Highly resistant starch rice screening and identification method and system

Abstract

The application discloses a high-resistant starch rice screening and identification and content detection method and system, and relates to the technical field of rice screening. The high-resistant starch rice screening and identification and content detection system comprises a rice screening and identification module and a rice content detection module. The application realizes non-destructive, high-throughput and high-precision comprehensive evaluation of the resistant starch of rice by constructing a multi-modal fusion screening system of starch metabolism fluctuation frequency analysis, three-dimensional seed twin modeling, virtual enzyme molecule diffusion simulation and bionic digestion dynamics school, can complete large-scale seed screening under the premise of not significantly increasing sample loss, and improves the ability to describe the dynamic behavior of starch metabolism by introducing frequency domain features and spatial topology analysis, so that the screening result not only reflects the static content difference, but also can represent the dynamic law of the formation process.

Description

Technical Field

[0001] This invention relates to the field of rice screening technology, and in particular to a method and system for screening, identifying, and detecting the content of high-resistant starch rice. Background Technology

[0002] With the development of functional rice breeding and precision agriculture technologies, rice with high resistant starch content has attracted widespread attention due to its potential nutritional value, such as reducing postprandial glycemic response and improving gut microbiota. Existing screening methods for high-resistant starch rice mostly rely on chemical staining, enzymatic hydrolysis, in vitro digestion experiments, or near-infrared spectroscopy to detect starch content. These methods typically require destructive sampling of rice seeds, resulting in long testing cycles, high sample consumption, and difficulty in continuously tracking and screening individual seeds. Furthermore, traditional detection schemes are mostly based on static starch content analysis, lacking research on the dynamic fluctuations in starch metabolism during rice growth, making it difficult to accurately reflect the relationship between the formation mechanism and metabolic behavior of resistant starch under different growth environments.

[0003] Therefore, a screening and identification method for highly resistant starch rice and its content detection method is needed to achieve non-destructive, high-precision and dynamic screening and identification of highly resistant starch rice. Summary of the Invention

[0004] This invention aims to provide a method and system for screening, identifying, and detecting the content of highly resistant starch rice. By integrating dynamic frequency analysis of starch metabolism, three-dimensional twin modeling of seeds, and biomimetic digestion calibration mechanism, it achieves non-destructive, high-precision, and dynamic screening and identification of highly resistant starch rice.

[0005] A method for screening, identifying, and detecting the content of highly resistant starch rice includes the following steps:

[0006] Acquire starch metabolism fluctuation data and current growth environment parameters of the rice to be screened; perform data analysis based on starch metabolism fluctuation data to obtain the characteristic fluctuation frequency of starch synthesis; screen the rice to be screened based on the characteristic fluctuation frequency of starch synthesis to obtain the rice screening growth group and the rice screening rejection group.

[0007] In the rice screening growth group, spectral scanning was performed on all single intact rice seeds to obtain internal structure data of rice seeds; a three-dimensional twin model of rice seeds was constructed based on the internal structure data of rice seeds; a virtual enzyme molecular diffusion model was introduced into the three-dimensional twin model of rice seeds for simulation calculation to obtain the initial predicted value of rice starch.

[0008] Meanwhile, some single intact rice seeds were extracted as rice growth control samples for destructive testing: a biomimetic digestion reactor was pre-designed; digestion simulation was performed based on the rice growth control samples and the biomimetic digestion reactor to obtain simulated digestion kinetic characteristics; the virtual enzyme molecular diffusion model was calibrated based on the simulated digestion kinetic characteristics to obtain a corrected virtual enzyme molecular diffusion model.

[0009] A modified virtual enzyme molecular diffusion model was used to simulate and calculate the remaining single intact rice seeds to obtain the modified rice starch prediction value; the screening and identification operation of the rice to be screened was completed based on the modified rice starch prediction value.

[0010] As a preferred embodiment of the present invention, the specific steps for data analysis based on starch metabolism fluctuation data include:

[0011] Time-series spectral data of rice to be screened within a preset time period are obtained as starch metabolism fluctuation data; the starch metabolism fluctuation data is divided into multiple metabolic observation steps based on a sliding window.

[0012] The fast Fourier transform is used to project the starch metabolism fluctuation data according to the metabolic observation step size to obtain the metabolic power spectral density map; based on the metabolic power spectral density map, specific frequency components associated with amylase are extracted to obtain the characteristic fluctuation frequency of starch synthesis; the energy distribution entropy value of the characteristic fluctuation frequency of starch synthesis within a preset frequency band is calculated; the characteristic fluctuation frequency of starch synthesis and the corresponding energy distribution entropy value are input into a preset unsupervised clustering model for matching, and the high resistance matching degree of the starch to be screened is output.

[0013] The rice sorting threshold was set based on the high resistance matching degree of the starch to be screened and the current growth environment parameters; the rice screening growth group and the rice screening discard group were divided according to the rice sorting threshold and the high resistance matching degree of the starch to be screened.

[0014] As a preferred embodiment of the present invention, the specific steps for introducing a virtual enzyme molecular diffusion model into a three-dimensional twin model of rice seeds for simulation calculations include:

[0015] The internal structure data of rice seeds is transformed to obtain three-dimensional point cloud data of rice. The kernel density estimation algorithm is mapped to obtain a three-dimensional twin model of rice seeds. A three-dimensional topological skeleton network reflecting the permeable path of enzyme molecules is extracted from the three-dimensional twin model of rice seeds, and geometric impedance weights are assigned to the edges in the three-dimensional topological skeleton network.

[0016] In a three-dimensional twin model of rice seeds, a particle potential energy function is defined; the amplitude of the particle potential energy function is proportional to the local crystallinity; a virtual enzyme molecule diffusion model contains several virtual enzyme molecules; each virtual enzyme molecule moves according to a preset diffusion characteristic; the virtual enzyme molecule diffusion model is placed on a three-dimensional topological skeleton network for simulation to obtain the region inaccessible to enzymatic hydrolysis.

[0017] The energy percentage of point cloud corresponding to the region inaccessible by enzymatic hydrolysis is inverted to obtain the initial predicted value of rice starch.

[0018] As a preferred embodiment of the present invention, the specific steps for simulating digestion based on rice growth control samples and a biomimetic digestion reactor include:

[0019] Rice growth control samples were loaded into a biomimetic digestion reactor, and simulated oral cavity mechanical shearing parameters, simulated stomach acid environment parameters, and simulated small intestinal enzyme gradient environment parameters were set. The biomimetic digestion reactor included an initial rapid digestion stage and a later slow digestion stage.

[0020] Time-series data of reducing sugar concentration were collected from rice growth control samples in a biomimetic digestion reactor; the first derivative of the time-series data of reducing sugar concentration was calculated to obtain the reducing sugar release rate curve; the instantaneous release slope vector representing the transition relationship between the initial rapid digestion stage and the later slow digestion stage was extracted from the reducing sugar release rate curve.

[0021] Feature identification is performed based on the instantaneous release slope vector to obtain simulated digestion kinetics characteristics.

[0022] As a preferred embodiment of the present invention, the specific steps for calibrating a virtual enzyme molecular diffusion model based on simulated digestion kinetics include:

[0023] A virtual enzyme molecular diffusion model was run under the initial preset diffusion characteristics to obtain a virtual reducing sugar release rate curve; the virtual reducing sugar release rate curve and the reducing sugar release rate curve were spatiotemporally aligned and compared to obtain the target error function.

[0024] The preset diffusion characteristics are decoupled into physical diffusion parameters and chemical barrier parameters. For the initial rapid digestion stage, the physical partial derivatives of the target error function with respect to the physical diffusion parameters are calculated, and the physical diffusion parameters are corrected based on the physical partial derivatives to obtain the corrected physical diffusion parameters. For the later slow digestion stage, the chemical partial derivatives of the target error function with respect to the chemical barrier parameters are calculated, and the chemical barrier parameters are corrected based on the chemical partial derivatives to obtain the corrected chemical barrier parameters.

[0025] Perform iterative correction operations until the target error function converges to the preset error threshold range. Use the final corrected physical diffusion parameters and corrected chemical barrier parameters as the corrected preset diffusion characteristics to obtain the corrected virtual enzyme molecular diffusion model.

[0026] As a preferred embodiment of the present invention, the specific steps for simulating the remaining single intact rice seeds using a modified virtual enzyme molecular diffusion model include:

[0027] The modified virtual enzyme molecular diffusion model was applied to the three-dimensional twin model of rice seeds with remaining single intact rice seeds to re-execute the virtual enzyme molecular diffusion simulation, and the modified rice starch prediction value of each remaining single intact rice seed was obtained.

[0028] Extract the initial rice starch prediction value corresponding to each remaining single intact rice seed; calculate the starch prediction residual between the corresponding initial rice starch prediction value and the corrected rice starch prediction value; perform feature extraction based on the starch prediction residual to obtain the starch deviation distribution characteristics;

[0029] The initial predicted rice starch value is used as the basic resistance dimension, and the corrected predicted rice starch value is used as the true resistance dimension. A two-dimensional starch resistance evaluation space is constructed based on the basic resistance dimension and the true resistance dimension. The prediction confidence weight of the remaining single intact rice seeds in the two-dimensional comprehensive resistance evaluation space is calculated using the starch deviation distribution characteristics. The basic resistance dimension and the true resistance dimension are weighted and fused according to the prediction confidence weight to obtain the comprehensive resistant starch identification index.

