Multi-dimensional data analysis method for breeding screening

Abstract

The invention discloses a multi-dimensional data analysis method for breeding screening, and relates to the technical field of breeding screening, and the method comprises the following steps: laying pepper seeds of a to-be-detected batch in a detection area, and dividing the to-be-detected area into a plurality of sub-areas; according to the method, the texture complexity index and the porosity index are introduced, and the pre-training machine learning model is combined to perform intelligent analysis on the surface microstructure characteristics of the pepper seeds, so that the screening precision is improved. The system flexibly adjusts the number and position of detection windows, accurately captures complex microstructure features, avoids misjudgment of high-quality seeds as poor-quality seeds, and improves the screening accuracy. Meanwhile, simple and complex surface structures are intelligently divided, resource allocation is optimized, the detection cost and time are saved, and an efficient and economical screening scheme is provided for the breeding process.

Description

Technical Field

[0001] The present invention relates to the technical field of breeding and screening, and in particular to a multidimensional data analysis method for breeding and screening. Background Art

[0002] Multidimensional data analysis for breeding screening refers to the collection and processing of data from multiple dimensions to evaluate and select seeds with superior genetic characteristics. In the screening process of pepper seeds, this analysis can help breeders comprehensively consider the physical characteristics of seeds, such as seed size, shape, color, density, and moisture content, and then judge their germination potential and growth performance. With high-precision pepper seed screening equipment, breeders can quickly measure and analyze these physical characteristics, thereby improving screening efficiency and accuracy.

[0003] In pepper seed quality screening, multidimensional data analysis can reveal the relationship between different seed characteristics and help breeders identify the best seed sources. For example, using optical sensors and imaging technology, pepper seed screening equipment can capture the surface characteristics and internal structure of seeds to assess their quality. Combined with machine learning algorithms, these data can not only be used to monitor the physical properties of seeds in real time, but also predict the performance of different seeds in the future growth process through the analysis of historical data, and further optimize breeding decisions.

[0004] The prior art has the following deficiencies:

[0005] The surface microstructure of pepper seeds refers to the subtle morphology and texture characteristics of the outer surface of the seeds, including surface bumps, pores and lines. These microstructures not only affect the appearance and touch of seeds, but also play an important role in the germination and growth of seeds. For example, certain microstructures help to enhance the adhesion between seeds and soil, improve the permeability of water and air, and thus promote rapid germination of seeds under suitable conditions. In addition, microstructures may also affect the resistance of seeds to diseases and environmental stress, thereby improving the survival rate and healthy growth of seeds.

[0006] In existing image recognition technology, a fixed number of observation areas are usually randomly selected for a batch of pepper seeds, and high-resolution image acquisition is performed to identify the surface microstructure features of the seeds. However, when the surface microstructure of some pepper seeds is too complex, the fixed number of observation areas may not be able to accurately capture these features. This deficiency may cause high-quality seeds to be misjudged as low-quality seeds, thus affecting the screening results, and then using unsuitable seeds during the planting process, ultimately affecting the growth and yield of the crop.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0008] The purpose of the present invention is to provide a multidimensional data analysis method for breeding screening, which introduces texture complexity index and porosity index, combines pre-trained machine learning model to intelligently analyze the surface microstructure characteristics of pepper seeds, and achieves the improvement of screening accuracy. The system flexibly adjusts the number and position of detection windows, accurately captures complex microstructure characteristics, avoids misjudging high-quality seeds as low-quality seeds, and improves the accuracy of screening. At the same time, it intelligently divides simple and complex surface structures, optimizes resource allocation, saves detection costs and time, and provides an efficient and economical screening solution for the breeding process to solve the problems in the above-mentioned background technology.

[0009] In order to achieve the above object, the present invention provides the following technical solution: a multidimensional data analysis method for breeding screening, comprising the following steps:

[0010] Spreading the pepper seeds of the batch to be tested in the testing area, and dividing the testing area into multiple sub-areas;

[0011] In the divided sub-areas, sub-areas are randomly selected based on the preset number of detection windows for data collection to obtain the seed surface structure information;

[0012] Extract the seed surface structure information in the sub-region. Based on the extracted seed surface microstructure features, use the pre-learned machine learning model to intelligently evaluate the complexity of the seed surface structure in the sub-region under the detection window to identify different surface structure types of the seeds.

[0013] Based on the evaluation results of the machine learning model, the surface structure of pepper seeds is divided into complex surface structure and simple surface structure;

[0014] For the simple surface structures identified, sub-areas are continuously randomly selected for data collection based on the preset number of detection windows;

[0015] For complex surface structures, based on the preset number of detection windows and the evaluation results, the number and position of detection windows are dynamically adjusted to comprehensively capture the complex microstructure features.

