Method and system for screening and determining epimedium seed maturity degree image

Abstract

The application discloses a method and system for screening and determining the maturity of Epimedium seed, and relates to the technical field of seed image recognition; the method comprises the following steps: collecting general basic data for calculating correction parameters within a set sample collection period; calculating correction parameters for appearance feature correction based on the general basic data; sampling and collecting the appearance features of the seeds in the batch to be screened within the sampling time, adjusting the correction parameters, and obtaining adjusted correction parameters; screening the seeds to be screened based on the adjusted correction parameters, and obtaining a maturity classification list of the seeds; the accuracy of the screening result is improved, the quality and efficiency of subsequent seed breeding, storage and seedling raising are guaranteed, and the waste of misclassified high-quality seeds and the mixing of unqualified seeds to affect the cultivation effect are avoided.

Description

Technical Field

[0001] This invention relates to the field of seed image recognition technology, and more specifically, to a method and system for screening and determining the maturity of Epimedium seeds from images. Background Technology

[0002] In actual screening operations of Epimedium seeds, there is often a gap of several hours between the time the seeds are harvested and when they enter the screening system, during which time they need to undergo temporary storage and transportation. The core principle of the currently used appearance screening technology is to collect the appearance characteristics of the seeds after harvesting, such as color, size, luster and texture, to construct a correlation model between these appearance characteristics and maturity, and then judge and classify the maturity of the seeds based on the correlation model.

[0003] However, this technology does not take into account the physiological dynamic changes that occur in Epimedium seeds in the short term after harvest. After harvest, the seeds will undergo processes such as after-ripening or dehydration stress. Immature seeds may shrink due to rapid water loss, making them appear plump, and their color may also darken due to dehydration, similar to the characteristics of mature seeds. If fully mature seeds are exposed to moisture after harvest, the seed coat may absorb water and swell, becoming bloated and appearing overripe and softened.

[0004] Because existing technologies use training data based solely on morphological characteristics measured immediately after harvest, without incorporating dynamic morphological data during the post-harvest physiologically active period, the constructed models are unable to identify these pseudo-maturity characteristics caused by physiological changes. This directly leads to the system misclassifying seeds exhibiting pseudo-maturity characteristics due to post-harvest physiological changes during actual screening, resulting in a significant decrease in the accuracy of screening results. This affects the quality and efficiency of subsequent seed breeding, storage, and seedling cultivation, potentially leading to the misclassification and waste of high-quality seeds, or the mixing of substandard seeds that negatively impacts cultivation outcomes.

[0005] In view of this, the present invention proposes a method and system for screening and determining the maturity of Epimedium seeds by image screening to solve the above problems. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a method for screening and determining the maturity of Epimedium seeds using images, comprising:

[0007] Within the set sample collection period, general basic data for calculating correction parameters are collected; the general basic data includes the appearance characteristics, physiological indicators, and seed collection information of seeds collected at each set time point within the sample collection period.

[0008] Based on general basic data, correction parameters for appearance feature correction are calculated; these correction parameters include a color change rate benchmark, a physiological-appearance correlation coefficient matrix, and a color feature correction factor; the method for obtaining the correction parameters for appearance feature correction includes:

[0009] The average appearance characteristics of seeds of the same variety and different batches at the same time point were calculated. The average rate of change of color characteristics was obtained by the ratio of the difference between adjacent time points to the time interval. The Pearson correlation coefficients between each appearance characteristic and each physiological index were calculated.

[0010] The arithmetic mean of the average rate of change of color characteristics of different batches of seeds at the same time interval is taken to obtain the benchmark of the rate of change of color of seeds of the corresponding variety type at the corresponding time interval.

[0011] By using appearance features as rows and physiological indicators as columns, the corresponding elements of the matrix are the Pearson correlation coefficients between the corresponding appearance features and physiological indicators, thus obtaining the physiological-appearance correlation coefficient matrix.

[0012] Based on the time interval of each hour node within the sample collection period, the color change rate benchmark of the corresponding time interval is called to calculate the cumulative color feature change from 0 hours after collection to each hour node after collection; the cumulative color feature change is multiplied by the corresponding Pearson correlation coefficient and summed to obtain the weighted cumulative color feature change; the weighted cumulative color feature change is divided by the corresponding time node value to obtain the color feature correction factor for each hour node.

[0013] The appearance characteristics of the batch of seeds to be screened are sampled within a set sampling time; and the correction parameters are adjusted based on the appearance characteristics collected within the set sampling time to obtain the adjusted correction parameters.

[0014] Based on the adjusted correction parameters, the seeds to be screened are screened by maturity to obtain a list of seed maturity classifications.

[0015] Furthermore, the appearance features include color features, shape features, and texture features; the color feature is the H value of the seed in the HSV space; the shape feature is the aspect ratio of the seed; and the texture feature is the energy of the seed calculated through the gray-level co-occurrence matrix.

[0016] The physiological indicators include seed embryo vigor, seed coat chlorophyll content, and endosperm soluble sugar content.

[0017] The seed collection information includes seed variety type, harvesting site and time, and the interval between harvest and screening.

[0018] Furthermore, the methods for setting the sample sampling period and the time nodes within the sample sampling period include:

[0019] Calculate the ratio of the current appearance characteristics of the seed to the time interval from seed harvest to the present to obtain the change in appearance characteristics per unit time;

[0020] The sample sampling period is set to fully cover the post-harvest physiologically active period of seeds. The physiologically active period is the fluctuation stage where the change in seed appearance characteristics per unit time exceeds the preset threshold for change per unit time corresponding to the variety type, and the stable stage where the change in seed appearance characteristics per unit time is lower than the preset benchmark threshold for change per unit time for multiple consecutive time nodes.

[0021] The density of time nodes within the sample sampling period is adjusted according to the change in seed appearance characteristics per unit time. The interval between time nodes in the fluctuation phase is set to be less than a preset proportion of the average change period of the fluctuation phase; the interval between time nodes in the stable phase is set to a preset multiple of the interval of the fluctuation phase.

[0022] Furthermore, the method of sampling the appearance characteristics of the batch of seeds to be screened within the sampling time and adjusting the correction parameters to obtain the adjusted correction parameters includes:

[0023] A predetermined number of samples were randomly selected from the batch of seeds to be screened. At the time point of 0 hours after harvest, the appearance characteristics and physiological indicators of the samples were collected on site.

[0024] Calculate the rate of change of color features and the temporary correlation coefficient matrix between appearance features and physiological indicators of the sample within a set time interval;

[0025] The difference between the color feature change rate within a set time interval and the color change rate reference for the corresponding time interval is divided by the color change rate reference to obtain the deviation ratio; the corrected color change rate reference = color change rate reference × (1 + deviation ratio).