[0030] The remaining single intact rice seeds were screened using a comprehensive resistant starch identification index according to a preset multi-level threshold range.

[0031] A screening, identification, and content detection system for high-resistant starch rice, including:

[0032] The rice screening and identification module includes a preliminary rice screening unit. The preliminary rice screening unit is used to acquire starch metabolism fluctuation data and current growth environment parameters of the rice to be screened. Based on the starch metabolism fluctuation data, data analysis is performed to obtain the characteristic fluctuation frequency of starch synthesis. Based on the characteristic fluctuation frequency of starch synthesis, the rice to be screened is screened to obtain the rice screening growth group and the rice screening rejection group.

[0033] The rice content detection module includes a rice content prediction unit and a rice correction screening unit. The rice content prediction unit performs spectral scanning on all single, intact rice seeds in the rice screening growth group to obtain internal structure data. Based on this data, a three-dimensional twin model of the rice seeds is constructed. A virtual enzyme molecular diffusion model is introduced into the three-dimensional twin model to perform simulation calculations and obtain an initial predicted value for rice starch. The rice correction screening unit extracts a portion of single, intact rice seeds as control samples for destructive testing: a pre-set biomimetic digestion reactor is used; digestion simulation is performed based on the control samples and the biomimetic digestion reactor to obtain simulated digestion kinetics; the virtual enzyme molecular diffusion model is calibrated based on these simulated digestion kinetics to obtain a corrected virtual enzyme molecular diffusion model; the corrected virtual enzyme molecular diffusion model is used to simulate calculations on the remaining single, intact rice seeds to obtain a corrected predicted value for rice starch; and the screening and identification of the rice to be screened is completed based on the corrected predicted value for rice starch.

[0034] The present invention has the following advantages:

[0035] 1. This invention achieves a non-destructive, high-throughput, and high-precision comprehensive evaluation of rice resistant starch by constructing a multimodal fusion screening system that integrates starch metabolism fluctuation frequency analysis, three-dimensional seed twin modeling, virtual enzyme molecular diffusion simulation, and biomimetic digestion kinetic calibration. It can complete large-scale seed screening without significantly increasing sample loss. Furthermore, by introducing frequency domain features and spatial topology analysis, it improves the ability to characterize the dynamic behavior of starch metabolism, enabling the screening results to not only reflect static content differences but also characterize the dynamic laws of its formation process.

[0036] 2. This invention effectively reduces model prediction bias and improves the accuracy and stability of resistant starch content prediction through a closed-loop calibration mechanism of virtual enzyme diffusion model and real digestion kinetics; through multi-scale structure reconstruction and hierarchical screening strategy, it significantly reduces the overall computational complexity while ensuring computational accuracy, making it more suitable for large-scale rice breeding screening scenarios; by introducing adaptive threshold and data-driven parameter optimization mechanism, the screening model can be dynamically adjusted according to different batches of samples and changes in the growth environment. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of the high-resistant starch rice screening, identification, and content detection system used in an embodiment of the present invention. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0039] Example 1, a method for screening, identifying, and detecting the content of highly resistant starch rice, includes the following steps:

[0040] Obtain data on starch metabolism fluctuations and current growth environment parameters of the rice to be screened;

[0041] When the rice to be screened is in the grain-filling stage, continuous spectral data is collected from rice grains at different growth time points. Starch metabolism fluctuation data refers to the time-series response data caused by changes in the concentrations of amylose, amylopectin, and related metabolites during the continuous synthesis and transformation of starch in rice. Essentially, it reflects the dynamic changes in the activity of starch metabolism within the rice plant. Specifically, a near-infrared spectral acquisition device can be used for non-contact scanning of the rice panicle, ensuring that the spectral signal covers the characteristic absorption bands related to starch bonding structures. Spectral curves at multiple time points are continuously collected at preset time intervals, and a time-series spectral matrix is ​​constructed according to the collection time sequence. The time-series spectral matrix was preprocessed, including using moving average filtering to eliminate random noise, using baseline correction to eliminate the influence of ambient light drift, and using normalization to unify the spectral amplitude range between different time points. After preprocessing, the intensity changes of characteristic bands of the same rice sample within a continuous time window were extracted as metabolic change vectors, and starch metabolism oscillation curves were calculated based on the differences between adjacent time points. Metabolic oscillations refer to the periodic changes of starch synthesis-related substances over time. All metabolic oscillation curves were combined in chronological order to obtain starch metabolism fluctuation data that reflects the dynamic changes in rice starch metabolism.

[0042] While collecting data on starch metabolism fluctuations, the environmental conditions of the rice planting area to be screened are monitored simultaneously. The current growth environment parameters refer to external environmental information that can affect the process of starch formation and accumulation in rice, including data such as air temperature, ambient humidity, light intensity, soil moisture content, carbon dioxide concentration, and diurnal temperature range.

[0043] In practical implementation, an environmental sensor network can be deployed within the planting area to collect environmental data from different locations in real time through multiple sensor nodes. Data output from different types of sensors is uniformly converted into standard digital signals and time-series aligned according to a unified timestamp. Time-series alignment refers to mapping data from different sources to the same time base to ensure consistency between environmental and metabolic data in subsequent analysis. After time synchronization, outlier data is identified and removed. Specifically, a local outlier detection algorithm is used to calculate the deviation between each environmental parameter and data from adjacent time windows. When the deviation exceeds a preset threshold, the corresponding data is identified as an outlier and interpolated for correction. All environmental parameters are standardized to convert data of different dimensions to a unified scale. The standardized environmental parameters are combined in chronological order to form an environmental state feature vector, which is then used as the current growth environment parameter input for subsequent screening and analysis to characterize the impact of different growth conditions on rice resistant starch formation behavior.

[0044] Data analysis was performed based on starch metabolism fluctuation data to obtain the characteristic fluctuation frequency of starch synthesis; the rice varieties to be screened were then screened based on the characteristic fluctuation frequency of starch synthesis to obtain the rice screening growth group and the rice screening rejection group.

[0045] The specific steps for data analysis based on starch metabolism fluctuation data include:

[0046] Time-series spectral data of rice to be screened within a preset time period are obtained as starch metabolism fluctuation data; the starch metabolism fluctuation data is divided into multiple metabolic observation steps based on a sliding window.

[0047] The acquired starch metabolism fluctuation data are arranged chronologically to form a continuous metabolic sequence. The sliding window method is a data processing technique that dynamically extracts local data intervals. It analyzes continuous data segment by segment over fixed-length time intervals to identify local metabolic change characteristics. Specifically, the window length and window movement step size are set, and data within the first time interval is extracted from the beginning of the metabolic sequence as the initial metabolic observation step size. The window is then moved forward by the preset step size, and the local data extraction operation is repeated to obtain multiple continuous metabolic observation step size data segments. During the window sliding process, each metabolic observation step size contains metabolic oscillation information within the corresponding time range. Local mean and variance calculations are performed on the data in each metabolic observation step size to measure the stability of metabolic activity within the current time interval. All metabolic observation steps are numbered and stored chronologically to provide a data foundation at a local time scale for subsequent frequency domain analysis.

[0048] The fast Fourier transform is used to project the starch metabolism fluctuation data according to the metabolic observation step size to obtain the metabolic power spectral density map; based on the metabolic power spectral density map, specific frequency components associated with amylase are extracted to obtain the characteristic fluctuation frequency of starch synthesis; the energy distribution entropy value of the characteristic fluctuation frequency of starch synthesis within a preset frequency band is calculated; the characteristic fluctuation frequency of starch synthesis and the corresponding energy distribution entropy value are input into a preset unsupervised clustering model for matching, and the high resistance matching degree of the starch to be screened is output.

[0049] The time-series data for each metabolic observation step is discretized to convert the continuous metabolic change curve into equally spaced digital signals. Fast Fourier Transform (FFT) is a data analysis method that converts time-domain signals into frequency-domain signals, its core function being to identify hidden periodic patterns in the data. Specifically, frequency decomposition is performed for each metabolic observation step, splitting the original time series into multiple combinations of sine and cosine waves of different frequencies, and calculating the amplitude value corresponding to each frequency component. The square of the amplitude of each frequency component is statistically analyzed to obtain the energy intensity at the corresponding frequency. Power spectral density refers to the energy distribution within a unit frequency range, reflecting the contribution of different frequency components to overall metabolic activity. The energy intensities corresponding to each frequency are arranged in frequency order, and a two-dimensional spectral image is constructed using grayscale mapping. In this image, the horizontal axis represents frequency change, the vertical axis represents different metabolic observation steps, and the color intensity represents the metabolic energy intensity at the corresponding frequency. This yields a metabolic power spectral density map that reflects the periodic changes in rice starch metabolism.