[0016] Preferably, feature extraction is performed on the seed surface structure information in the sub-area, and the extracted features include the degree of diversity change of the surface texture and the distribution characteristics of the number of pores on the surface. Under the monitoring window, the degree of diversity change of the extracted surface texture and the distribution characteristics of the number of pores on the surface are analyzed, and then a texture complexity index and a porosity index are generated respectively. The texture complexity index is used to measure the diversity and irregularity of the seed surface texture, and the porosity index is used to measure the number, size and distribution characteristics of the pores on the seed surface. The texture complexity index and the porosity index are input into a pre-learned machine learning model, and a microstructure coefficient is generated by the machine learning model. The complexity of the seed surface structure is intelligently evaluated by the microstructure coefficient to identify different surface structure types of seeds.

[0017] Preferably, the specific steps of analyzing the degree of diversity change of surface textures under the detection window and generating the texture complexity index are as follows:

[0018] Under the detection window, the collected seed surface data is preprocessed, and the texture features are extracted using the gray level co-occurrence matrix technology for the preprocessed data. The calculation expression is as follows:

[0019]

[0020] , where G(i,j) is the element of the gray-level co-occurrence matrix, which indicates the frequency of adjacent appearance of pixels with gray-level values ​​i and j in the image, and Q(i,j|x,y) indicates the conditional probability of the occurrence of a pair of pixels with gray-level values ​​i and j at position (x,y);

[0021] The texture energy and texture contrast are calculated using the extracted gray level co-occurrence matrix. The texture energy calculation expression is as follows:

[0022]

[0023] , where E is the texture energy, which measures the uniformity of the texture;

[0024] The texture contrast calculation expression is as follows:

[0025]

[0026] , where C is the texture contrast, describing the degree of texture change;

[0027] Calculate the entropy value of the texture and further evaluate the complexity of the texture. The calculation expression is as follows:

[0028]

[0029] , where H is the texture entropy value, ∈ is a small constant used to avoid the zero value problem in logarithmic calculation;

[0030] Combining texture energy E, texture contrast C and texture entropy value H, the texture complexity index is generated. The calculation expression is as follows:

[0031]

[0032] , where TCI is the texture complexity index, max(E) is the maximum texture energy, max(C) is the maximum texture contrast, max(H) is the maximum texture entropy, A is the adjustment parameter of texture energy E, B is the adjustment parameter of texture contrast C, F is the power adjustment parameter of texture contrast, δ is the offset adjustment parameter of texture entropy value H, λ is the power adjustment parameter of texture entropy value H, and η is a non-zero positive number used to control the smoothness of the final result.

[0033] Preferably, under the detection window, the specific steps of analyzing the distribution characteristics of the number of pores on the surface and generating the porosity index are as follows:

[0034] In the detection window, the acquired image information is preprocessed, and the image processing algorithm is used to identify pores on the preprocessed image information. The expression is as follows:

[0035]

[0036] , where A p is the total pore area, N p is the number of pores, A y is the area of ​​the yth pore;

[0037] The calculation formula for porosity is:

[0038]

[0039] , where P is the porosity, A is t is the total area of ​​the observation region;

[0040] On the basis of porosity, the pore shape factor is introduced and the porosity is corrected by its influence. The calculation expression is as follows:

[0041]

[0042] , where S is the shape factor, N p is the number of pores, L y is the perimeter of the yth aperture, A y is the area of ​​the yth pore;

[0043] The porosity is corrected based on the influence of the shape factor S, and the calculation expression is as follows:

[0044]

[0045] , where P adj is the corrected porosity;

[0046] Finally, the porosity index is generated, and the calculation expression is as follows:

[0047]

[0048] , where PI is the porosity index, k is the normalization constant, and 1-P adj represents the proportion of non-porous area in the detection area, and ω are tuning parameters, Used to adjust the corrected porosity P adj The influence of ω is used to adjust the influence of the non-pore area in the detection area, e -ρS is the exponential weight of the shape factor, ρ is the weight of the shape factor, e is the natural base, and δ is the uniformity adjustment parameter.

[0049] Preferably, a microstructure coefficient generated after analyzing the seed surface structure information in the sub-region under the detection window is compared and analyzed with a preset microstructure coefficient reference threshold, and the seed surface structure state in the sub-region is divided;

[0050] If the microstructure coefficient is greater than or equal to a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a complex surface structure;

[0051] If the microstructure coefficient is less than a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a simple surface structure.