[0026] The temporary correlation coefficient matrix is ​​compared with the general physiological-appearance correlation coefficient matrix. Coefficients with differences exceeding a preset difference threshold are replaced. Based on the corrected rate of change benchmark and the replaced physiological-appearance correlation coefficient matrix, the feature correction factor for the corresponding time interval is recalculated to obtain the adjusted correction parameters.

[0027] Furthermore, based on the adjusted correction parameters, the method for screening the seeds by maturity to obtain a list of seed maturity classifications includes:

[0028] For color features, the current color feature value is corrected by a color feature correction factor to obtain the color feature 0 hours after sampling;

[0029] For shape and texture features, the aspect ratio and energy are corrected according to the feature change pattern corresponding to the postharvest interval, based on the physiological-appearance correlation coefficient matrix in the modified physiological calibration model, to obtain the shape and texture features at 0 hours postharvest.

[0030] The corrected equivalent appearance characteristics at 0 hours post-harvest are input into the preset classifier. After judging each seed in the entire batch, the final seed maturity classification list is obtained.

[0031] Methods for obtaining color characteristics at 0 hours post-harvest include:

[0032] Color feature value 0 hours after harvest = current color feature value - (color feature correction factor × time interval from harvest to screening).

[0033] Furthermore, methods for obtaining shape and texture features at 0 hours post-harvest include:

[0034] The difference in shape and texture features between different batches of seeds of the same variety at adjacent time points is divided by the time interval to obtain the rate of change of shape and texture features of a single batch of seeds at that interval; the arithmetic mean of the rate of change of shape and texture features of multiple batches of seeds at the same time interval is taken as the average rate of change of shape and texture features of the same variety at the corresponding time interval.

[0035] Based on the time interval from harvest to screening of the seeds to be corrected, determine the time interval to which the seeds belong, and obtain the average rate of change of shape and texture features of the corresponding time interval. The cumulative change of shape and texture features = average rate of change of shape and texture features × time interval from harvest to screening.

[0036] Extract the Pearson correlation coefficients between shape and texture features and corresponding physiological indicators from the physiological-appearance correlation coefficient matrix; use the Pearson correlation coefficients as weights to perform weighted correction on the cumulative changes of shape and texture features to obtain the weighted correction of the cumulative changes of shape and texture features.

[0037] Equivalent shape and texture feature value after 0 hours = current shape and texture feature value - (cumulative change in weighted and corrected shape and texture features).

[0038] Furthermore, the classification methods of the preset classifier include:

[0039] Based on the physiological-appearance correlation coefficient matrix, strongly correlated appearance features and their corresponding threshold ranges are determined. The strongly correlated appearance feature refers to the appearance feature corresponding to the coefficient with the largest absolute value among the Pearson correlation coefficients of various appearance features corresponding to a certain physiological index in the physiological-appearance correlation coefficient matrix. The equivalent postharvest 0-hour threshold range of mature seeds corresponding to each strongly correlated appearance feature is determined based on the numerical distribution of the equivalent postharvest 0-hour strongly correlated appearance features of mature seeds in the general basic data.

[0040] An image screening and determination system for the maturity of Epimedium seeds, including:

[0041] The sample acquisition module is used to collect general basic data for calculating correction parameters within a set sample acquisition period.

[0042] The parameter calculation module calculates correction parameters for appearance feature correction based on general basic data.

[0043] The parameter adjustment module is used to sample and collect the appearance characteristics of the batch of seeds to be screened within a set sampling time; and to adjust the correction parameters based on the appearance characteristics collected within the set sampling time to obtain the adjusted correction parameters.

[0044] The classification and screening module, based on the adjusted correction parameters, filters the seeds to be screened by maturity and obtains a list of seed maturity classifications.

[0045] Compared with the prior art, the technical effects and advantages of the Epimedium seed maturity image screening and determination method and system of the present invention are as follows:

[0046] This invention sets a sample collection cycle covering the physiologically active period of seeds after harvest, collects appearance characteristics, physiological indicators and collection information at multiple time points, and calculates correction parameters including a color change rate benchmark, a physiological-appearance correlation coefficient matrix and a color feature correction factor. The correction parameters are adjusted by sampling to adapt to new batches of seeds, and then the color, shape and texture features are corrected to the state at 0 hours after harvest based on the adjusted parameters. Finally, maturity screening is completed based on strong correlation features and thresholds.

[0047] This invention solves the problems of misjudging false maturity characteristics due to failure to consider short-term physiological dynamic changes after harvest, low screening accuracy for seeds with different intervals, and high error rate in cross-variety screening. It improves the accuracy of screening results, ensures the quality and efficiency of subsequent seed breeding, storage and seedling raising, and avoids the waste of high-quality seeds due to misclassification and the mixing of unqualified seeds, which affects the cultivation effect. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the Epimedium seed maturity image screening and determination system according to an embodiment of the present invention;

[0049] Figure 2 This is a flowchart of the method for screening and determining the maturity of Epimedium seeds according to an embodiment of the present invention;

[0050] Figure 3 This is a flowchart illustrating the method for setting the sample sampling period and each time node within the sample sampling period according to an embodiment of the present invention.

[0051] Figure 4 This is a flowchart illustrating a method for calculating correction parameters for appearance feature modification according to an embodiment of the present invention.

[0052] Figure 5 This is a schematic diagram of appearance feature acquisition according to an embodiment of the present invention. Detailed Implementation

[0053] To further clarify the technical problem to be solved by this application and its background, before proceeding with specific implementation methods, the relevant reaction mechanism, the limitations of the prior art, and the practical difficulties faced by those skilled in the art in solving the problem will be explained in detail.

[0054] Specifically, although the ripening process of Epimedium seeds ceases after harvest, physiological stress responses are triggered short-term by changes in environmental temperature and humidity. For example, immature seeds have thinner seed coats and higher cell water content. Within 0-24 hours after harvest, they are prone to shrinkage due to water loss, resulting in a lighter surface texture. Simultaneously, the chlorophyll in the seed coat rapidly degrades, causing the color to change from light green to dark brown, resembling the appearance of mature seeds. Mature seeds have a stable seed coat structure and low water content, but if exposed to moisture after harvest, the seed coat will absorb water and swell, resulting in an abnormally high length-to-width ratio and even softening, appearing as overripe. Furthermore, the soluble sugar content in the endosperm shows a short-term increase in immature seeds within 0-12 hours after harvest due to cell rupture, while remaining stable in mature seeds. This fluctuation in physiological indicators further exacerbates the deviation between appearance characteristics and true maturity.