[0050] High-energy regions in the metabolic power spectral density map are identified, and frequency ranges with energy intensities exceeding a preset energy threshold are extracted. Specific frequency components refer to periodic change signals that consistently appear and have stable energy peaks during starch metabolism. Based on an existing database of amylase activity variation patterns, different frequency ranges are matched accordingly. Amylase is a crucial metabolic enzyme involved in starch decomposition and reconstruction, and its activity changes cause periodic oscillations during starch metabolism. Specifically, the stability of each frequency component within a continuous metabolic observation step is calculated, and the frequency stability coefficient is used to measure the recurrence rate of that frequency across different time intervals. Frequency components with high stability coefficients are screened, and the correlation coefficient between the corresponding frequency component and historical high-resistant starch samples is further calculated. The frequency component with the highest correlation and satisfactory stability is identified as the characteristic fluctuation frequency of starch synthesis, reflecting the corresponding metabolic rhythm characteristics during the formation of high-resistant starch.

[0051] A preset frequency band is established centered on the characteristic fluctuation frequency of starch synthesis; where a frequency band refers to a continuously distributed frequency range within a certain range. The energy values ​​corresponding to each frequency point within the frequency band are statistically analyzed, and all energy values ​​are normalized so that the sum of the energy proportions of each frequency point equals one. The information entropy is calculated based on the normalized energy proportion distribution; where the energy distribution entropy value is used to measure the dispersion of frequency energy distribution, a low entropy value indicates that energy is concentrated in a small number of frequencies, indicating that metabolic activity has strong regularity, while a high entropy value indicates that energy distribution is more dispersed, indicating that metabolic activity fluctuations are more complex. In specific implementation, the logarithmic information content of the energy proportion of each frequency point is calculated, and the information content of all frequency points is summed to obtain the energy distribution entropy value within the corresponding frequency band; the energy distribution entropy value is used as a dynamic characteristic parameter characterizing the stability of starch metabolism.

[0052] A two-dimensional feature vector is formed by combining the characteristic fluctuation frequency of starch synthesis with the corresponding energy distribution entropy value. Unsupervised clustering is a data analysis model that automatically classifies data based on the similarity between samples without requiring manual labeling. In practice, a distance calculation algorithm is first used to calculate the feature distance between different samples. This feature distance represents the similarity between different rice samples in terms of metabolic frequency and energy distribution. Based on the distance, similar samples are gradually clustered into multiple feature clusters. During clustering, the center position of each feature cluster is continuously updated, and the sample classification operation is repeated until the center positions of each feature cluster stabilize. The feature cluster to which the rice sample to be screened belongs is compared with historical high-resistance starch sample clusters, and the distance deviation between it and the target high-resistance feature center is calculated. The distance deviation is used to generate a high-resistance matching degree for the starch to be screened, representing the potential probability of the current rice sample forming high-resistance starch.

[0053] The rice sorting threshold was set based on the high resistance matching degree of the starch to be screened and the current growth environment parameters; the rice screening growth group and the rice screening discard group were divided according to the rice sorting threshold and the high resistance matching degree of the starch to be screened.

[0054] The process involves obtaining the starch resistance matching degree of the rice variety to be screened and simultaneously collecting growth environment parameters; conducting environmental sensitivity analysis on different environmental parameters to calculate the influence of factors such as temperature, humidity, light, and soil moisture content on starch formation stability; environmental sensitivity analysis refers to the data analysis process of assessing the strength of the influence of different environmental variables on the target metabolic outcome; in practice, an environmental impact weight matrix is ​​established using historical sample data, and the changing trend of the probability of high-resistant starch formation under different environmental conditions is calculated; the basic matching threshold is dynamically adjusted according to the environmental impact weight corresponding to the current environmental parameters; for example, the screening threshold is appropriately increased under high temperature and high humidity conditions to reduce the risk of misjudgment of screening results due to environmental disturbances; and a rice sorting threshold suitable for the current growth environment is obtained.

[0055] The high starch resistance matching degree of the rice samples to be screened is compared with the dynamically generated rice sorting threshold. When the high starch resistance matching degree is higher than the rice sorting threshold, the corresponding rice sample is judged to have high resistant starch formation potential; otherwise, it is judged to have insufficient resistant starch formation potential. Rice samples that meet the screening conditions are marked and classified into the rice screening growth group. The rice screening growth group refers to the set of target samples that will be retained for further testing, breeding or cultivation. At the same time, rice samples that do not meet the screening conditions are classified into the rice screening discard group. Corresponding sample numbers and screening record databases are established for the two groups of rice samples for subsequent germplasm management and resistant starch tracking analysis.

[0056] In the rice screening growth group, spectral scanning was performed on all single intact rice seeds to obtain internal structure data of rice seeds; a three-dimensional twin model of rice seeds was constructed based on the internal structure data of rice seeds; a virtual enzyme molecular diffusion model was introduced into the three-dimensional twin model of rice seeds for simulation calculation to obtain the initial predicted value of rice starch.

[0057] Single, intact rice seeds entering the rice selection growth group were numbered, and a fixed-position device was used to spatially locate the seeds to reduce data errors caused by posture deviation during subsequent scanning. A multi-angle hyperspectral scanning device was used to scan the rice seeds layer by layer, allowing incident light of different wavelengths to penetrate the seed's internal structure and acquiring reflection and absorption intensity information at different depths. The internal structure data of the rice seed refers to digitized data reflecting the distribution of starch granules, protein matrix density, pore structure, and spatial state of local crystalline regions within the seed. Specifically, based on the attenuation characteristics of light of different wavelengths propagating within the seed, a spectral absorption matrix was established for the corresponding location, and the spatial structure distribution within the seed was restored using tomographic reconstruction. The acquired raw spectral data underwent noise filtering and interlayer correction. Interlayer correction compensates for light intensity deviations between different scanning depths to ensure the consistency of the overall structural reconstruction. The structural data from all scanned layers were fused according to spatial coordinates to obtain internal structure data of the rice seed that reflects the microscopic structural state of the rice.

[0058] Spatial coordinate standardization is performed on the internal structure data of rice seeds to uniformly map the structural positions in different scanning layers to the same three-dimensional coordinate system. A three-dimensional twin model of rice seeds refers to a virtual three-dimensional model constructed digitally in a computer that corresponds to the internal structure of real rice seeds, reflecting the spatial structural relationships and material distribution within the real seed. Different tissue regions in the internal structure data are segmented and identified, including starch aggregation regions, protein-encapsulated regions, and microporous regions. Specifically, gray-level gradient changes are used to identify different structural boundaries, and spatial region division is completed based on the continuity of neighboring pixels. The structural information corresponding to different regions is converted into three-dimensional voxel units; where a voxel is the smallest structural unit in three-dimensional space, its function is similar to a pixel in a two-dimensional image. A three-dimensional spatial mesh is established based on the spatial connection relationships between voxels, and a continuous three-dimensional structural surface is generated using a surface reconstruction algorithm. The generated three-dimensional spatial structure is associated and mapped with the corresponding physical property parameters to obtain a three-dimensional twin model of rice seeds that reflects the internal structural characteristics of real rice.

[0059] The specific steps for introducing a virtual enzyme molecular diffusion model into a three-dimensional twin model of rice seeds for simulation calculations include:

[0060] The internal structure data of rice seeds is transformed to obtain three-dimensional point cloud data of rice. The kernel density estimation algorithm is mapped to obtain a three-dimensional twin model of rice seeds. A three-dimensional topological skeleton network reflecting the permeable path of enzyme molecules is extracted from the three-dimensional twin model of rice seeds, and geometric impedance weights are assigned to the edges in the three-dimensional topological skeleton network.

[0061] Each spatial location in the internal structure data of rice seeds is assigned a corresponding three-dimensional coordinate value; key structural points are extracted based on the spectral reflectance intensity and structural density information of different locations; the three-dimensional point cloud data refers to a three-dimensional digital set composed of a large number of spatial coordinate points, each point corresponding to an actual structural location inside the rice; in practice, all key structural points are arranged according to spatial coordinates, and attribute information such as local density value, crystallinity value, and pore characteristic value are added to each point; outlier removal is performed on the point cloud data to eliminate abnormal spatial points caused by scanning errors; the spatial distance relationship between each point is calculated using a neighborhood search algorithm, and local missing regions are repaired based on distance continuity; thus, three-dimensional point cloud data of rice that can reflect the topological relationship of the internal microstructure of rice is obtained.

[0062] Each spatial point in the 3D point cloud data of rice is used as the local structure sampling center. Kernel density estimation is a probabilistic analysis method used to estimate the continuity of spatial data distribution, inferring the density of the overall spatial structure based on the distribution of surrounding neighboring points. Specifically, a local search radius is established with each spatial point as the center, and the spatial distribution of other points within the search range is counted. Different weights are assigned based on the distance between neighboring points and the center point, with closer points receiving higher weights. The weights of all neighboring points are superimposed to obtain the local density value at the current spatial location. The density estimation operation is repeated for all points in the entire 3D space to form a continuous spatial density field. After constructing the density field, the spatial contours of different structural regions inside the seed are extracted based on the density change boundaries, and a continuous 3D structural surface is established. All structural regions are spatially fused to obtain a complete 3D twin model of the rice seed.

[0063] The process involves identifying porous and low-density connected regions in a three-dimensional twin model of rice seeds. The three-dimensional topological skeleton network, extracted from the complex three-dimensional structure, represents the central path network reflecting internal connectivity and is primarily used to describe possible diffusion channels for enzyme molecules within the seed. All connected regions are refined, gradually shrinking the complex spatial structure into central skeleton lines. The connections between these skeleton lines are converted into a network graph structure, where nodes represent structural intersections and edges represent accessible paths for enzyme molecules. A geometric impedance weight is calculated for each edge, representing the degree of spatial obstruction encountered by enzyme molecules during diffusion along that path. In practice, factors such as path length, pore width, local curvature, and surrounding structural density are considered to calculate a comprehensive impedance value for each path. All path impedance parameters are mapped onto the three-dimensional topological skeleton network for subsequent enzyme diffusion simulation.

