the processing capabilities of the system, the final optimization objective is expressed as: Constraint conditions: The load constraint of each channel, expressed as: L i ≤L max , i = 1, 2, …, P The total load constraint, expressed as: The task allocation ratio constraint of each channel, expressed as:

2. The visual-based agricultural seed screening method according to claim 1, characterized in that: The acquisition of three-dimensional image data of seeds using 3D imaging technology includes acquiring three-dimensional image data of seeds using structured light, laser scanning, or multi-view stereo vision technology. The three-dimensional image data is used to generate a three-dimensional point cloud model of the seeds; Define P SG is the point cloud obtained by structured light, and P LS is 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: P fused = λ SG ·P SG + λ LS ·P LD Among them, λ SG and λ LS are the weighting coefficients of structured light and laser scanning respectively, and P fused is the three-dimensional point coordinates after fusion.

3. The method for visually-based agricultural seed screening according to claim 2, characterized in that: The image processing algorithm includes adjusting the reconstruction of each point through a weighted centroid method based on local curvature information weighting, expressed as: Among them, P new 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; Using an optimized Monte Carlo method to calculate the three-dimensional volume of seeds. For the case of irregular seed surfaces, an adaptive sampling and weighted sampling strategy is used, expressed as: where V unit (P volume,o ) is the estimated value per unit volume, P volume,o is the o-th sampling point used to calculate the seed volume, is an indicator function, if P volume,o falls inside the seed, then otherwise it is 0; N is the number of sampling points; V seed is the seed volume; Using the combination of local volume and mass, and adopting a weighted average method to calculate the overall density of seeds, expressed as: Among them, is the density of each local volume unit, is the volume of the unit, N volume is the total number of volume units, and ρ is the density of the seeds.

4. The visual-based agricultural seed screening method according to claim 3, characterized in that: The screening based on the features of seeds includes: By comprehensively considering volume and density features, constructing a multi-dimensional feature fusion model to further screen out seeds that do not meet the requirements, expressed as: Among them, V seed is the seed volume, is the volume mean of all seeds, σ V is the standard deviation of volume; is the density of the seeds, is the density mean of all seeds, is the standard deviation of density; θ1, θ2 are weight coefficients used to adjust the contribution degrees of volume and density; If F volume exceeds the set threshold, the seed is determined to be unqualified; F volume represents the comprehensive score of volume and density characteristics; By detecting the aspect ratio and curvature features of cracks at different scales to comprehensively analyze the severity of cracks. The crack detection model is as follows: Perform multi-scale processing on the image and utilize different standard deviations σ k Perform edge detection at multiple scales, and the edge intensity is expressed as: Among them, E k represents the edge intensity value at the k-th layer scale, represents the o3-th pixel point for edge detection in the image, represents the indicator function, represents the gradient value of the image at the position, and σ k is the standard deviation at the k-th layer scale, and the gradient is obtained by calculating the rate of change of the image luminance function I at this point, and N edge represents the total number of points for edge detection; Based on the edge information of the crack, calculate the length L of the crack through morphological analysis crack and the area A crack , and combine the aspect ratio and curvature to judge the severity C of the crack severity , expressed as: where Curvature(L crack ) is the curvature of the crack length, reflecting the degree of crack bending; The detection of lesions and foreign objects uses a deep learning model. Convolution operations are used to extract the texture features of the image, and then the defect score of each seed image is output through a fully connected layer as the lesion and foreign object score, expressed as: If S defect exceeds the set threshold, the seed is determined to be defective; Among them, S defect is the score of the seed surface defect; is the k1-th feature extracted; is the corresponding weight coefficient; N feature is the total number of features for calculating the scores of the lesions and foreign objects; Using a model that combines local texture analysis and wavelet transform to evaluate the surface irregularity of seeds through multi-scale texture feature extraction and feature fusion, expressed as: Among them, N wavelet is the total number of wavelet scales, and N k is the number of sampling points at this scale, is the ripple texture feature at the k-th layer scale, and μ k is the mean value of the texture at this scale; After all features are fused, a multi-dimensional screening standard is formed, expressed as: F total = w1·F volume + w2·C severity + w3·S defect + w4·R texture Among them, C severity is the crack severity score, S defect is the lesion and foreign object score, R texture is the surface irregularity score, and w1, w2, w3, w4 are the weights of each feature; F total represents the comprehensive score of the seeds.

5. The method for visually-based agricultural seed screening according to claim 4, wherein: The screening according to the characteristics of the seeds includes that when the volume and density of the seeds are both within the preset range, and there are no obvious cracks, disease spots, foreign objects or morphological irregularities on the surface, the seeds are determined to be qualified seeds; if the seeds meet the above criteria, they will directly enter the storage, marking and subsequent processing procedures; if there are minor defects on the surface of the seeds, and the volume and density are close to the lower limit of the standard, the quality of the seeds will be further verified through secondary detection; If the verification result is qualified, they will be retained as qualified seeds, otherwise they will be transferred to the defective category for processing; When the volume of the seeds is less than the predetermined standard range, or the density is lower than the set value, and there are no serious cracks, disease spots or foreign objects, the seeds are determined to be substandard seeds; if the volume is less than the preset range and the density is lower than the preset range, but there are no obvious defects on the surface, the seeds will enter the reprocessing stage, and their quality will be restored through drying or other processing methods; If the volume and density of the seeds are restored to the qualified standard after processing, they will be re-evaluated as qualified seeds; if the seeds have surface defects such as cracks, disease spots, foreign objects or morphological irregularities, they should be removed; if the volume and density of the seeds do not meet the standards and there are irreparable defects on the surface, they will be directly removed; When there are cracks, disease spots, foreign object defects on the surface of the seeds, the seeds are determined to be defective seeds; if the crack depth is lower than the preset range or the disease spot is smaller than the preset range, but it does not affect the growth potential of the seeds, through further defect analysis, that is, crack morphology analysis or disease spot treatment, it is decided whether to repair or retain them; if the crack exceeds the preset range or the disease spot is larger than the preset range interval and the seeds cannot be restored, they should be directly removed; If there are foreign objects on the surface of the seeds and the foreign objects cannot be removed, they should be removed; For seeds with minor defects, repair treatment will be carried out, and after repair, it will be re-evaluated whether they are qualified. If they still do not meet the standards, they will be removed.