[0055] Existing appearance screening technologies, which rely on static appearance characteristics immediately after harvest, have significant limitations. For example, they cannot distinguish between true maturity and pseudo-maturity. Immature seeds, due to water loss, darken in color and increase in aspect ratio, closely resemble the natural appearance of mature seeds. Static models struggle to identify these pseudo-features caused by physiological changes, leading to a high misclassification rate. In actual production, the time interval between seed harvesting and screening varies. Existing models are trained only on data from 0 hours after harvest. For seeds with longer intervals, the screening accuracy drops significantly due to dynamic changes in appearance. Furthermore, different Epimedium varieties exhibit significant differences in post-harvest physiological rhythms. Static models lack variety-specific parameters, increasing the error rate when screening across varieties.

[0056] To address the aforementioned issues, dynamic feature acquisition over a long sample collection period is a key design consideration for balancing various needs. For example, Epimedium seeds undergo a rapid physiological activity period after harvest. Within 0-24 hours, immature seeds experience significant changes in appearance characteristics such as seed coat color and aspect ratio due to water loss, easily generating "false maturity" signals. Mature seeds, however, tend to stabilize their appearance characteristics within 24-48 hours. Setting a 48-hour sample collection period allows for the complete capture of the entire process of appearance characteristics changing and stabilizing after harvesting. This ensures that the dynamic patterns learned by the model cover all possible appearance fluctuations from harvest to screening, avoiding the omission of crucial data during the later characteristic stabilization stage due to an excessively short sample collection period. Furthermore, in actual production, the time interval between seed harvesting and entry into the screening system fluctuates, potentially being delayed by 12-48 hours due to transportation, temporary storage, and other factors. If a 48-hour sample collection period is set, the dynamic feature library can support the system in performing feature correction on seeds at any interval. If seeds are screened 36 hours after harvest, the system can call upon the feature change patterns of the 36-hour node to accurately restore the current appearance characteristics to the true state at 0 hours post-harvest. However, if the sample collection period is shorter than 48 hours, it cannot cover screening scenarios exceeding the corresponding time, leading to correction failure. Finally, different Epimedium varieties have different post-harvest physiological rhythms, and some varieties may experience secondary appearance changes around 36 hours. A 48-hour sample collection period can fully encompass these variety-specific fluctuations, ensuring that the model can accurately adapt to the dynamic characteristics of seeds from multiple varieties. If the sample collection period is too short, it may miss key change nodes for specific varieties, leading to correction biases when screening across varieties. In summary, dynamic feature collection with a long sample collection period provides full-cycle data support for subsequent feature correction and model training, which is a core prerequisite for ensuring screening accuracy.

[0057] In addressing the aforementioned issues, those skilled in the art face multiple technical challenges. If the sample collection period is too short, it is impossible to capture the stable characteristics of mature seeds 24 hours later and the later changes in immature seeds; if the sample collection period is too long, it will increase data collection costs, and after more than 48 hours, seeds may become moldy, leading to distorted characteristics, making it difficult to balance integrity and practicality; the correlation between appearance characteristics and physiological indicators changes dynamically with the time after harvest, and traditional static correlation models cannot adapt to this dynamic relationship, making it difficult to establish accurate mappings.

[0058] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0059] Example 1

[0060] Please see Figure 1 As shown in the figure, this embodiment discloses an image screening and judgment system for the maturity of Epimedium seeds, including a sample collection module, a parameter calculation module, a parameter adjustment module, and a classification and screening module. Each module is connected by wired and / or wireless means to realize data transmission.

[0061] The sample acquisition module is used to collect general basic data for calculating correction parameters within a set sample acquisition period.

[0062] The general basic data includes the appearance characteristics and physiological indicators of seeds collected at each set time point within the sample collection period, as well as seed collection information; the appearance characteristics include color characteristics, shape characteristics, and texture characteristics; the physiological indicators include seed embryo viability, seed coat chlorophyll content, and endosperm soluble sugar content; the seed collection information includes seed variety type, harvesting plot and time, and the interval between harvest and screening.

[0063] Please see Figure 3 As shown, the methods for setting the sample sampling period and the time nodes within the sample sampling period include:

[0064] The ratio of the current appearance characteristics of the seed to the time interval from seed harvest to the present is calculated to obtain the change in appearance characteristics per unit time. The sample sampling period is set to fully cover the physiologically active period after seed harvest. The physiologically active period is the fluctuation stage where the change in seed appearance characteristics per unit time exceeds the preset threshold for change per unit time corresponding to the variety type, and the stabilization stage where the change in seed appearance characteristics per unit time is lower than the preset benchmark threshold for change per unit time for multiple consecutive time nodes. The time node density within the sample sampling period is adjusted according to the change in seed appearance characteristics per unit time. The interval length of the time nodes in the fluctuation stage is set to be less than a preset proportion of the average change period of the fluctuation stage. The interval length of the time nodes in the stabilization stage is set to a preset multiple of the interval of the fluctuation stage.

[0065] The preset threshold for change per unit time and the preset baseline threshold for change per unit time need to be determined based on historical data of the change in appearance characteristics per unit time of multiple batches of seeds of the same variety during the postharvest physiological active period. By analyzing the distribution of appearance characteristic changes in the fluctuating and stable phases in these data, the critical value that can distinguish the two phases is selected as the corresponding threshold. The preset proportion of changes less than the average change period of the fluctuating phase needs to be calculated first, which is the average change period of the fluctuating phase of the corresponding variety type of seeds, that is, the average time interval between the change in appearance characteristics per unit time exceeding the preset threshold for change per unit time of the variety type. Then, combined with the requirement for accuracy in capturing dynamic changes, the proportion that can completely reflect the fluctuation rhythm in historical data is set. The preset multiple of the interval of the fluctuating phase needs to be set based on the characteristic that the change in appearance characteristics per unit time in the stable phase is more gradual than that in the fluctuating phase. Combined with the difference in the rate of change of appearance characteristics in the fluctuating and stable phases in historical data, the multiple that can reduce the node density while ensuring the integrity of the stable phase data is set. All preset values ​​need to be adapted to the postharvest physiological characteristics of the variety type and determined by calibration through the change patterns in the general basic data.

[0066] By setting a sampling cycle covering the post-harvest physiologically active period of seeds, and collecting appearance characteristics, physiological indicators, and collection information at multiple time points to obtain general basic data, this approach can fully capture the dynamic patterns of seeds from the fluctuating stage of significant changes in appearance characteristics to the stable stage of gradual changes after harvest, rather than relying solely on static data immediately after harvest. By integrating the characteristics and physiological indicators of different time points, it can systematically record the fluctuation patterns of appearance characteristics caused by post-harvest water loss or moisture absorption, providing full-cycle dynamic data support for subsequent calculation and correction parameters. To address the misjudgment of false maturity characteristics caused by neglecting short-term post-harvest physiological dynamic changes, the dynamically collected general basic data can distinguish between true maturity characteristics and false characteristics caused by physiological changes, providing a basis for the fluctuation patterns of correction parameters. Regarding the issue of low seed screening accuracy at different intervals, the basic data covering the entire cycle can support feature correction for seeds at any post-harvest interval, ensuring restoration to the true state at 0 hours post-harvest. To address the issue of high cross-variety screening error rates, by collecting dynamic characteristics of different varieties, it can capture variety-specific physiological rhythms, making correction parameters adaptable to multiple varieties and reducing cross-variety screening errors.