[0064] In a three-dimensional twin model of rice seeds, a particle potential energy function is defined; the amplitude of the particle potential energy function is proportional to the local crystallinity; a virtual enzyme molecule diffusion model contains several virtual enzyme molecules; each virtual enzyme molecule moves according to a preset diffusion characteristic; the virtual enzyme molecule diffusion model is placed on a three-dimensional topological skeleton network for simulation to obtain the region inaccessible to enzymatic hydrolysis.

[0065] Local crystallinity analysis was performed on different structural regions in a three-dimensional twin model of rice seeds. Local crystallinity refers to the proportion of ordered crystalline structures within starch granules; a higher value indicates a denser region that is more difficult for enzyme molecules to penetrate. A corresponding spatial potential field was constructed based on the crystallinity of the local regions. The particle potential energy function is a mathematical function used to describe the ease of enzyme molecule movement in different spatial regions, essentially simulating the restrictive effect of spatial structure on enzyme diffusion behavior. In practice, regions with higher crystallinity were assigned higher potential energy values, making it more difficult for virtual enzyme molecules to enter these regions during movement, while regions with lower crystallinity corresponded to lower potential energy values. Continuous interpolation of the potential energy values ​​throughout the three-dimensional space was performed to form a complete three-dimensional potential energy distribution field. The particle potential energy function was embedded into the three-dimensional twin model of rice seeds to construct the subsequent virtual enzyme molecule movement environment.

[0066] Multiple virtual enzyme molecules are randomly generated in the outer boundary region of a 3D twin model of rice seeds. These virtual enzyme molecules are digital particle objects used to simulate the diffusion behavior of real digestive enzymes within rice. Each virtual enzyme molecule is assigned corresponding diffusion parameters, including diffusion velocity, directional perturbation coefficient, potential energy sensitivity coefficient, and path turning probability. The preset diffusion features refer to a set of rules used to describe the motion behavior of enzyme molecules within the complex internal structure. In practice, each virtual enzyme molecule calculates its motion direction based on the current spatial potential energy gradient at each time step and generates a new spatial position by combining random perturbations. When a virtual enzyme molecule enters a high-impedance region, its movement velocity decreases, while when it enters a low-impedance region, its diffusion velocity increases. The motion trajectories of all virtual enzyme molecules are recorded in real time, and their dwell time and diffusion coverage in different regions are statistically analyzed. This forms a virtual enzyme molecule diffusion model that reflects the diffusion behavior of enzyme molecules within rice.

[0067] A virtual enzyme molecule diffusion model is loaded into a three-dimensional topological backbone network, and enzyme molecule diffusion simulation is performed in a time-iterative manner. In each simulation cycle, the next motion path of the virtual enzyme molecule is calculated based on its current position, local potential energy value, and geometric impedance weight. The spatial regions that all virtual enzyme molecules can cover throughout the entire simulation time are statistically analyzed. Among them, the enzymatically inaccessible region refers to the spatial region that the virtual enzyme molecule cannot effectively enter or cannot reach for a long time within the preset simulation time. This region usually corresponds to a high crystallinity or high structural density region. In practice, spatial positions that are not covered by virtual enzyme molecules for a long time are marked, and the degree of connectivity between them and the surrounding regions is calculated. Cluster analysis is performed on all low-coverage regions to form multiple continuous enzymatically inaccessible regions. All enzymatically inaccessible regions are mapped back to the three-dimensional twin model for subsequent resistant starch content prediction analysis.

[0068] The energy percentage of point cloud corresponding to the region inaccessible by enzymatic hydrolysis is inverted to obtain the initial predicted value of rice starch.

[0069] Three-dimensional point cloud data corresponding to all regions inaccessible to enzymatic degradation are extracted, and the structural density and local potential energy values ​​of the point clouds within each region are statistically analyzed. The energy contribution value is calculated based on the local structural stability of each point. The point cloud energy percentage refers to the proportion of stable energy in the overall three-dimensional structure of a certain region, reflecting its resistance to enzymatic degradation. Specifically, the point cloud energy values ​​in all inaccessible regions are summed, and their proportion of the total structural energy of the entire rice seed is calculated. A mapping relationship between energy percentage and resistant starch content is established based on a historical resistant starch sample database. The current point cloud energy percentage is converted into the corresponding predicted resistant starch value using an inversion calculation method. Inversion refers to the process of deriving the target physical quantity from known result characteristics. The initial predicted rice starch value for the current single intact rice seed is output.

[0070] Meanwhile, some single intact rice seeds were extracted as rice growth control samples for destructive testing: a biomimetic digestion reactor was pre-designed; digestion simulation was performed based on the rice growth control samples and the biomimetic digestion reactor to obtain the simulated digestion kinetic characteristics;

[0071] A portion of single, intact rice seeds were randomly selected from the rice growth group according to a preset sampling ratio as rice growth control samples. These rice growth control samples are standard test samples used to establish reference data for realistic digestion behavior, providing a practical calibration basis for the subsequent virtual enzyme diffusion model. The selected rice seeds were numbered and registered, and their corresponding initial predicted rice starch values, planting environment parameters, and growth cycle information were recorded. The rice growth control samples underwent destructive pretreatment, i.e., mechanically crushing the seeds to destroy their original intact structure and fully expose the internal starch granules. Destructive testing refers to a detection method that obtains more accurate internal component information by altering the original physical structure of the sample. Specifically, the crushed samples were sieved to homogenize particle size, reducing the impact of particle size differences on subsequent digestion behavior. The treated samples were placed in a constant-temperature, sealed environment for pre-equilibrium treatment to stabilize the internal moisture distribution. The pretreated rice growth control samples were used as input for subsequent biomimetic digestion simulations.

[0072] The specific steps for simulating digestion based on rice growth control samples and a biomimetic digestion reactor include:

[0073] Rice growth control samples were loaded into a biomimetic digestion reactor, and simulated oral cavity mechanical shearing parameters, simulated stomach acid environment parameters, and simulated small intestinal enzyme gradient environment parameters were set. The biomimetic digestion reactor included an initial rapid digestion stage and a later slow digestion stage.

[0074] A multi-stage reaction device was constructed to simulate the human digestive environment. The biomimetic digestion reactor is an experimental system capable of simulating the physical and chemical environmental changes during the digestive stages of the human mouth, stomach, and small intestine, aiming to reproduce the actual starch digestion process as closely as possible. Specifically, a temperature control unit, an acid-base regulation unit, a mechanical stirring unit, and an enzyme injection unit are installed inside the reactor. The temperature control module maintains the reactor at a constant temperature close to the human digestive environment. A fluid circulation system simulates the liquid flow process in the gastrointestinal tract, and periodic mechanical disturbances simulate chewing and gastric peristalsis. Simultaneously, staged reaction zones are set inside the reactor to correspond to the oral pretreatment stage, the gastric acid hydrolysis stage, and the small intestinal enzymatic hydrolysis stage, respectively. Parameters such as pH, enzyme concentration, and reaction time for each stage are preset, and an automatic switching mechanism is established to automatically adjust the reaction environment over time. The environmental initialization configuration of the biomimetic digestion reactor is completed.

[0075] Pretreated rice growth control samples are added to the sample reaction chamber of the biomimetic digestion reactor, and a preset proportion of simulated digestive liquid is added to the reaction chamber. The process then enters a simulated oral cavity mechanical shearing stage. The mechanical shearing parameters are control parameters used to simulate the squeezing, friction, and crushing effects on food during human chewing. Specifically, a mechanical stirring device periodically changes the rotation speed and shearing direction, causing continuous collisions and shear deformation of the rice particles in the liquid, thus simulating the structural destruction behavior during oral chewing. Next, the process enters a simulated gastric acid environment stage. An acid-regulating module gradually reduces the pH of the reaction system and controls the addition rate of the simulated pepsin solution, keeping the sample in a continuously acidic environment. Gastric acid environment parameters include the rate of pH change, acid concentration, and gastric residence time. Finally, the process enters a simulated small intestinal enzyme gradient environment stage. A multi-stage injection method gradually increases the amylase concentration, creating a low-to-high concentration gradient of digestive enzymes in the reaction system. The enzyme gradient environment refers to the reaction state where enzyme concentration gradually changes with spatial location or time. An automatic switching between digestion stages is achieved through a stage switching control module.

[0076] After the biomimetic digestion reactor is started, the overall digestion process is divided into stages based on the trend of reducing sugar release. The initial rapid digestion stage refers to the stage where the surface structure of starch granules is easily decomposed by enzymes, resulting in rapid release of reducing sugars. The later slow digestion stage is the stage where the digestion rate decreases significantly as the enzymes gradually enter the highly crystalline regions. In practice, at the initial stage of the reaction, due to the large number of amorphous starch areas on the sample surface, enzyme molecules can quickly contact and decompose the starch structure, thus the system records a high sugar release rate. As the easily digestible areas gradually decrease, enzyme molecules begin to be hindered by the highly crystalline regions and dense structures, leading to a gradual decrease in the reducing sugar release rate. The system automatically identifies the stage inflection points based on the change in sugar release rate per unit time. When the release rate drops below a preset proportion, the system determines that it has entered the later slow digestion stage from the initial rapid digestion stage. Corresponding time interval markers are established for different stages for subsequent kinetic analysis.