[0052] Preferably, for complex surface structures, based on a preset number of detection windows, the number and position of the detection windows are dynamically adjusted according to the evaluation results to comprehensively capture the complex microstructure features. The specific steps are as follows:

[0053] According to the evaluation results, i.e., the complex surface structure, the number of detection windows is dynamically adjusted. Considering the influence of the changes of the texture complexity index TCI and the porosity index PI on the number of detection windows, a weighted adjustment method is adopted. The calculation expression is as follows:

[0054]

[0055] , where N adjusted is the number of detection windows after adjustment, N initial is the preset number of detection windows, TCI rt is the preset texture complexity index reference threshold, PI rtis the preset porosity index reference threshold, α and β are adjustment factors, which reflect the influence of texture complexity and porosity change on the number of detection windows, respectively;

[0056] After adjusting the number of windows, the position of the new window is determined. Based on the distribution of complex surface features, the cluster analysis method is used to identify areas with large changes in surface features. The expression is as follows:

[0057] P selected =Top(N adjusted ,{P r |r∈rank(C r )})

[0058] , where P selected is the selected detection window position set, P r is the position of the candidate detection window, the rth position in the position set of all candidate detection windows, P r = {P1, P2, P3, ..., P u}, u is the total number of candidate detection windows, C r is the candidate detection window P r The complexity evaluation is the rth candidate detection window position P r The complexity index, rank(C r ) is the order of complexity evaluation;

[0059] Based on the feedback of complex surface features, the position of the window is evaluated again. If some positions fail to fully cover the features, the position can be adjusted. The calculation expression is as follows:

[0060] P revised =P selected +ΔP

[0061] , where P revised is the adjusted detection window position, ΔP is the position offset, which is used to adjust the window position after feedback;

[0062] Finally, loop optimization is performed based on the new number and position of windows. The calculation expression is as follows:

[0063]

[0064] , where N final is the number of detection windows after final adjustment, γ is the feedback adjustment factor, C new is the current complexity index, C old is the complexity index of the last evaluation.

[0065] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0066] The present invention can intelligently analyze the surface microstructure characteristics of pepper seeds by introducing texture complexity index and porosity index and inputting them into a pre-trained machine learning model. This method allows the screening process to not only focus on the basic morphology of the seed surface, but also accurately identify different microstructure types and their complexity. This automated evaluation process avoids the limitations of traditional random sampling methods and reduces the risk of high-quality seeds being misjudged as low-quality seeds because the complex microstructure is not fully captured. At the same time, by dynamically adjusting the number and position of the detection windows, the system can be flexibly adjusted according to the complexity of the actual microstructure to ensure that the complex microstructure characteristics are fully captured, thereby improving the accuracy and reliability of the screening process.

[0067] The present invention intelligently divides the complexity of the seed surface structure, and divides the surface microstructure into "complex surface structure" and "simple surface structure", so that suitable collection and detection resources can be allocated for different seed characteristics. For simple surface structures, random collection is only required based on a fixed number of detection windows, saving unnecessary resource investment; while for complex surface structures, by dynamically adjusting the number and position of the collected detection windows, the seed characteristics can be analyzed more comprehensively to achieve efficient use of resources. This method significantly improves the overall detection efficiency, which not only reduces time and cost consumption in the screening process, but also ensures that resources can be concentrated on the identification of complex surface structures, ultimately providing a high-quality, cost-effective and efficient screening solution for the breeding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0069] Figure 1 The present invention is a method flow chart of a multidimensional data analysis method for breeding screening. DETAILED DESCRIPTION

[0070] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0071] The present invention provides Figure 1 A multidimensional data analysis method for breeding screening is shown, comprising the following steps:

[0072] Spreading the pepper seeds of the batch to be tested in the testing area, and dividing the testing area into multiple sub-areas;

[0073] Divide the inspection area into multiple sub-areas. Divide according to the number and size of seeds, usually using a gridding method to make each sub-area of ​​moderate size for subsequent random sampling. The divided sub-areas not only help to manage samples systematically, but also improve the efficiency and accuracy of image acquisition, ensuring that the collected data can fully reflect the characteristics of the entire batch of seeds.

[0074] In the divided sub-areas, sub-areas are randomly selected based on the preset number of detection windows for data collection to obtain the seed surface structure information;

[0075] In the divided sub-regions, random sampling is performed based on the preset number of detection windows. This process is designed to ensure that each sub-region can represent the characteristics of the entire batch of seeds. In specific operations, a random number generation algorithm can be used to select specific sub-regions so that the probability of each sub-region being extracted is equal. Such random sampling not only avoids the bias of sample selection, but also effectively covers the surface characteristics of various types of seeds, thereby enhancing the representativeness and reliability of the data.

[0076] After random sampling, the data collection stage begins. Usually, high-resolution image acquisition equipment, such as a microscope or a high-resolution camera, is used to capture images of the selected sub-areas. At this time, it is necessary to ensure that the light source is uniform and the focal length is adjusted appropriately so that the microstructural features of the seed surface can be clearly captured. Through image processing software, the collected images can be further pre-processed, such as denoising and contrast enhancement, to improve the accuracy of subsequent feature extraction. This combination of random sampling and efficient data collection helps to obtain high-quality sample data and lay a solid foundation for subsequent analysis and evaluation.