[0067] For example, in this embodiment, the sample collection period is set to 48 hours, and the time nodes are set to five time nodes: 0 hours, 6 hours, 12 hours, 24 hours, and 48 hours post-collection. Three to five batches of seeds are collected in advance. For each batch, at each time node, in an environment with stable temperature and humidity, the appearance characteristics of the seeds are collected using an industrial camera with fixed parameters and a light source. Simultaneously, the physiological indicators of the corresponding seeds are recorded at the five time nodes. The appearance characteristics and physiological indicators collected at the five time nodes are integrated to obtain general basic data. Please refer to [link / reference]. Figure 5 As shown, the specific method for collecting appearance features includes: a vibratory feeder uses high-frequency vibration to neatly sort seeds according to their shape, ensuring that the radicles face the same direction; an optical sensor is installed at the outlet to detect the seed spacing; a visual positioning system captures real-time images of the seeds at the vibratory feeder outlet, extracts contour features, calculates the center of gravity coordinates, and generates a robotic arm grasping path; after the robotic arm picks up the seeds according to the path, it accurately places the seeds into the corresponding holes on the carrier plate according to a preset coordinate matrix; each seed placed triggers the system to record the carrier plate number, hole coordinates, and feeding time; the system automatically integrates the data to generate a seed-carrier plate number-hole coordinate-feeding time binding table, which is synchronously stored in the database. When the set time node is reached, the system drives the conveyor belt to transport the carrier plate to the shooting station; the positioning holes on the edge of the carrier plate are aligned with the station sensor and automatically limited; an industrial camera captures an image of the entire plate from a vertical direction; after image preprocessing, the software segments the individual seed areas according to the hole coordinates; appearance features are extracted from each area, and the binding database is called in conjunction with the carrier plate number to match the feature data with the corresponding seed ID.

[0068] The color feature includes the H value of the seed in the HSV color space; the HSV color space is a standardized color space used to describe color, quantifying color from different dimensions. The H value in the HSV color space refers to the hue of the color, ranging from 40 to 160, and is used to reflect the transition of hue from green to brown. The color feature is directly related to the color change during seed maturation; immature seeds have lower H values, while mature seeds have higher H values.

[0069] The shape characteristics include the length-to-width ratio of the seed; the length-to-width ratio is the ratio of the seed's major diameter to its minor diameter, and the value ranges from 1.2 to 3.5, which is used to reflect the elliptical shape of the seed. The length-to-width ratio is more stable in mature seeds.

[0070] The texture feature is the seed energy calculated through the gray-level co-occurrence matrix; the energy value ranges from 0.1 to 0.5, and is used to reflect the uniformity of the seed texture. The higher the energy value, the smoother the seed surface; mature seeds have high energy, while immature or overripe seeds have low energy and a rougher texture.

[0071] Among the physiological indicators, embryo viability is collected using fluorescence analysis. Seed samples are taken in their intact, undamaged natural state and placed in the detection area of ​​a fluorescence detector. The seeds are irradiated with specific wavelengths of excitation light, such as ultraviolet or blue light. The instrument captures the fluorescence signal emitted by the excited seeds and records characteristic parameters such as fluorescence intensity and wavelength distribution. Based on the correlation between fluorescence characteristics and embryo viability, embryo viability is determined. For example, highly viable seeds typically exhibit stable and strong fluorescence signals. The entire process requires no cutting or sampling of the seeds and does not damage the embryo. Seed coat chlorophyll content is collected using a portable chlorophyll meter: the instrument's detection head is gently placed against the seed coat surface for non-contact measurement of the relative chlorophyll content. Each seed is measured three times, and the average value is taken to avoid damaging the seed coat. Endosperm soluble sugar content is collected using a near-infrared spectrometer: the seeds are placed in the spectrometer's detection window and scanned within the wavelength range of 780-2526 nm. The endosperm soluble sugar content is calculated using a pre-established sugar content model. The entire process is non-destructive and does not affect the subsequent use of the seeds. The establishment of a sugar content model requires the prior collection of multiple batches of seed samples at different maturity levels. Spectral data are obtained by scanning the samples with a near-infrared spectrometer within the wavelength range of 780-2526 nm. Simultaneously, the endosperm soluble sugar content of these samples is determined using conventional detection methods as a reference value. The spectral data is then correlated with the corresponding reference values, and the spectral data is preprocessed to eliminate interference. Based on the correlation between the two, an algorithm is used to construct a mapping relationship between spectral characteristics and soluble sugar content, forming a sugar content model. Finally, the sugar content model is validated using new sample data to ensure that it can accurately calculate the endosperm soluble sugar content based on near-infrared spectral data.

[0072] Seed variety types are collected before harvesting by verifying planting records. Each batch of seeds is accompanied by a variety label upon harvest, such as Epimedium sagittatum and Epimedium pubescens. The label information is simultaneously entered into the system database to ensure accurate association with subsequent seed batches. Harvesting plots and times are recorded using handheld terminals by harvesters. Plot information includes specific plot numbers and latitude and longitude coordinates, and harvesting times are accurate to the year, month, day, hour, and minute. This data is uploaded to the system in real time and linked to the seed batch. The interval between harvesting and screening is collected through barcode scanning records when seeds enter the temporary storage stage. After harvesting, seeds are placed in containers with unique QR codes. The system automatically records the storage time when the code is scanned. Before screening begins, the code is scanned again, and the system automatically calculates the time difference between the two scans, storing this interval data in the corresponding batch file.

[0073] The parameter calculation module calculates correction parameters for appearance feature correction based on general basic data.

[0074] The correction parameters include a color change rate benchmark, a physiological-appearance correlation coefficient matrix, and a color feature correction factor.

[0075] First, the appearance characteristics and physiological indicators of 3-5 batches of seeds were preprocessed for 48 hours to remove outliers caused by collection errors, such as color values ​​that significantly deviated from the population characteristics and abnormal physiological indicators, ensuring data consistency. Then, the appearance characteristics of each batch of seeds at five time points (0 hours, 6 hours, 12 hours, 24 hours, and 48 hours) were correlated with the corresponding physiological indicators along a time axis to form a three-dimensional dataset of time, appearance, and physiology.