[0077] Time-series data of reducing sugar concentration were collected from rice growth control samples in a biomimetic digestion reactor; the first derivative of the time-series data of reducing sugar concentration was calculated to obtain the reducing sugar release rate curve; the instantaneous release slope vector representing the transition relationship between the initial rapid digestion stage and the later slow digestion stage was extracted from the reducing sugar release rate curve.

[0078] During the operation of the biomimetic digestion reactor, a small amount of reaction solution sample is automatically extracted at preset time intervals. The time-series data of reducing sugar concentration refers to the sequence of changes in reducing sugar concentration recorded over a continuous time dimension, reflecting the dynamic process of starch being gradually broken down by enzymes into small sugar molecules such as glucose. The extracted reaction solution sample undergoes a colorimetric reaction treatment, causing a color change in the reducing sugar and the detection reagent. The intensity of the corresponding color change is measured using an optical absorption detection device, and the reducing sugar concentration value at the corresponding time node is calculated based on a preset standard curve. The concentration values ​​corresponding to all time nodes are arranged in chronological order to form a continuous reducing sugar concentration change curve. Noise filtering and smoothing are performed on the concentration data to eliminate data anomalies caused by liquid sampling errors and detection fluctuations, resulting in complete time-series data of reducing sugar concentration.

[0079] The time-series data of reducing sugar concentration are calculated by difference between adjacent time points. The first derivative, a mathematical calculation method used to represent the rate of change of data, is used in this embodiment to reflect the rate of increase of reducing sugar concentration per unit time. Specifically, the concentration difference between adjacent time points is calculated and divided by the corresponding time interval to obtain the sugar release rate within each time interval. The release rates corresponding to all time intervals are continuously connected to form a reducing sugar release rate curve. This curve reflects the changing trend of enzymatic hydrolysis efficiency in different digestion stages. To avoid local abnormal fluctuations affecting the overall analysis results, the release rate curve is smoothed and fitted, and a local weighted regression algorithm is used to eliminate abrupt noise, resulting in a stable and continuous reducing sugar release rate curve.

[0080] Local slope analysis was performed on the reducing sugar release rate curve. The instantaneous release slope refers to the trend of the release rate curve near a certain time point, reflecting the real-time changes in the enzymatic hydrolysis rate. Specifically, a continuous sliding analysis window was established on the release rate curve, and local linear fitting was performed on the data within the window to obtain the local slope value at the corresponding time point. The local slope values ​​corresponding to all time points were arranged in chronological order to form a slope sequence. Abrupt regions in the slope sequence were identified, and the magnitude of change between adjacent slopes was calculated. Specifically, a significant decrease in slope is usually observed in the release rate curve when transitioning from the initial rapid digestion stage to the later slow digestion stage. Multiple consecutive slope values ​​in the corresponding transition region were combined to form an instantaneous release slope vector, thus obtaining an instantaneous release slope vector that characterizes the digestion stage transition behavior.

[0081] The simulated digestion kinetics are obtained by feature identification based on the instantaneous release slope vector. The instantaneous release slope vector is normalized to allow for comparison of slope changes across different samples on a uniform scale. Features such as peak position, descent rate, stage duration, and fluctuation stability are extracted from the slope vector. The simulated digestion kinetics refer to a comprehensive set of parameters characterizing the overall digestive behavior of rice samples in a biomimetic digestion environment. Specifically, a dynamic pattern recognition algorithm is used to analyze the changing trends in the slope vector and identify the transition features between the rapid and slow release stages. Corresponding kinetic parameters, including maximum release rate, transition time, sustained release duration, and decay coefficient, are established based on the transition position and intensity of change. All kinetic parameters are combined to form a digestion behavior feature vector, which is then compared with a historical database of highly resistant starch samples. This yields simulated digestion kinetics that reflect the actual enzymatic hydrolysis behavior of the current rice samples.

[0082] The virtual enzyme molecular diffusion model is calibrated based on simulated digestion kinetics to obtain a corrected virtual enzyme molecular diffusion model. The specific steps for calibrating the virtual enzyme molecular diffusion model based on simulated digestion kinetics include:

[0083] A virtual enzyme molecular diffusion model was run under the initial preset diffusion characteristics to obtain a virtual reducing sugar release rate curve; the virtual reducing sugar release rate curve and the reducing sugar release rate curve were spatiotemporally aligned and compared to obtain the target error function.

[0084] The previously constructed three-dimensional twin model of rice seeds was loaded into the virtual enzyme molecule diffusion model, and initial preset diffusion features were imported. These initial preset diffusion features refer to a set of enzyme molecule motion rules parameters artificially set at the initial stage of model building, used to describe the basic diffusion behavior of virtual enzyme molecules within the internal structure of rice. Specifically, the initial preset diffusion features include parameters such as enzyme molecule diffusion velocity, path turning probability, local residence time, potential energy sensitivity, and enzyme reactivity coefficient. Multiple virtual enzyme molecules were released into the three-dimensional topological framework network, and diffusion simulation was performed according to discrete time steps. Within each time step, the system calculated the diffusion behavior of virtual enzyme molecules based on their diffusion rate. The geometric impedance weight and local potential energy value corresponding to the current position are used to calculate the next direction of movement and the distance to move; the diffusion coverage of all virtual enzyme molecules in different regions is statistically analyzed, and the corresponding virtual starch degradation amount is calculated according to the position where the enzyme molecule comes into contact with the starch structure; whereby the virtual starch degradation amount refers to the theoretical degree of starch decomposition obtained by mathematical simulation; the corresponding virtual reducing sugar production amount is calculated based on the change of virtual starch degradation amount per unit time, and the production amount at all time points is arranged in chronological order to form a virtual reducing sugar concentration change curve; the first derivative of the concentration change curve is calculated to obtain the virtual reducing sugar release rate curve.

[0085] The study acquires both a virtual reducing sugar release rate curve generated by a virtual enzyme molecular diffusion model and a reducing sugar release rate curve actually measured by a biomimetic digestion reactor. The two sets of curves undergo time-axis unification processing to eliminate data offsets caused by different sampling frequencies or start times. Spatiotemporal alignment refers to the data processing procedure that establishes a correspondence between the two sets of dynamic curves in the time and trend dimensions. Specifically, a time interpolation algorithm is first used to resample the two sets of curves to the same time interval. A dynamic path matching algorithm is then used to calculate the optimal correspondence between the two sets of curves, ensuring that similar trends overlap as much as possible. After time-dimensional alignment, spatial feature correspondence analysis is performed on the local peak positions, slope trends, and stage inflection points of the two sets of curves. The difference deviation between the two sets of curves at each corresponding time node is calculated, and all deviation values ​​are squared and accumulated. The target error function is a mathematical function used to quantify the degree of difference between the virtual model output and the actual experimental results; a smaller value indicates that the model is closer to the actual digestion behavior. This target error function is then used for subsequent parameter calibration.

[0086] The preset diffusion characteristics are decoupled into physical diffusion parameters and chemical barrier parameters. For the initial rapid digestion stage, the physical partial derivatives of the target error function with respect to the physical diffusion parameters are calculated, and the physical diffusion parameters are corrected based on the physical partial derivatives to obtain the corrected physical diffusion parameters. For the later slow digestion stage, the chemical partial derivatives of the target error function with respect to the chemical barrier parameters are calculated, and the chemical barrier parameters are corrected based on the chemical partial derivatives to obtain the corrected chemical barrier parameters.

[0087] The parameters in the initial preset diffusion characteristics are classified by functional attributes. Decoupling refers to breaking down the originally coupled complex parameter system into multiple subsets of parameters with independent physical meaning. In practice, parameters that mainly affect the spatial mobility of enzyme molecules are classified as physical diffusion parameters, including diffusion velocity, path probability, spatial penetration distance, and random motion perturbation coefficient. Parameters that mainly affect the difficulty of enzyme molecules reacting with starch structures are classified as chemical barrier parameters, including local reaction activation difficulty, crystallization region resistance coefficient, and enzyme reaction sensitivity coefficient. The chemical barrier refers to the energy barrier encountered by enzyme molecules when undergoing enzymatic hydrolysis in highly crystalline structural regions. Independent parameter matrices are established for the two types of parameters, and their correlation with different digestion stages is recorded. This forms an independent physical diffusion parameter system and a chemical barrier parameter system, providing a basis for subsequent staged calibration.

[0088] Time interval data corresponding to the initial rapid digestion stage were extracted from the reducing sugar release rate curve. This initial rapid digestion stage mainly corresponds to the rapid diffusion of enzyme molecules and their preferential degradation of low-barrier regions, making it most sensitive to physical diffusion capabilities. While keeping the chemical barrier parameter constant, small perturbations were applied to different physical diffusion parameters, and the virtual enzyme molecule diffusion model was rerun. The changes in the target error function before and after the perturbation were compared, and the rate of change of the target error function relative to each physical diffusion parameter was calculated. The physical partial derivative is a mathematical description of the sensitivity of the target error function to changes in a certain physical diffusion parameter; a larger value indicates a more significant impact of that parameter on the model error. In practice, the corresponding parameter values ​​were adjusted according to the positive or negative direction of the physical partial derivative: decreasing the corresponding parameter when the error increases and increasing the corresponding parameter when the error decreases. The parameter perturbation and error calculation process was repeated for all physical diffusion parameters, and the parameter matrix was updated step by step. This yielded corrected physical diffusion parameters that more accurately reflect the actual initial digestion behavior.