[0077] Extract the seed surface structure information in the sub-region. Based on the extracted seed surface microstructure features, use the pre-learned machine learning model to intelligently evaluate the complexity of the seed surface structure in the sub-region under the detection window to identify different surface structure types of the seeds.

[0078] Feature extraction is performed on the seed surface structure information in the sub-region. The extracted features include the diversity variation degree of the surface texture and the distribution characteristics of the number of pores on the surface. Under the monitoring window, the diversity variation degree of the extracted surface texture and the distribution characteristics of the number of pores on the surface are analyzed, and then a texture complexity index and a porosity index are generated respectively. The texture complexity index is used to measure the diversity and irregularity of the seed surface texture, and the porosity index is used to measure the number, size and distribution characteristics of the pores on the seed surface. The texture complexity index and the porosity index are input into a pre-learned machine learning model, and a microstructure coefficient is generated through the machine learning model. The complexity of the seed surface structure is intelligently evaluated through the microstructure coefficient to identify different surface structure types of seeds.

[0079] When the diversity of surface textures obtained in a sub-region changes abnormally, it can indeed indicate that the surface microstructure of seeds in the sub-region is in a complex state. This is because the surface microstructure of seeds is composed of a variety of morphological features, including uneven surfaces, pores of different sizes and shapes, and diverse texture patterns. Under normal circumstances, the diversity of surface textures will remain within a certain range, reflecting the health and consistency of seeds. However, when this diversity increases significantly or becomes abnormal, it means that the morphological characteristics of the seed surface may become more complex, which may be due to genetic variation, environmental factors, or abnormal changes in seed development. Complex microstructures may include more delicate lines, abnormal protrusions, and higher roughness, making the identification and classification of surface features more difficult. Therefore, abnormal surface texture diversity not only reflects the complexity of the microstructure, but also may indicate that there are significant differences in the morphology of seeds, which complicates the judgment of seed quality and affects the effective screening and evaluation of seeds in the sub-region. By monitoring this change, seeds with complex surface microstructures can be better identified, which provides an important basis for subsequent research and application.

[0080] Under the detection window, the diversity of surface texture is analyzed and the specific steps of generating texture complexity index are as follows:

[0081] Under the detection window, the collected seed surface data is preprocessed, and the texture features are extracted using the gray level co-occurrence matrix (GLCM) technology for the preprocessed data. The calculation expression is as follows:

[0082]

[0083] , where G(i, j) is an element of the gray-level co-occurrence matrix, indicating the frequency of adjacent appearance of pixels with gray-level values ​​i and j in the image, and Q(i, j|x, y) indicates the conditional probability of the occurrence of a pair of pixels with gray-level values ​​i and j at position (x, y);

[0084] Gray-Level Co-occurrence Matrix (GLCM) is a statistical method used to describe the spatial relationship of image grayscale. It reflects the texture characteristics of an image by calculating the spatial co-occurrence frequency of pixel pairs with different grayscale levels in the image. Specifically, each element G(i, j) of the gray-level co-occurrence matrix represents the number or probability of adjacent pixel pairs with grayscale values ​​i and j in the image. By defining the relative position between pixels (e.g. horizontal, vertical or diagonal direction), the gray-level co-occurrence matrix can reveal the spatial relationship between pixels in the image and capture the texture information of the image.

[0085] Here, the role of the gray-level co-occurrence matrix is ​​to extract the texture features of the pepper seed surface for generating a texture complexity index. By calculating the energy, contrast, entropy and other indicators of the gray-level co-occurrence matrix, the texture complexity of the seed surface can be quantified and simple and complex surface structures can be distinguished. These texture features can help evaluate the quality of seeds, because the texture characteristics of the seed surface are closely related to its germination and growth potential. As a basic tool for texture analysis, the gray-level co-occurrence matrix can provide key data support for seed screening.

[0086] The texture energy and texture contrast are calculated using the extracted gray level co-occurrence matrix. The texture energy calculation expression is as follows:

[0087]

[0088] , where E is the texture energy, which measures the uniformity of the texture;

[0089] The texture contrast calculation expression is as follows:

[0090]

[0091] , where C is the texture contrast, describing the degree of texture change;

[0092] Texture energy is used to measure the uniformity and repeatability of texture in an image. The higher the texture energy value, the higher the frequency of pixel pairs with the same grayscale value in the image, that is, the more uniform and repeatable the texture of the image is. For the surface texture of pepper seeds, high texture energy usually means that the surface structure is relatively flat or regular, which may indicate that the seeds have good quality because surface uniformity may help the uniform absorption of water and nutrients.