[0076] The average appearance characteristics of seeds of the same variety and different batches at the same time point were calculated. The average rate of change of color characteristics was obtained by the ratio of the difference between adjacent time points to the time interval, which was used to capture the regular fluctuation of color characteristics caused by seed dehydration or moisture absorption after harvest. By calculating the Pearson correlation coefficient between appearance characteristics and physiological indicators, the negative correlation coefficient between seed coat chlorophyll content and H value, the positive correlation coefficient between endosperm soluble sugar content and aspect ratio, and the positive correlation coefficient between embryo vigor and texture energy were obtained, thus clarifying the indicative weight of appearance characteristics on physiological maturity.

[0077] Seeds of the same variety exhibit a consistent rate of change in color characteristics under the same postharvest conditions. For example, the H value of immature seeds may increase at a rate of 5–8 every 6 hours due to rapid water loss within 0–24 hours, indicating a transition in seed coat color from light green to dark brown. In mature seeds, due to the stable seed coat structure, the rate of change in H value slows significantly, with a change of no more than 2 every 6 hours within 24–48 hours. By calculating the average rate of change of color characteristics, the fluctuation rhythm of different variety types can be quantified, providing a correspondence between postharvest time and the average rate of change of color characteristics for a general physiological calibration model.

[0078] When a new batch of seeds enters the screening process, the actual state at 0 hours post-harvest is calculated by reverse engineering based on the post-harvest interval of that batch and the average rate of change of color characteristics. For example, if the H value of a seed 12 hours post-harvest is 80, and the average rate of change of color characteristics of immature seeds of that variety within 12 hours is an increase of 6 every 6 hours, then the H value at 0 hours post-harvest can be corrected to 80 - (12 / 6 × 6) = 68. This eliminates the interference of post-harvest color characteristic fluctuations and ensures that maturity judgment is based on the true initial state of the seed. Therefore, the average rate of change of color characteristics is the core basis for characteristic correction and is indispensable for eliminating false maturity signals caused by post-harvest physiological changes.

[0079] Please see Figure 4 As shown, the methods for calculating correction parameters for appearance feature correction based on general basic data include:

[0080] The color change rate benchmark is the average rate of color characteristic change of seeds of the same variety within a specific time interval. The calculation method is as follows: For 3-5 batches of seeds of the same variety, extract the H-value change of each batch within four time intervals: 0-6 hours, 6-12 hours, 12-24 hours, and 24-48 hours. The change at each interval is calculated by subtracting the color characteristic value of the previous time interval from the color characteristic value at the next time point, and then dividing by the interval length (e.g., 6 hours, 12 hours, etc.) to obtain the color change rate of a single batch of seeds within that interval. The arithmetic mean of the rates for the same time interval across 3-5 batches of seeds is taken as the color change rate benchmark for that variety within that time interval, such as the average H-value change rate from 0 to 6 hours.

[0081] The physiological-appearance correlation coefficient matrix reflects the degree of linear correlation between appearance features and physiological indicators. It is calculated by: calculating the Pearson correlation coefficient between each appearance feature and each physiological indicator; then, with appearance features as rows and physiological indicators as columns, the corresponding elements of the matrix represent the Pearson correlation coefficients between the respective appearance features and physiological indicators, thus obtaining the physiological-appearance correlation coefficient matrix.

[0082] The color feature correction factor is used to reverse-correct the appearance feature value at any post-harvest time to the appearance feature value at 0 hours post-harvest. The calculation method is as follows: based on the color change rate benchmark and the physiological-appearance correlation coefficient matrix, the deviation correction coefficient between the appearance feature of each hour node within 0 to 48 hours and the feature at 0 hours post-harvest is calculated. Specifically, firstly, according to the time interval of each hour node within 0 to 48 hours, the color change rate benchmark of the corresponding interval is called to calculate the cumulative color feature change from 0 hours to the corresponding node. For example, the 10th hour node is in the time interval of 6 to 12 hours. The cumulative color feature change is multiplied by the corresponding Pearson correlation coefficient, and the results are summed to obtain the weighted cumulative change. The weighted cumulative change is divided by the corresponding time node to obtain the color feature correction factor for each hour node.

[0083] The core purpose of the correction parameters is to provide dynamic correction and prediction basis for seed maturity screening, which is reflected in two aspects: First, by using the color change rate benchmark, physiological-appearance correlation coefficient matrix and color feature correction factor, the appearance characteristics of seeds at any time after harvest are reversed to the true state at 0 hours after harvest, eliminating false maturity signals caused by post-harvest physiological changes such as water loss and dampness; Second, as a basic template, it provides common parameters for the same variety type for new batches of seeds, reducing the cost of repeatedly collecting full data, while ensuring the consistency of screening standards for different batches.

[0084] Traditional methods, trained solely on immediate post-harvest appearance characteristics, cannot adapt to the dynamic appearance changes of seeds from 0 to 48 hours post-harvest, resulting in low screening accuracy for seeds with long intervals. Furthermore, different batches and varieties of seeds exhibit common post-harvest physiological rhythms. By integrating full-cycle data from different batches of seeds to calculate correction parameters, the common patterns of post-harvest physiological rhythms can be systematically captured, providing a unified dynamic correction standard for subsequent screening and avoiding correction biases caused by the randomness of single-batch data. In addition, correction parameters can quantify the dynamic correlation between appearance characteristics and physiological indicators, shifting maturity judgment from solely relying on appearance to precise matching based on physiological essence, significantly reducing the misjudgment rate.

[0085] The parameter adjustment module is used to sample and collect the appearance characteristics of the batch of seeds to be screened within a set sampling time; and to adjust the correction parameters based on the appearance characteristics collected within the set sampling time to obtain the adjusted correction parameters.

[0086] A predetermined number of samples are randomly selected from the seed batch to be screened. At a specific time point of 0 hours post-collection, the appearance characteristics and physiological indicators of the samples are collected on-site. The rate of change of color characteristics and the temporary correlation coefficient matrix between appearance characteristics and physiological indicators are calculated within a set time interval. The difference between the rate of change of color characteristics within the set time interval and the baseline of the rate of change of color characteristics within the corresponding time interval is divided by the baseline of the rate of change of color characteristics to obtain the deviation ratio. The corrected baseline of the rate of change of color characteristics = baseline of the rate of change of color characteristics × (1 + deviation ratio). The temporary correlation coefficient matrix is ​​compared with the general physiological-appearance correlation coefficient matrix. Coefficients with differences exceeding a preset difference threshold are replaced. Based on the corrected baseline of the rate of change of color characteristics and the replaced physiological-appearance correlation coefficient matrix, the feature correction factor within the corresponding time interval is recalculated to obtain the adjusted correction parameters.