[0089] The time interval data corresponding to the slow digestion stage in the later stage were extracted from the reducing sugar release rate curve. This slow digestion stage mainly corresponds to the digestion process after enzyme molecules gradually enter the highly crystalline and dense structural regions, making it more sensitive to chemical barrier parameters. While keeping the corrected physical diffusion parameters constant, small changes were applied to each chemical barrier parameter, and the enzyme diffusion simulation was re-executed. The error difference of the target error function before and after parameter changes was statistically analyzed, and the rate of change of the target error function relative to each chemical barrier parameter was calculated. The chemical partial derivatives are used to reflect the strength of the influence of different chemical barrier parameters on the later enzymatic hydrolysis behavior. In practice, when an increase in a chemical barrier parameter leads to a decrease in error, the weight of that parameter is increased; conversely, the weight of that parameter is decreased. Cyclic perturbation and error evaluation operations were performed on all chemical barrier parameters, and the corresponding parameter values ​​were updated step by step. This yielded corrected chemical barrier parameters that more accurately reflect the later slow-release behavior.

[0090] Perform iterative correction operations until the target error function converges to the preset error threshold range. Use the final corrected physical diffusion parameters and corrected chemical barrier parameters as the corrected preset diffusion characteristics to obtain the corrected virtual enzyme molecular diffusion model.

[0091] The modified physical diffusion parameters and modified chemical barrier parameters are recombined to form a new set of diffusion characteristic parameters, and the virtual enzyme molecular diffusion model is rerun using this new parameter set. The corresponding virtual reducing sugar release rate curve is regenerated and spatiotemporally aligned and compared with the real reducing sugar release rate curve. A new target error function value is recalculated, and it is determined whether it enters the preset error threshold range. Error convergence refers to the process where, as the parameters are continuously adjusted, the target error function gradually stabilizes and its amplitude of change continuously decreases. Specifically, when the target error function is still higher than the preset threshold, the system continues to perform physical diffusion parameter correction and chemical barrier parameter correction operations. The parameter change trend is recorded during each correction cycle, and the parameter adjustment amplitude is controlled using an error decay coefficient to prevent parameter oscillation. When the amplitude of the target error function change is lower than the preset stability threshold for several consecutive cycles, the model is determined to have reached a stable convergence state. The finally converged modified physical diffusion parameters and modified chemical barrier parameters are used as the modified preset diffusion characteristics, and a modified virtual enzyme molecular diffusion model is generated based on this parameter system.

[0092] The modified virtual enzyme molecular diffusion model was used to simulate and calculate the remaining single intact rice seeds, obtaining the modified rice starch prediction value. Based on the modified rice starch prediction value, the screening and identification of the rice to be screened was completed. The specific steps for simulating and calculating the remaining single intact rice seeds using the modified virtual enzyme molecular diffusion model include:

[0093] The modified virtual enzyme molecular diffusion model was applied to the three-dimensional twin model of rice seeds with remaining single intact rice seeds to re-execute the virtual enzyme molecular diffusion simulation, and the modified rice starch prediction value of each remaining single intact rice seed was obtained.

[0094] The calibrated and corrected virtual enzyme diffusion model was loaded into the three-dimensional twin model of each remaining intact rice seed. The corrected virtual enzyme diffusion model refers to an enzyme diffusion simulation model calibrated using real biomimetic digestion kinetics data, whose internal parameters more accurately reflect real enzymatic hydrolysis behavior. The release location, diffusion direction, and diffusion rate of the virtual enzyme molecule were reinitialized based on the internal structural characteristics of each rice seed. Discrete-time simulation was performed in the corresponding three-dimensional topological framework network, allowing the virtual enzyme molecule to diffuse and hydrolyze in different structural regions according to the corrected diffusion characteristics. Specifically, within each time step, the system calculated the interaction state between the virtual enzyme molecule and the local starch structure in real time, and dynamically adjusted the diffusion ability of the enzyme molecule based on local crystallinity, geometric impedance weights, and chemical barrier parameters. The amount of virtual reducing sugar generated throughout the entire simulation cycle was cumulatively counted, and the corresponding resistant starch retention level was calculated based on the remaining proportion of regions inaccessible to enzymatic hydrolysis. The resistant starch retention result for each rice seed was converted into a corrected rice starch prediction value to characterize the current true resistant starch level of the rice seed.

[0095] Extract the initial rice starch prediction value corresponding to each remaining single intact rice seed; calculate the starch prediction residual between the corresponding initial rice starch prediction value and the corrected rice starch prediction value; perform feature extraction based on the starch prediction residual to obtain the starch deviation distribution characteristics;

[0096] The initial rice starch prediction values ​​for each intact single rice seed are retrieved from the database of results from the previously uncalibrated virtual enzyme molecular diffusion model. These initial rice starch prediction values ​​refer to the preliminary resistant starch prediction results obtained based on the original virtual enzyme diffusion rules, primarily reflecting the theoretical prediction level before actual digestion calibration. The initial prediction records for each rice seed are indexed and matched according to their unique seed number, establishing a one-to-one correspondence between the initial prediction values ​​and the corrected prediction values. All initial prediction values ​​undergo data integrity verification to confirm the absence of missing data, duplicate records, or abnormal offsets. Specifically, an interval anomaly detection algorithm is used to identify abnormal prediction values ​​that significantly deviate from the overall distribution range, and these abnormal records are re-verified. The verified initial prediction values ​​are then uniformly arranged according to seed number order, forming a set of initial rice starch prediction values ​​corresponding to each remaining intact single rice seed.

[0097] The difference between the initial and corrected predicted starch values ​​for the same rice seed is calculated. The starch prediction residual refers to the deviation between the uncalibrated prediction and the prediction corrected for actual digestion behavior, reflecting the degree of difference between the original model and actual enzymatic digestion. Statistical analysis is performed on the starch prediction residuals for all rice seeds, and a residual distribution sequence is established. Local clustering analysis is conducted on the residual distribution sequence to identify concentrated distribution patterns in different residual intervals. Specifically, statistical characteristics such as the residual mean, residual fluctuation amplitude, residual skewness, and residual peak density are calculated. Skewness indicates whether the residual distribution has an overall tendency to be too high or too low, while peak density reflects the stability of concentrated residual regions. The residual change trend is divided into continuous intervals, and high-stability and high-volatility residual regions are identified. All residual statistical results are combined to form a starch deviation distribution characteristic, used to characterize the overall stability and reliability of the current prediction system.

[0098] The initial predicted rice starch value is used as the basic resistance dimension, and the corrected predicted rice starch value is used as the true resistance dimension. A two-dimensional starch resistance evaluation space is constructed based on the basic resistance dimension and the true resistance dimension. The prediction confidence weight of the remaining single intact rice seeds in the two-dimensional comprehensive resistance evaluation space is calculated using the starch deviation distribution characteristics. The basic resistance dimension and the true resistance dimension are weighted and fused according to the prediction confidence weight to obtain the comprehensive resistant starch identification index.

[0099] The initial predicted rice starch value of each rice seed is defined as the basic resistance dimension; the basic resistance dimension is used to characterize the original resistant starch characteristics of rice seeds under the theoretical diffusion model conditions; the corresponding corrected predicted rice starch value is defined as the true resistance dimension; the true resistance dimension is used to reflect the resistant starch level that is closer to the actual digestion result after calibration by real biomimetic digestion behavior; using the basic resistance dimension as the horizontal axis and the true resistance dimension as the vertical axis, all rice seeds are mapped to a two-dimensional space; the two-dimensional starch resistance evaluation space refers to the data evaluation space that simultaneously reflects the theoretical prediction ability and the actual resistance performance using a dual-dimensional approach; in specific implementation, the sample distribution density in the two-dimensional space is statistically analyzed, and the clustering regions corresponding to different resistance levels are identified; the degree of resistance difference between different samples is analyzed based on the positional relationship in the two-dimensional space; thus, a two-dimensional starch resistance evaluation space is obtained for comprehensively evaluating the resistant starch level of a single intact rice seed.

[0100] The prediction confidence level is calculated based on the magnitude of the starch prediction residuals and the stability of residual fluctuations for each rice seed. The prediction confidence weight is a weighting parameter representing the reliability of the current prediction result; a higher value indicates a more stable and reliable prediction result. Samples located in the low residual stability region are assigned higher prediction confidence weights, while samples located in the high fluctuation region have their corresponding weights reduced. The weighted contribution ratios of the basic resistance dimension and the true resistance dimension are calculated based on the prediction confidence weights. In practice, when the stability of the corrected prediction result for a sample is high, the weight ratio of the true resistance dimension is increased; conversely, the reference weight of the basic resistance dimension is appropriately retained. A weighted fusion operation is performed on the two dimensions, and the corresponding comprehensive resistance score is calculated. The comprehensive resistant starch identification index is a comprehensive evaluation parameter used to comprehensively reflect the resistant starch level and prediction confidence of rice seeds. The comprehensive resistant starch identification index of each remaining single intact rice seed is output.