[0093] Texture contrast is used to describe the degree of texture change in an image, reflecting the size of the grayscale value difference in the image. The higher the contrast value, the greater the grayscale difference in the image, and the more complex or irregular the texture. For the seed surface, a higher texture contrast may mean that there are more subtle bumps or texture features on the surface, which may affect the contact between the seed and the soil and its ability to absorb water. Combining texture energy and contrast can more comprehensively evaluate the surface quality of seeds, thereby helping to screen out seeds with excellent germination potential.

[0094] Calculate the entropy value of the texture and further evaluate the complexity of the texture. The calculation expression is as follows:

[0095]

[0096] , where H is the texture entropy value, ∈ is a small constant used to avoid the zero value problem in logarithmic calculation;

[0097] Combining texture energy E, texture contrast C and texture entropy value H, the texture complexity index is generated. The calculation expression is as follows:

[0098]

[0099] , where TCI is the texture complexity index, max(E) is the maximum texture energy, max(C) is the maximum texture contrast, max(H) is the maximum texture entropy, A is the adjustment parameter of texture energy E, B is the adjustment parameter of texture contrast C, F is the power adjustment parameter of texture contrast, δ is the offset adjustment parameter of texture entropy value H, λ is the power adjustment parameter of texture entropy value H, and η is a non-zero positive number used to control the smoothness of the final result;

[0100] The expression for calculating the texture complexity index shows that, under the detection window, the larger the value of the texture complexity index generated after analyzing the degree of diversity change of the surface texture, the more complex the seed surface microstructure is. This is because a higher complexity index usually reflects the presence of diverse morphological features on the seed surface, such as bumps and lines of different shapes, sizes and arrangements, indicating that the complexity of the surface structure has increased. Relatively speaking, a lower complexity index indicates that the surface texture is relatively simple, lacking obvious diversity and change, indicating that the seed surface microstructure is in a simple structure. Therefore, the texture complexity index can effectively quantify and reflect the complexity of the seed surface microstructure, thereby providing an important basis for the analysis of seed characteristics.

[0101] When the number and distribution characteristics of the pores on the seed surface obtained in a sub-region are abnormal, this usually indicates that the surface microstructure of the seed in this sub-region is in a complex state. The number and distribution characteristics of pores are one of the important indicators for evaluating surface microstructure. Under normal circumstances, the pores on the seed surface should show a certain regularity and uniformity. However, when the number of pores increases or decreases significantly, or their distribution shows uneven and random characteristics, it means that the formation of the surface microstructure may be affected by multiple factors, such as genetic variation, environmental conditions or different growth stages. This abnormal pore distribution often reflects the complex changes in the morphology, texture or physical properties of the surface, which may lead to an increase in the diversity of surface characteristics and the formation of complex microstructures. Complex microstructures may include pores of multiple sizes and shapes. This diversity makes the interaction between the surface and the surrounding environment more complex. Therefore, the abnormal pore number distribution characteristics are not only a direct reflection of the complexity of the microstructure, but also the response of seeds to adapt to environmental pressure or other biological factors, further emphasizing the complex state of the seed surface microstructure.

[0102] Under the detection window, the specific steps for analyzing the distribution characteristics of the pore quantity on the surface and generating the porosity index are as follows:

[0103] In the detection window, the acquired image information is preprocessed, and the image processing algorithm is used to identify pores on the preprocessed image information. The expression is as follows:

[0104]

[0105] , where A p is the total pore area, N p is the number of pores, A y is the area of ​​the yth pore;

[0106] The calculation formula for porosity is:

[0107]

[0108] , where P is the porosity, A is t is the total area of ​​the observation region;

[0109] On the basis of porosity, the pore shape factor is introduced and the porosity is corrected by its influence. The calculation expression is as follows:

[0110]

[0111] , where S is the shape factor, N p is the number of pores, L y is the perimeter of the yth aperture, A y is the area of ​​the yth pore;

[0112] The shape factor of pores plays a key role in the calculation of porosity index because it not only reflects the geometric characteristics of the pores, but also affects the permeability of the pores to water, air and nutrients. The shape factor is usually expressed by the ratio of the pore perimeter to the area. The more complex the shape of the pore, the larger the surface area to volume ratio, which helps to increase the contact area between the seed and the soil and improve its adhesion and stability. In addition, pores with complex shapes can often better regulate the exchange of water and gas, thereby enhancing the adaptability and resistance of seeds during growth. Therefore, the introduction of the shape factor enables the porosity index to more comprehensively reflect the actual performance of the seeds and provide a more accurate quality assessment.