[0087] The preset difference threshold needs to be set based on the fluctuation range of the physiological-appearance correlation coefficients of different batches of seeds in the general baseline data. By analyzing the differences in correlation coefficients among multiple batches of seeds of the same variety, a critical value that can distinguish between normal fluctuations and differences is determined. Specifically, the degree of variation of the physiological-appearance correlation coefficients in historical batches is statistically analyzed, and a value that can reflect the true difference in the strength of the correlation between batches is selected as the threshold. This ensures that when the difference between the temporary correlation coefficient matrix and the general matrix exceeds this threshold, it is determined to be a change in the correlation pattern caused by the specificity of the new batch, and then the coefficients are replaced so that the adjusted correlation coefficient matrix better reflects the actual correlation characteristics of the new batch of seeds.

[0088] For example, 20-30 seeds are randomly selected from the batch to be screened, and their appearance characteristics and physiological indicators are collected at three time points: 0 hours, 3 hours, and 6 hours post-harvest. Initial appearance images of the entire batch of seeds are taken at 0 hours post-harvest for final screening. Taking images of the sampled seeds at 0 hours post-harvest can be achieved by setting up a temporary collection point at the harvesting site. Specifically, portable image acquisition equipment, such as a small industrial camera with a fixed light source, is provided at the seed harvesting site. Immediately after harvesting, samples are randomly selected from the batch, and 0-hour appearance images are taken on-site, with the harvesting time simultaneously recorded as the 0-hour time point.

[0089] First, the appearance characteristics and physiological indicators of 20-30 seeds from the new batch were pre-treated within 0-6 hours to remove appearance characteristic values ​​or physiological indicator values ​​that deviated from the characteristics of the batch population.

[0090] Extract the H value change within 0 to 6 hours, subtract the feature value of the previous time node from the feature value of the next time node, and then divide by 6 hours to obtain the actual color change rate of the batch within 0 to 6 hours; at the same time, calculate the Pearson correlation coefficient between the appearance characteristics of the batch within 0 to 6 hours and the corresponding physiological indicators to form a temporary correlation coefficient matrix.

[0091] The actual color change rate of the new batch from 0 to 6 hours is compared with the color change rate benchmark from 0 to 6 hours to calculate the deviation value. The color change rate benchmark from 0 to 6 hours is then corrected according to the deviation ratio. The temporary correlation coefficient matrix is ​​compared with the general physiological-appearance correlation coefficient matrix. For positions where the difference exceeds 10%, the corresponding coefficients of the general matrix are replaced with the coefficients of the temporary matrix. Based on the corrected color change rate benchmark from 0 to 6 hours and the adjusted correlation coefficients, the color feature correction factor for each hour node within 0 to 6 hours is recalculated to ensure that it matches the actual feature changes of the new batch.

[0092] After making the above adjustments to the correction parameters, the adjusted correction parameters are obtained. The adjusted correction parameters are then verified using data from 5–8 seeds in the new batch that were not included in the calculations. If the prediction error of the physiological indicators decreases by more than 30% compared to the original correction parameters, the adjustment is considered effective.

[0093] The adjustment of the correction parameters is mainly to address the batch-specific differences that exist when adapting the correction parameters to new batches of seeds. The correction parameters are constructed based on the common patterns of 3 to 5 batches of seeds. However, there are subtle differences in the short-term physiological changes after harvest among different batches of seeds, such as the rate of color change from 0 to 6 hours and the correlation strength between appearance characteristics and physiological indicators. Directly using the correction parameters will lead to a decrease in screening accuracy.

[0094] The classification and screening module, based on the adjusted correction parameters, filters the seeds to be screened by maturity and obtains a list of seed maturity classifications.

[0095] First, the initial appearance features of the entire batch of seeds are standardized, and the appearance features of each seed are extracted to ensure that the format of the appearance features is consistent with the input requirements of the modified physiological calibration model.

[0096] For color features, the current color feature value is corrected by a color feature correction factor to obtain the color feature at 0 hours post-harvest. The correction formula is: Color feature value at 0 hours post-harvest = Current color feature value - (Color feature correction factor × Interval time from harvest to screening).

[0097] For shape and texture features, the physiological-appearance correlation coefficient matrix in the modified physiological calibration model is combined with the feature change patterns corresponding to the post-harvest interval. The aspect ratio and energy are then corrected for deviations to obtain the shape and texture features at 0 hours post-harvest. The specific method is as follows:

[0098] First, the difference in shape and texture features between different batches of seeds of the same variety at adjacent time points is taken and divided by the time interval to obtain the rate of change of shape and texture features of a single batch of seeds at that interval. Then, the arithmetic mean of the rate of change of shape and texture features of multiple batches of seeds at the same time interval is taken as the average rate of change of shape and texture features of the same variety at the corresponding time interval.

[0099] Secondly, based on the interval between harvest and screening of the seeds to be corrected, determine the time interval to which they belong. For example, if the interval between harvest and screening is 9 hours, it belongs to the 6-12 hour interval. Call the average rate of change of shape and texture features of the corresponding time interval and calculate the cumulative change of shape and texture features from 0 hours after harvest to the current time node. The cumulative change of shape and texture features = average rate of change of shape and texture features × interval between harvest and screening.

[0100] Subsequently, Pearson correlation coefficients between shape and texture features and corresponding physiological indicators were extracted from the physiological-appearance correlation coefficient matrix. Using the Pearson correlation coefficients as weights, a weighted correction was applied to the cumulative changes in shape and texture features. Specifically, the larger the absolute value of the Pearson correlation coefficient for a shape or texture feature, the higher the weight of its cumulative change in shape and texture in the correction, making the correction result more consistent with the intrinsic relationship between features and physiological maturity.

[0101] Finally, the equivalent post-harvest 0-hour feature value is calculated using a modified formula: Equivalent post-harvest 0-hour shape and texture feature value = Current shape and texture feature value - (Weighted modified cumulative change in shape and texture features). This modified formula eliminates shape and texture deviations caused by physiological changes in the seed shape and texture during the interval between harvest and screening, restoring the true shape and texture features to those at 0 hours post-harvest.

[0102] The corrected equivalent 0-hour post-harvest appearance features are input into the classifier. Based on the physiological-appearance correlation coefficient matrix, strongly correlated appearance features and their corresponding threshold ranges are determined. The strongly correlated appearance features refer to the appearance feature corresponding to the appearance feature with the largest absolute value among the Pearson correlation coefficients of various appearance features corresponding to a certain physiological indicator in the physiological-appearance correlation coefficient matrix. The equivalent 0-hour post-harvest threshold range for each strongly correlated appearance feature is determined based on the numerical distribution of the equivalent 0-hour post-harvest strongly correlated appearance features of mature seeds in general basic data. For example, the appearance feature most strongly correlated with embryo vigor is texture, and the equivalent 0-hour post-harvest energy value of mature seeds must be greater than or equal to 0.35; the appearance feature most strongly correlated with seed coat chlorophyll content is color, and the equivalent 0-hour post-harvest H value of mature seeds must be greater than or equal to 100; the appearance feature most strongly correlated with endosperm soluble sugar content is shape, and the equivalent 0-hour post-harvest aspect ratio of mature seeds must be less than or equal to 1.8.