[0101] The remaining intact single-grain rice seeds were screened using a comprehensive resistant starch identification index (CRSI) according to preset multi-level threshold intervals. A multi-level resistance evaluation standard was established based on a historical database of rice samples with high resistant starch, and multiple consecutive resistance threshold intervals were preset. These multi-level threshold intervals refer to continuous numerical ranges used to classify different levels of resistant starch. The CRSI of each remaining intact single-grain rice seed was compared with its corresponding threshold interval. When the CRSI was in the highest threshold interval, the corresponding rice seed was classified as a high-priority, high-resistant starch seed. When the CRSI was in the middle threshold interval, it was classified as a potentially high-resistant seed. Samples below the lowest threshold interval were identified as low-resistant starch seeds. In practice, corresponding germplasm tags were established for the screening results of different levels, and the corresponding three-dimensional structural characteristics, metabolic frequency characteristics, and digestive kinetic characteristics were recorded simultaneously. This completed the grading, screening, and identification of the remaining intact single-grain rice seeds.

[0102] In practice, high-precision 3D reconstruction and complete simulation calculations are not performed on all rice seeds individually. Specifically, a large number of rice seeds are first screened using time-series spectral data, metabolic fluctuation frequencies, and growth environment parameters. Candidate seed populations with high resistant starch potential are identified through an unsupervised clustering model. High-precision 3D twin modeling and biomimetic digestion calibration are performed only on a small number of representative samples that enter the candidate population, thereby reducing the overall computational load. Subsequently, a mapping relationship between structural features, digestion kinetics, and resistant starch content is established based on representative samples. The calibrated model is then extended to other seeds with similar structural features using a transfer mapping algorithm, achieving rapid screening at the population level. Therefore, the screening results in this embodiment do not rely on the absolutely independent modeling of individual seeds, but rather on a comprehensive screening and identification based on population structural similarity and metabolic characteristic similarity.

[0103] In practical implementation, to address the issue of insufficient accuracy in 3D models due to the small size of individual rice seeds, a method is first used: hyperspectral scanning is employed to obtain macroscopic structural distribution information, followed by local super-resolution interpolation algorithms to refine and reconstruct key regions. Super-resolution interpolation utilizes the continuity of neighborhood structures to spatially compensate for low-resolution regions, thereby restoring finer internal structural boundaries. Simultaneously, kernel density estimation and local topological constraints are introduced during 3D point cloud generation. The continuity of neighborhood points restores the internal pore paths and starch aggregation regions of the seed, thus avoiding structural breaks caused by insufficient single-point scanning accuracy. Furthermore, the 3D twin model in this embodiment is essentially a "functional digital twin model," focusing on reflecting enzyme molecule reachable paths, local crystallization resistance, and diffusion trends, rather than absolute micron-level morphological replication. Therefore, it allows for effective characterization of resistant starch behavior even under limited resolution conditions.

[0104] In this embodiment, all calculations and simulations are jointly implemented based on a frequency domain analysis model, a spatial topology diffusion model, and a kinetic calibration model. Specifically, the time-series spectral analysis process is completed using the Fast Fourier Transform algorithm: first, the time-series spectral data is discretized at fixed time intervals; then, the time-domain signal is converted to a frequency-domain signal using a frequency decomposition method; subsequently, the energy distribution at different frequencies is calculated, and a metabolic power spectral density map is generated. The three-dimensional twin modeling process is implemented based on a kernel density estimation model and a point cloud topology reconstruction algorithm: first, the spectral tomography data is converted to spatial point cloud coordinates; then, the spatial density distribution is calculated using a local neighborhood search; and finally, based on density changes... The continuous three-dimensional structure is restored. The virtual enzyme molecule diffusion simulation process is based on a combination of a random diffusion model and a potential field constraint model. Specifically, virtual enzyme molecules are first generated in a three-dimensional topological skeleton network. Then, the next trajectory of the enzyme molecule is calculated based on the local potential energy gradient, geometric impedance weight, and random perturbation coefficient. The coverage area of ​​the enzyme molecule and the starch degradation behavior are statistically analyzed. The kinetic calibration process is based on an error function iterative optimization model. Specifically, the virtual reducing sugar release curve is first aligned with the real reducing sugar release curve in time. Then, the error function between the two is calculated. Subsequently, the partial derivative changes corresponding to the physical diffusion parameters and the chemical barrier parameters are obtained respectively. The parameters are cyclically adjusted until the error converges. To address the issue of inconsistent data units, this embodiment employs a unified data standardization process. Specifically: First, a unified unit mapping table is established for data from different sources. For example, length is uniformly converted to spatial scale units, time to standard time units, and concentration to mass concentration units. Then, data with different dimensions are normalized, mapping different data to a unified numerical range using maximum and minimum value intervals. Next, the normalized data undergoes standard deviation correction to ensure similar weights for different data types during model training. Finally, standardized data is used for matrix operations during model computation, and the results are then converted back to corresponding physical units during output, ensuring unified computation and joint analysis across different data sources.

[0105] In this embodiment, all preset models and preset threshold parameters are determined using a data-driven initialization combined with iterative adaptive optimization, rather than fixed settings based on human experience. Specifically, the selection of preset models begins with constructing a training set based on a historical rice sample database. This training set includes spectral time-series data, three-dimensional structural feature data, and corresponding results of actual resistant starch content measurements for different rice varieties. Subsequently, for the metabolic fluctuation analysis module, various candidate frequency domain analysis models are compared and screened, such as the fast Fourier transform model, short-time frequency analysis model, and wavelet decomposition model. The stability of frequency feature extraction, noise resistance, and mean square error of correlation with actual resistant starch are used as evaluation indicators to select the appropriate model. The model with the lowest overall error and the highest correlation was selected as the final preset frequency domain analysis model. For the 3D twin modeling and structural reconstruction model, the model that achieves the optimal balance between pore path recovery accuracy and computational complexity was selected as the preset structural model by comparing the structural restoration error and topological connectivity preservation ability of the kernel density estimation model, voxel reconstruction model and point cloud interpolation reconstruction model. For the virtual enzyme molecule diffusion model, the parameters were calibrated by introducing simulation experimental data, and the performance of the random walk model, Brownian motion model and potential field constrained diffusion model in fitting error of reducing sugar release curve was compared. The hybrid diffusion model that can explain both the initial rapid digestion and the later slow digestion behavior was selected as the preset diffusion model.

[0106] To determine the preset threshold parameters, this embodiment adopts a combination of statistical distribution modeling and adaptive quantile optimization. First, a feature distribution space is constructed using historical high-resistant starch samples and ordinary samples. Key indicators such as starch synthesis characteristic fluctuation frequency, energy distribution entropy value, and comprehensive resistance index are statistically modeled to obtain the probability distribution curves of different categories of samples. Then, with the minimum classification error rate as the objective function, different candidate thresholds are traversed and calculated, and the initial threshold interval is determined using grid search or binary search methods. On this basis, a dynamic quantile adjustment mechanism is introduced, that is, the threshold position is automatically adjusted according to the current batch sample distribution. For example, the threshold for the high-resistant category is set to the top 85% to top 90% interval of the historical high-resistant sample distribution, and offset correction is made according to changes in environmental parameters. At the same time, by introducing a cross-validation mechanism, the samples are randomly divided into training subsets and validation subsets, and the classification accuracy and stability under different threshold combinations are repeatedly tested, so as to select the threshold with the smallest error and the best robustness as the final preset parameter.

[0107] Furthermore, during the entire system operation, all preset model parameters and thresholds support an online update mechanism. That is, when new experimental data or new digestion kinetic curves are added to the database, the system will recalculate the changes in feature distribution and use gradient descent or expectation-maximization algorithms to iteratively update the model parameters. At the same time, the thresholds will be adaptively drift-corrected to ensure that the model always maintains consistency with the latest data distribution, thereby avoiding screening bias caused by static parameters and improving the applicability and stability of the overall screening and identification method under different batches and environmental conditions.

[0108] Example 2, Screening, Identification and Content Detection System for High-Resistant Starch Rice, see [link / reference] Figure 1 As shown, it includes:

[0109] The rice screening and identification module includes a preliminary rice screening unit. The preliminary rice screening unit is used to acquire starch metabolism fluctuation data and current growth environment parameters of the rice to be screened. Based on the starch metabolism fluctuation data, data analysis is performed to obtain the characteristic fluctuation frequency of starch synthesis. Based on the characteristic fluctuation frequency of starch synthesis, the rice to be screened is screened to obtain the rice screening growth group and the rice screening rejection group.

[0110] The rice content detection module includes a rice content prediction unit and a rice correction screening unit. The rice content prediction unit performs spectral scanning on all single, intact rice seeds in the rice screening growth group to obtain internal structure data. Based on this data, a three-dimensional twin model of the rice seeds is constructed. A virtual enzyme molecular diffusion model is introduced into the three-dimensional twin model to perform simulation calculations and obtain an initial predicted value for rice starch. The rice correction screening unit extracts a portion of single, intact rice seeds as control samples for destructive testing: a pre-set biomimetic digestion reactor is used; digestion simulation is performed based on the control samples and the biomimetic digestion reactor to obtain simulated digestion kinetics; the virtual enzyme molecular diffusion model is calibrated based on these simulated digestion kinetics to obtain a corrected virtual enzyme molecular diffusion model; the corrected virtual enzyme molecular diffusion model is used to simulate calculations on the remaining single, intact rice seeds to obtain a corrected predicted value for rice starch; and the screening and identification of the rice to be screened is completed based on the corrected predicted value for rice starch.