[0113] The porosity is corrected based on the influence of the shape factor S, and the calculation expression is as follows:

[0114]

[0115] , where P adj is the corrected porosity;

[0116] Finally, the porosity index is generated, and the calculation expression is as follows:

[0117]

[0118] , where PI is the porosity index, k is a normalization constant used to adjust the final index to a suitable range, 1-P adj represents the proportion of non-porous area in the detection area, and ω are tuning parameters, Used to adjust the corrected porosity P adj The influence of ω is used to adjust the influence of the non-pore area in the detection area, e -ρS is the exponential weight of the shape factor, ρ is the weight of the shape factor, e is the natural base, and δ is the uniformity adjustment parameter used to control the effect of the uniformity of pore area distribution on the index;

[0119] From the porosity index calculation expression, it can be seen that within the detection window, the larger the porosity index value generated after analyzing the distribution characteristics of the number of pores on the surface, the more complex the microstructure of the seed surface is. This is because higher porosity means that there are more pores and irregular shapes on the surface, and this diversity reflects the complexity and variability of the surface structure. Relatively speaking, if the porosity index is low, it means that the seed surface presents fewer pores and more regular shapes, showing a relatively simple microstructure.

[0120] The machine learning model is not limited here, and any machine learning model that can perform a comprehensive analysis of the texture complexity index TCI and the porosity index PI to generate a microstructure coefficient MC is acceptable. To implement the technical solution of the present invention, the present invention provides a specific implementation method:

[0121] The calculation formula for the microstructure coefficient MC is as follows:

[0122]

[0123] , where k1 and k2 are the preset proportional coefficients of the texture complexity index TCI and the porosity index PI, respectively, and both k1 and k2 are greater than 0.

[0124] It can be seen from the calculation expression of the microstructure coefficient that, under the detection window, the larger the texture complexity index expression value generated after analyzing the diversity change degree of the surface texture, the larger the porosity index expression value generated after analyzing the distribution characteristics of the number of pores on the surface, that is, the larger the microstructure coefficient expression value generated under the detection window, indicating that the surface microstructure of the seed in this sub-region is in a complex state, otherwise it indicates that the surface microstructure of the seed in this sub-region is relatively simple.

[0125] Based on the evaluation results of the machine learning model, the surface structure of pepper seeds is divided into complex surface structure and simple surface structure;

[0126] The microstructure coefficient generated after analyzing the seed surface structure information in the sub-region under the detection window is compared and analyzed with a preset microstructure coefficient reference threshold, and the seed surface structure state in the sub-region is divided;

[0127] If the microstructure coefficient is greater than or equal to a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a complex surface structure;

[0128] If the microstructure coefficient is less than a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a simple surface structure;

[0129] Complex surface structure refers to the presence of diverse and irregular morphological features on the surface of pepper seeds, including pores of various sizes and shapes, uneven textures, and complex microstructures, while simple surface structure is characterized by relatively uniform and regular surface features, usually containing fewer pores and relatively smooth textures;

[0130] For the simple surface structures identified, sub-areas are randomly selected for data collection based on the preset number of detection windows;

[0131] This process can effectively reduce the interference to the complex structure, while ensuring the continuity and stability of the screening process. Through continuous collection, the seed screening strategy can be further verified and optimized. This process can effectively reduce the interference to the complex structure, while ensuring the continuity and stability of the screening process. Through continuous collection, the seed screening strategy can be further verified and optimized.

[0132] For complex surface structures, based on the preset number of detection windows, multi-scale image analysis is used to dynamically adjust the number and position of detection windows according to the evaluation results to comprehensively capture the complex microstructure features;

[0133] For complex surface structures, based on the preset number of detection windows and the evaluation results, the number and position of detection windows are dynamically adjusted to fully capture the complex microstructure features. The specific steps are as follows:

[0134] According to the evaluation results, i.e., the complex surface structure, the number of detection windows is dynamically adjusted. Considering the influence of the changes of the texture complexity index TCI and the porosity index PI on the number of detection windows, a weighted adjustment method is adopted. The calculation expression is as follows:

[0135]

[0136] , where N adjusted is the number of detection windows after adjustment, N initial is the preset number of detection windows, TCI rt is the preset texture complexity index reference threshold, PI rt is the preset porosity index reference threshold, α and β are adjustment factors, which reflect the influence of texture complexity and porosity change on the number of detection windows, respectively;

[0137] After adjusting the number of windows, the position of the new window is determined. Based on the distribution of complex surface features, the cluster analysis method is used to identify areas with large changes in surface features. The expression is as follows:

[0138] P selected =Top(N adjusted ,{P r |r∈rank(C r )})

[0139] , where P selected is the selected detection window position set, P r is the position of the candidate detection window, the rth position in the position set of all candidate detection windows, P r = {P1, P2, P3, ..., P u}, u is the total number of candidate detection windows, C r is the candidate detection window P rThe complexity evaluation is the rth candidate detection window position P r The complexity index, rank(C r ) is the order of complexity evaluation;

[0140] Select the N most complex candidate detection window positions. adjusted (i.e. the number of detection windows after adjustment) positions, forming P selected This method ensures that when resources are limited, the most complex and representative surface areas can be detected first, thereby fully capturing the complex microstructural features of the seed surface and improving the accuracy of the analysis.