[0103] The classifier matches the equivalent appearance characteristics of each seed at 0 hours post-harvest one by one: first, it checks whether the energy value is greater than or equal to 0.35, corresponding to the seed embryo vigor meeting the standard; then it checks whether the H value is greater than or equal to 100, corresponding to the seed coat chlorophyll content meeting the standard; finally, it checks whether the length-to-width ratio is less than or equal to 1.8, corresponding to the endosperm soluble sugar content meeting the standard.

[0104] When all of the above appearance features meet the corresponding threshold range, the seed is determined to match a mature seed; if any appearance feature does not meet the corresponding threshold range, the seed is determined to match an immature seed. This multi-appearance feature collaborative verification method ensures that maturity judgment is based on the accurate correlation between physiological indicators and appearance features, reducing the possibility of misjudgment based solely on immediate appearance features.

[0105] After judging each seed in the entire batch, the judgment results of all seeds are summarized to obtain the final seed maturity classification list.

[0106] Example 2

[0107] This embodiment provides a system for screening and determining the maturity of Epimedium seeds from images, and also includes:

[0108] The parameter update module is used to recollect general basic data from the sample collection period and update and correct parameters when screening for new varieties of seeds or when environmental mutations cause changes in seed characteristics.

[0109] First, identify the special case type. If the special case type is selecting seeds of a new variety, then compare the variety types in the system database to confirm that there is no historical basic model data for the new variety type. If the special case type is environmental mutation, then select 30 seeds, collect 0 hours of appearance characteristics and physiological indicators, and compare them with the data of the same variety type in the same period of the historical batch. If the difference in appearance characteristics and physiological indicators exceeds 20%, it is determined to be a change in characteristics.

[0110] For confirmed special cases, 3 to 5 batches of seeds were selected for 48-hour full data collection. For each batch of seeds, at five time points after harvest (0 hours, 6 hours, 12 hours, 24 hours, and 48 hours), the appearance characteristics of the seeds were collected using an industrial camera with fixed parameters and a light source in an environment with stable temperature and humidity. Physiological indicators at the corresponding time points were recorded simultaneously. The collection method was consistent with the general basic data collection of the sample collection module.

[0111] The collected 48-hour full data was preprocessed to remove outliers caused by collection errors. Then, the appearance characteristics and physiological indicators of each batch of seeds were correlated along the time axis to form a new time-appearance-physiological three-dimensional dataset. Based on this dataset, the color change rate benchmark, the physiological-appearance correlation coefficient matrix, and the color feature correction factor were recalculated using the same methods as the parameter correction parameters in the parameter calculation module.

[0112] Replace the original corrected parameters with the newly generated corrected parameters, update the parameters in the system database, and record the update time, reason, and batch information to ensure that the updated corrected parameters are used when processing new batches of seeds. Verify the new parameters with validation data. If the prediction accuracy of physiological indicators reaches the set accuracy threshold, the update is confirmed to be effective.

[0113] The accuracy threshold needs to be determined based on historical data of the physiological indicator prediction accuracy of multiple effective models in the general baseline data. The specific processing procedure is as follows:

[0114] First, collect prediction accuracy data for physiological indicators from 3-5 batches of seeds of the same variety that have been confirmed as effective during the parameter adjustment or model validation phase. The physiological indicator prediction accuracy data comes from historically validated modified parameters; that is, the accuracy value corresponding to the degree of agreement between actual measured values ​​and predicted values ​​when the modified parameters are used to predict seed physiological indicators.

[0115] Secondly, statistical analysis was performed on the collected historical accuracy data to calculate its distribution characteristics, including the mean, median, and range of variation. By analyzing the data distribution, a critical value was determined that can distinguish between valid and invalid updates: this critical value must be higher than the highest accuracy of invalid models in historical data and lower than the lowest accuracy of valid models in historical data, to ensure that when the prediction accuracy of the new parameters reaches this value, it can be determined that it can stably reflect the correlation between seed physiological indicators and appearance characteristics.

[0116] Finally, considering the accuracy requirements of seed selection, this critical value is set as the accuracy threshold. For example, if the physiological indicator prediction accuracy of historically effective models is all above 90%, while the accuracy of ineffective models is mostly below 85%, then the accuracy threshold is set to 90% to ensure that the newly generated corrected parameters are validated when their physiological indicator prediction accuracy reaches 90%.

[0117] Example 3

[0118] Please see Figure 2 As shown in the figure, this embodiment provides a method for screening and determining the maturity of Epimedium seeds from images, including:

[0119] Within the set sample collection period, general basic data for calculating correction parameters are collected; the general basic data includes the appearance characteristics, physiological indicators, and seed collection information of seeds collected at each set time point within the sample collection period.

[0120] Based on general basic data, correction parameters for appearance feature correction are calculated; the correction parameters include color change rate benchmark, physiological-appearance correlation coefficient matrix and color feature correction factor;

[0121] The appearance characteristics of the batch of seeds to be screened are sampled within a set sampling time; and the correction parameters are adjusted based on the appearance characteristics collected within the set sampling time to obtain the adjusted correction parameters.

[0122] Based on the adjusted correction parameters, the seeds to be screened are screened by maturity to obtain a list of seed maturity classifications.

[0123] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0124] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for screening and determining the maturity of Epimedium seeds from images, characterized in that, include: Within the set sample collection period, collect general basic data used to calculate correction parameters; The general basic data includes the appearance characteristics, physiological indicators, and seed collection information of seeds collected at each set time point during the sample collection period. Based on general basic data, correction parameters for appearance feature correction are calculated; the correction parameters include color change rate benchmark, physiological-appearance correlation coefficient matrix and color feature correction factor; Methods for obtaining correction parameters for appearance feature modification include: The average appearance characteristics of seeds of the same variety and different batches at the same time point were calculated. The average rate of change of color characteristics was obtained by the ratio of the difference between adjacent time points to the time interval. The Pearson correlation coefficients between each appearance characteristic and each physiological index were calculated. The arithmetic mean of the average rate of change of color characteristics of different batches of seeds at the same time interval is taken to obtain the benchmark of the rate of change of color of seeds of the corresponding variety type at the corresponding time interval. By treating appearance features as rows and physiological indicators as columns, the corresponding elements of the matrix are the Pearson correlation coefficients between the corresponding appearance features and physiological indicators, thus obtaining the physiological-appearance correlation coefficient matrix. Based on the time interval of each hour node within the sample collection period, the color change rate benchmark of the corresponding time interval is called to calculate the cumulative color feature change from 0 hours after collection to each hour node after collection; the cumulative color feature change is multiplied by the corresponding Pearson correlation coefficient and summed to obtain the weighted cumulative color feature change; the weighted cumulative color feature change is divided by the corresponding time node value to obtain the color feature correction factor for each hour node. The appearance characteristics of the batch of seeds to be screened are sampled within a set sampling time; and the correction parameters are adjusted based on the appearance characteristics collected within the set sampling time to obtain the adjusted correction parameters. Based on the adjusted correction parameters, the seeds to be screened are screened by maturity to obtain a list of seed maturity classifications.