6. The method for visually-based agricultural seed screening according to claim 5, wherein: The environmental adaptability model includes: Predict the adaptability of seeds in a specific environment through a machine learning model, and set the seed characteristics as X c ={x c1 ,x c2 ,…,x cn}, where X c is the feature vector of the c-th seed, including morphological characteristics and environmental condition information; set the target environment as E j′ ={e j'1 ,e j′2 ,…,e j′m}, where E j′ is the feature vector of the j'-th environment; Set the environmental adaptability output to y c , representing the adaptability score of the c-th 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. Suppose there are T decision trees, and each tree outputs a prediction value. The final adaptability prediction is the average of the prediction values of each tree, which is expressed as: Among them, represents the output of the t1-th tree, based on the seed feature X c and the environmental feature E j′ ; T rf 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. Using SVM for regression, it is expressed as: y c = w T φ(X c , E j′ ) + b Among them, w is the weight vector, representing the linear relationship between the seed features and the environmental conditions; φ(X c ,E j′ ) is a non-linear mapping function that maps the original features to a high-dimensional space; b is the bias term, and w T represents the transpose of the weight vector; If the neural network algorithm is used, that is, the neural network model is constructed by a multi-layer perceptron MLP, which can capture the non-linear relationship between input features. Assume that the neural network has L layers, and the output of the l-th layer is h (l) , then the output y of the neural network i , is expressed as: h (l) = σ(W (l) h (l-1) + b (l) ) y i = W (L) h (L) + b (L) Among them, h (0) = X i is the input layer; h (l) represents the output of the l-th layer, σ represents the activation function, W (l) represents the weight matrix of the l-th layer, b (l) represents the bias term of the l-th layer, which represents the adjustment value of the output of the neurons in this layer. h (l-1) represents the output of the (l - 1)-th layer, which is the activation value of the neurons in the previous layer. W (L) represents the weight matrix of the output layer, connecting the last hidden layer and the output layer. h (L) represents the output of the output layer, that is, the adaptability score of the seed in a specific environment. b (L) 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 the optical flow algorithm, and inferring the movement direction and speed of the seeds based on the pixel changes between image frames. Suppose the seeds in the image are located at the position (x, y), and tracking is carried out between time t and t + 1. The optical flow algorithm is expressed as: Wherein, u and v are respectively the velocity components of the seed along the x-axis and y-axis in the image, representing the displacement velocity of the seed during the screening process; and represents the displacement rate of the seed between times t and t + 1; At each point (x, y) in the image, the change in the seed position is represented by the following optical flow constraint equation: I(x,y,t)=I(x+u,y+v,t+1) Where, I(x, y, t) is the brightness value of the point (x, y) in the image at time t; I(x + u, y + v, t + 1) is the brightness value of the point (x + u, y + v) in the image at time t + 1, representing the displacement of this point; The Kalman filter is used for state estimation of a dynamic system to track the movement of the seeds. The state vector of the seeds is set as x(k2), which includes the current position x′(k2) and velocity v'(k2) of the seeds. The state update equation of the Kalman filter is as follows: x(k2) = F(k2 - 1)x(k2 - 1) + B(k2 - 1)u(k2 - 1) + w(k2 - 1) Among them, represents the state vector at time k2, including the position and velocity of the seed; F(k2 - 1) is the state transition matrix, describing the system dynamics from time k2 - 1 to time k2; B(k2 - 1) is the control input matrix, representing the influence of external control on the state; u(k2 - 1) is the control input; w(k2 - 1) is the process noise, representing the uncertainty in the system; The measurement update equation is expressed as: z(k2) = H(k2)x(k2) + v(k2) where z(k2) is the measurement data, representing the measured position of the seeds at time k2; H(k2) is the observation matrix, which maps the state vector x(k2) to the measurement space; v(k2) is the measurement noise, representing the sensor error; State update: x(k2|k2) = x(k2|k2 - 1) + K(k2)(z(k2) - H(k2)x(k2|k2 - 1)) where x(k2|k2 - 1) is the predicted state estimate, representing the current state predicted based on the estimate of the previous time step; K(k2) is the Kalman gain; x(k2|k2) is the updated state estimate; Screening path adjustment is expressed as: p(k2) = p(k2 - 1) + Δp(k2) where p(k2) is the position of the filter at time k2, Δp(k2) is the path offset dynamically adjusted according to the state vector x(k2) of the seeds, and p(k2 - 1) represents the position of the filter in the previous time step.

7. A system adopting a vision-based agricultural seed screening method as described in any one of claims 1 to 6, characterized in that: Including, An image acquisition module for obtaining three-dimensional image data of the seeds using 3D imaging technology; An image processing module for extracting the morphology, surface features, and internal features of the seeds through image processing algorithms; A feature screening module for screening according to the features of the seeds; A parallel screening module for parallel screening of multiple seeds through a multi-channel synchronous screening system; An environment screening module for screening seeds based on an environmental adaptability model in combination with specific environmental conditions; A screening correction module for ensuring the screening accuracy by adopting dynamic tracking and correction techniques during the screening process.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method described in any one of claims 1 to 6.

Images