[0141] Based on the feedback of complex surface features, the position of the window is evaluated again. If some positions fail to fully cover the features, the position can be adjusted. The calculation expression is as follows:

[0142] P revised =P selected +ΔP

[0143] , where P revised is the adjusted detection window position, ΔP is the position offset, which is used to adjust the window position after feedback;

[0144] Finally, loop optimization is performed based on the new number and position of windows. The calculation expression is as follows:

[0145]

[0146] , where N final is the number of detection windows after final adjustment, γ is the feedback adjustment factor, which controls the sensitivity of the adjustment amplitude, that is, the degree of adjustment of the number of windows when the surface complexity changes greatly, C new is the current complexity index, C old It is the complexity index of the last evaluation, the result of a previous complexity analysis, and serves as a benchmark for complexity changes.

[0147] The present invention can intelligently analyze the surface microstructure characteristics of pepper seeds by introducing texture complexity index and porosity index and inputting them into a pre-trained machine learning model. This method allows the screening process to not only focus on the basic morphology of the seed surface, but also accurately identify different microstructure types and their complexity. This automated evaluation process avoids the limitations of traditional random sampling methods and reduces the risk of high-quality seeds being misjudged as low-quality seeds because the complex microstructure is not fully captured. At the same time, by dynamically adjusting the number and position of the detection windows, the system can be flexibly adjusted according to the complexity of the actual microstructure to ensure that the complex microstructure characteristics are fully captured, thereby improving the accuracy and reliability of the screening process.

[0148] The present invention intelligently divides the complexity of the seed surface structure, and divides the surface microstructure into "complex surface structure" and "simple surface structure", so that suitable collection and detection resources can be allocated for different seed characteristics. For simple surface structures, random collection is only required based on a fixed number of detection windows, saving unnecessary resource investment; while for complex surface structures, by dynamically adjusting the number and position of the collected detection windows, the seed characteristics can be analyzed more comprehensively to achieve efficient use of resources. This method significantly improves the overall detection efficiency, which not only reduces time and cost consumption in the screening process, but also ensures that resources can be concentrated on the identification of complex surface structures, ultimately providing a high-quality, cost-effective and efficient screening solution for the breeding process.

[0149] The above only describes some exemplary embodiments of the present invention in a general manner. It is undoubted that, for those skilled in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A multidimensional data analysis method for breeding screening, characterized in that: The following steps are involved: Spreading the pepper seeds of the batch to be tested in the testing area, and dividing the testing area into multiple sub-areas; In the divided sub-areas, sub-areas are randomly selected based on the preset number of detection windows for data collection to obtain the seed surface structure information; Extract the seed surface structure information in the sub-region. Based on the extracted seed surface microstructure features, use the pre-learned machine learning model to intelligently evaluate the complexity of the seed surface structure in the sub-region under the detection window to identify different surface structure types of the seeds. Based on the evaluation results of the machine learning model, the surface structure of pepper seeds is divided into complex surface structure and simple surface structure; For the simple surface structures identified, sub-areas are randomly selected for data collection based on the preset number of detection windows; For complex surface structures, based on the preset number of detection windows and the evaluation results, the number and position of detection windows are dynamically adjusted to comprehensively capture the complex microstructure features.

2. The multidimensional data analysis method for breeding screening according to claim 1, characterized in that: Feature extraction is performed on the seed surface structure information in the sub-area. The extracted features include the diversity and variation degree of the surface texture and the distribution characteristics of the number of pores on the surface. Under the monitoring window, the diversity and variation degree of the extracted surface texture and the distribution characteristics of the number of pores on the surface are analyzed, and then the texture complexity index and porosity index are generated respectively. The texture complexity index is used to measure the diversity and irregularity of the seed surface texture, and the porosity index is used to measure the number, size and distribution characteristics of the pores on the seed surface. The texture complexity index and porosity index are input into a pre-learned machine learning model, and a microstructure coefficient is generated through the machine learning model. The complexity of the seed surface structure is intelligently evaluated through the microstructure coefficient to identify different surface structure types of seeds.