2. The method for screening and determining the maturity of Epimedium seeds according to claim 1, characterized in that: The appearance features include color features, shape features, and texture features; the color feature is the H value of the seed in the HSV space; the shape feature is the aspect ratio of the seed; and the texture feature is the energy of the seed calculated through the gray-level co-occurrence matrix. The physiological indicators include seed embryo vigor, seed coat chlorophyll content, and endosperm soluble sugar content. The seed collection information includes seed variety type, harvesting site and time, and the interval between harvest and screening.

3. The method for screening and determining the maturity of Epimedium seeds according to claim 1, characterized in that, The methods for setting the sample sampling period and the time nodes within the sample sampling period include: Calculate the ratio of the current appearance characteristics of the seed to the time interval from seed harvest to the present to obtain the change in appearance characteristics per unit time; The sample sampling period is set to fully cover the post-harvest physiologically active period of seeds. The physiologically active period is the fluctuation stage where the change in seed appearance characteristics per unit time exceeds the preset threshold for change per unit time corresponding to the variety type, and the stable stage where the change in seed appearance characteristics per unit time is lower than the preset benchmark threshold for change per unit time for multiple consecutive time nodes. The density of time nodes within the sample sampling period is adjusted according to the change in seed appearance characteristics per unit time. The interval between time nodes in the fluctuation phase is set to be less than a preset proportion of the average change period of the fluctuation phase; the interval between time nodes in the stable phase is set to a preset multiple of the interval of the fluctuation phase.

4. The method for screening and determining the maturity of Epimedium seeds according to claim 1, characterized in that, Methods for sampling the appearance characteristics of the batch of seeds to be screened within a sampling time, adjusting the correction parameters, and obtaining the adjusted correction parameters include: A predetermined number of samples were randomly selected from the batch of seeds to be screened. At the time point of 0 hours after harvest, the appearance characteristics and physiological indicators of the samples were collected on site. Calculate the rate of change of color features and the temporary correlation coefficient matrix between appearance features and physiological indicators of the sample within a set time interval; The difference between the color feature change rate within a set time interval and the color change rate reference for the corresponding time interval is divided by the color change rate reference to obtain the deviation ratio; the corrected color change rate reference = color change rate reference × (1 + deviation ratio). The temporary correlation coefficient matrix is ​​compared with the general physiological-appearance correlation coefficient matrix. Coefficients with differences exceeding a preset difference threshold are replaced. Based on the corrected rate of change benchmark and the replaced physiological-appearance correlation coefficient matrix, the feature correction factor for the corresponding time interval is recalculated to obtain the adjusted correction parameters.

5. The method for screening and determining the maturity of Epimedium seeds according to claim 2, characterized in that, Based on the adjusted correction parameters, methods for screening seeds by maturity to obtain a list of seed maturity classifications include: For color features, the current color feature value is corrected by a color feature correction factor to obtain the color feature 0 hours after sampling; For shape and texture features, the aspect ratio and energy are corrected according to the feature change pattern corresponding to the postharvest interval, based on the physiological-appearance correlation coefficient matrix in the modified physiological calibration model, to obtain the shape and texture features at 0 hours postharvest. The corrected equivalent appearance characteristics at 0 hours post-harvest are input into the preset classifier. After judging each seed in the entire batch, the final seed maturity classification list is obtained.

6. The method for screening and determining the maturity of Epimedium seeds according to claim 5, characterized in that, Methods for obtaining color characteristics at 0 hours post-harvest include: Color feature value 0 hours after harvest = current color feature value - (color feature correction factor × time interval from harvest to screening).

7. The method for screening and determining the maturity of Epimedium seeds according to claim 5, characterized in that, Methods for obtaining shape and texture features at 0 hours post-harvest include: The difference in shape and texture features between different batches of seeds of the same variety at adjacent time points is divided by the time interval to obtain the rate of change of shape and texture features of a single batch of seeds at that interval; the arithmetic mean of the rate of change of shape and texture features of multiple batches of seeds at the same time interval is taken as the average rate of change of shape and texture features of the same variety at the corresponding time interval. Based on the time interval from harvest to screening of the seeds to be corrected, determine the time interval to which the seeds belong, and obtain the average rate of change of shape and texture features of the corresponding time interval. The cumulative change of shape and texture features = average rate of change of shape and texture features × time interval from harvest to screening. Extract the Pearson correlation coefficients between shape and texture features and corresponding physiological indicators from the physiological-appearance correlation coefficient matrix; use the Pearson correlation coefficients as weights to perform weighted correction on the cumulative changes of shape and texture features to obtain the weighted correction of the cumulative changes of shape and texture features. Equivalent shape and texture feature value after 0 hours = current shape and texture feature value - (cumulative change in weighted and corrected shape and texture features).

8. The method for screening and determining the maturity of Epimedium seeds according to claim 5, characterized in that, The preset classification methods for the classifier include: Based on the physiological-appearance correlation coefficient matrix, strongly correlated appearance features and their corresponding threshold ranges are determined. The strongly correlated appearance feature refers to the appearance feature corresponding to the coefficient with the largest absolute value among the Pearson correlation coefficients of various appearance features corresponding to a certain physiological index in the physiological-appearance correlation coefficient matrix. The equivalent postharvest 0-hour threshold range of mature seeds corresponding to each strongly correlated appearance feature is determined based on the numerical distribution of the equivalent postharvest 0-hour strongly correlated appearance features of mature seeds in the general basic data.

9. A system for screening and determining the maturity of Epimedium seeds from images, used to implement the method for screening and determining the maturity of Epimedium seeds from images according to any one of claims 1-8, characterized in that, include: The sample acquisition module is used to collect general basic data for calculating correction parameters within a set sample acquisition period. The parameter calculation module calculates correction parameters for appearance feature correction based on general basic data. The parameter adjustment module is used to sample and collect the appearance characteristics of the batch of seeds to be screened within a set sampling time. The correction parameters are adjusted based on the appearance features collected within a set sampling time to obtain the adjusted correction parameters; The classification and screening module, based on the adjusted correction parameters, filters the seeds to be screened by maturity and obtains a list of seed maturity classifications.

Images