3. A multidimensional data analysis method for breeding screening according to claim 2, characterized in that: Under the detection window, the diversity of surface texture is analyzed and the specific steps of generating texture complexity index are as follows: Under the detection window, the collected seed surface data is preprocessed, and the texture features are extracted using the gray level co-occurrence matrix technology for the preprocessed data. The calculation expression is as follows: ， Where G(i,j) is an element of the gray-level co-occurrence matrix, which indicates the frequency of adjacent appearance of pixels with gray-level values ​​i and j in the image, and Q(i,j|x,y) indicates the conditional probability of the occurrence of a pair of pixels with gray-level values ​​i and j at position (x,y); The texture energy and texture contrast are calculated using the extracted gray level co-occurrence matrix. The texture energy calculation expression is as follows: ， Where E is the texture energy, which measures the uniformity of the texture; The texture contrast calculation expression is as follows: ， Where C is the texture contrast, which describes the degree of texture change; Calculate the entropy value of the texture and further evaluate the complexity of the texture. The calculation expression is as follows: ， Where H is the texture entropy value, ∈ is a small constant used to avoid the zero value problem in logarithmic calculation; Combining texture energy E, texture contrast C and texture entropy value H, the texture complexity index is generated. The calculation expression is as follows: ， Where TCI is the texture complexity index, max(E) is the maximum texture energy, max(C) is the maximum texture contrast, max(H) is the maximum texture entropy, A is the adjustment parameter of texture energy E, B is the adjustment parameter of texture contrast c, F is the power adjustment parameter of texture contrast, δ is the offset adjustment parameter of texture entropy value H, λ is the power adjustment parameter of texture entropy value H, and η is a non-zero positive number used to control the smoothness of the final result.

4. A multidimensional data analysis method for breeding screening according to claim 2, characterized in that: Under the detection window, the specific steps for analyzing the distribution characteristics of the pore quantity on the surface and generating the porosity index are as follows: In the detection window, the acquired image information is preprocessed, and the image processing algorithm is used to identify pores on the preprocessed image information. The expression is as follows: ， In the formula, A p is the total pore area, N p is the number of pores, A y is the area of ​​the yth pore; The calculation formula of porosity is: ， Where P is the porosity, A is t is the total area of ​​the observation region; On the basis of porosity, the pore shape factor is introduced and the porosity is corrected by its influence. The calculation expression is as follows: ， Where S is the shape factor, N p is the number of pores, L y is the perimeter of the yth aperture, A y is the area of ​​the yth pore; The porosity is corrected based on the influence of the shape factor S, and the calculation expression is as follows: ， Where P adj is the corrected porosity; Finally, the porosity index is generated, and the calculation expression is as follows: ， Where PI is the porosity index, k is the normalization constant, and 1-P adj represents the proportion of non-porous area in the detection area, and ω are tuning parameters, Used to adjust the corrected porosity P adj The influence of ω is used to adjust the influence of the non-pore area in the detection area, e -ρS is the exponential weight of the shape factor, ρ is the weight of the shape factor, e is the natural base, and δ is the uniformity adjustment parameter.

5. The multidimensional data analysis method for breeding screening according to claim 2, characterized in that: The microstructure coefficient generated after analyzing the seed surface structure information in the sub-region under the detection window is compared and analyzed with a preset microstructure coefficient reference threshold, and the seed surface structure state in the sub-region is divided; If the microstructure coefficient is greater than or equal to a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a complex surface structure; If the microstructure coefficient is less than a preset microstructure coefficient reference threshold, the surface structure state of the seed in the sub-region is classified as a simple surface structure.

6. A multidimensional data analysis method for breeding screening according to claim 5, characterized in that: For complex surface structures, based on the preset number of detection windows and the evaluation results, the number and position of detection windows are dynamically adjusted to fully capture the complex microstructure features. The specific steps are as follows: According to the evaluation results, i.e., the complex surface structure, the number of detection windows is dynamically adjusted. Considering the influence of the changes of the texture complexity index TCI and the porosity index PI on the number of detection windows, a weighted adjustment method is adopted. The calculation expression is as follows: ， Where N adjusted is the number of detection windows after adjustment, N initial is the preset number of detection windows, TCI rt is the preset texture complexity index reference threshold, PI rt is the preset porosity index reference threshold, α and β are adjustment factors, reflecting the effects of texture complexity and porosity changes on the number of detection windows, respectively; After adjusting the number of windows, the position of the new window is determined. Based on the distribution of complex surface features, the cluster analysis method is used to identify areas with large changes in surface features. The expression is as follows: P selected =Top(N adjusted ,{P r |r∈rank(C r )}), Where P selected is the selected detection window position set, P r is the position of the candidate detection window, the rth position in the position set of all candidate detection windows, P r = {P1, P2, P3, ..., P u }, u is the total number of candidate detection windows, C r is the candidate detection window P r The complexity evaluation is the rth candidate detection window position P r The complexity index, rank(C r ) is the order of complexity evaluation; Based on the feedback of complex surface features, the position of the window is evaluated again. If some positions fail to fully cover the features, the position can be adjusted. The calculation expression is as follows: P revised =P revised +ΔP, Where P revised is the adjusted detection window position, ΔP is the position offset, which is used to adjust the window position after feedback; Finally, loop optimization is performed based on the new number and position of windows. The calculation expression is as follows: ， Where N final is the number of detection windows after final adjustment, γ is the feedback adjustment factor, C new is the current complexity index, C old is the complexity index of the last evaluation.

Images