[0111] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for screening, identifying, and detecting the content of high-resistant starch rice, characterized in that, Includes the following steps: Acquire starch metabolism fluctuation data and current growth environment parameters of the rice to be screened; perform data analysis based on starch metabolism fluctuation data to obtain the characteristic fluctuation frequency of starch synthesis. Based on the fluctuation frequency of starch synthesis characteristics, rice varieties were screened to obtain a rice screening growth group and a rice screening rejection group. In the rice screening growth group, spectral scanning was performed on all single intact rice seeds to obtain internal structure data of rice seeds; a three-dimensional twin model of rice seeds was constructed based on the internal structure data of rice seeds; a virtual enzyme molecular diffusion model was introduced into the three-dimensional twin model of rice seeds for simulation calculation to obtain the initial predicted value of rice starch. Meanwhile, some single intact rice seeds were extracted as rice growth control samples for destructive testing: a biomimetic digestion reactor was pre-designed; digestion simulation was performed based on the rice growth control samples and the biomimetic digestion reactor to obtain simulated digestion kinetic characteristics; the virtual enzyme molecular diffusion model was calibrated based on the simulated digestion kinetic characteristics to obtain a corrected virtual enzyme molecular diffusion model. The modified virtual enzyme molecular diffusion model was used to simulate and calculate the remaining single intact rice seeds, and the modified predicted value of rice starch was obtained. The screening and identification of rice varieties was completed based on the corrected predicted rice starch values.

2. The method for screening, identifying, and detecting the content of highly resistant starch rice according to claim 1, characterized in that, The specific steps for data analysis based on starch metabolism fluctuation data include: Time-series spectral data of rice to be screened within a preset time period are obtained as starch metabolism fluctuation data; the starch metabolism fluctuation data is divided into multiple metabolic observation steps based on a sliding window. The fast Fourier transform is used to project the starch metabolism fluctuation data according to the metabolic observation step size to obtain the metabolic power spectral density map; based on the metabolic power spectral density map, specific frequency components associated with amylase are extracted to obtain the characteristic fluctuation frequency of starch synthesis; the energy distribution entropy value of the characteristic fluctuation frequency of starch synthesis within a preset frequency band is calculated; the characteristic fluctuation frequency of starch synthesis and the corresponding energy distribution entropy value are input into a preset unsupervised clustering model for matching, and the high resistance matching degree of the starch to be screened is output. The rice sorting threshold was set based on the high resistance matching degree of the starch to be screened and the current growth environment parameters; the rice screening growth group and the rice screening discard group were divided according to the rice sorting threshold and the high resistance matching degree of the starch to be screened.

3. The method for screening, identifying, and detecting the content of highly resistant starch rice according to claim 2, characterized in that, The specific steps for introducing a virtual enzyme molecular diffusion model into a three-dimensional twin model of rice seeds for simulation calculations include: The internal structure data of rice seeds is transformed to obtain three-dimensional point cloud data of rice. The kernel density estimation algorithm is mapped to obtain a three-dimensional twin model of rice seeds. A three-dimensional topological skeleton network reflecting the permeable path of enzyme molecules is extracted from the three-dimensional twin model of rice seeds, and geometric impedance weights are assigned to the edges in the three-dimensional topological skeleton network. In a three-dimensional twin model of rice seeds, a particle potential energy function is defined; the amplitude of the particle potential energy function is proportional to the local crystallinity; a virtual enzyme molecule diffusion model contains several virtual enzyme molecules; each virtual enzyme molecule moves according to a preset diffusion characteristic; the virtual enzyme molecule diffusion model is placed on a three-dimensional topological skeleton network for simulation to obtain the region inaccessible to enzymatic hydrolysis. The energy percentage of point cloud corresponding to the region inaccessible by enzymatic hydrolysis is inverted to obtain the initial predicted value of rice starch.

4. The method for screening, identifying, and detecting the content of highly resistant starch rice according to claim 3, characterized in that, The specific steps for simulating digestion based on rice growth control samples and a biomimetic digestion reactor include: Rice growth control samples were loaded into a biomimetic digestion reactor, and simulated oral cavity mechanical shearing parameters, simulated stomach acid environment parameters, and simulated small intestinal enzyme gradient environment parameters were set. The biomimetic digestion reactor included an initial rapid digestion stage and a later slow digestion stage. Time-series data of reducing sugar concentration were collected from rice growth control samples in a biomimetic digestion reactor; the first derivative of the time-series data of reducing sugar concentration was calculated to obtain the reducing sugar release rate curve; the instantaneous release slope vector representing the transition relationship between the initial rapid digestion stage and the later slow digestion stage was extracted from the reducing sugar release rate curve. Feature identification is performed based on the instantaneous release slope vector to obtain simulated digestion kinetics characteristics.

5. The method for screening, identifying, and detecting the content of highly resistant starch rice according to claim 4, characterized in that, The specific steps for calibrating a virtual enzyme molecular diffusion model based on simulated digestion kinetics include: A virtual enzyme molecular diffusion model was run under the initial preset diffusion characteristics to obtain a virtual reducing sugar release rate curve; the virtual reducing sugar release rate curve and the reducing sugar release rate curve were spatiotemporally aligned and compared to obtain the target error function. The preset diffusion characteristics are decoupled into physical diffusion parameters and chemical barrier parameters. For the initial rapid digestion stage, the physical partial derivatives of the target error function with respect to the physical diffusion parameters are calculated, and the physical diffusion parameters are corrected based on the physical partial derivatives to obtain the corrected physical diffusion parameters. For the later slow digestion stage, the chemical partial derivatives of the target error function with respect to the chemical barrier parameters are calculated, and the chemical barrier parameters are corrected based on the chemical partial derivatives to obtain the corrected chemical barrier parameters. Perform iterative correction operations until the target error function converges to the preset error threshold range. Use the final corrected physical diffusion parameters and corrected chemical barrier parameters as the corrected preset diffusion characteristics to obtain the corrected virtual enzyme molecular diffusion model.

6. The method for screening, identifying, and detecting the content of highly resistant starch rice according to claim 5, characterized in that, The specific steps for simulating the remaining single intact rice seeds using a modified virtual enzyme molecular diffusion model include: The modified virtual enzyme molecular diffusion model was applied to the three-dimensional twin model of rice seeds with remaining single intact rice seeds to re-execute the virtual enzyme molecular diffusion simulation, and the modified rice starch prediction value of each remaining single intact rice seed was obtained. Extract the initial rice starch prediction value corresponding to each remaining single intact rice seed; calculate the starch prediction residual between the corresponding initial rice starch prediction value and the corrected rice starch prediction value; perform feature extraction based on the starch prediction residual to obtain the starch deviation distribution characteristics; The initial predicted rice starch value is used as the basic resistance dimension, and the corrected predicted rice starch value is used as the true resistance dimension. A two-dimensional starch resistance evaluation space is constructed based on the basic resistance dimension and the true resistance dimension. The prediction confidence weight of the remaining single intact rice seeds in the two-dimensional comprehensive resistance evaluation space is calculated using the starch deviation distribution characteristics. The basic resistance dimension and the true resistance dimension are weighted and fused according to the prediction confidence weight to obtain the comprehensive resistant starch identification index. The remaining single intact rice seeds were screened using a comprehensive resistant starch identification index according to a preset multi-level threshold range.

7. A system for screening, identifying, and detecting the content of high-resistant starch in rice, characterized in that, The system employs the screening, identification, and content detection method for high-resistant starch rice according to any one of claims 1-6, including: The rice screening and identification module includes a preliminary rice screening unit. The preliminary rice screening unit is used to acquire starch metabolism fluctuation data and current growth environment parameters of the rice to be screened. Based on the starch metabolism fluctuation data, data analysis is performed to obtain the characteristic fluctuation frequency of starch synthesis. Based on the characteristic fluctuation frequency of starch synthesis, the rice to be screened is screened to obtain the rice screening growth group and the rice screening rejection group. The rice content detection module includes a rice content prediction unit and a rice correction screening unit. The rice content prediction unit performs spectral scanning on all single, intact rice seeds in the rice screening growth group to obtain internal structure data. Based on this data, a three-dimensional twin model of the rice seeds is constructed. A virtual enzyme molecular diffusion model is introduced into the three-dimensional twin model to perform simulation calculations and obtain an initial predicted value for rice starch. The rice correction screening unit extracts a portion of single, intact rice seeds as control samples for destructive testing: a pre-set biomimetic digestion reactor is used; digestion simulation is performed based on the control samples and the biomimetic digestion reactor to obtain simulated digestion kinetics; the virtual enzyme molecular diffusion model is calibrated based on these simulated digestion kinetics to obtain a corrected virtual enzyme molecular diffusion model; the corrected virtual enzyme molecular diffusion model is used to simulate calculations on the remaining single, intact rice seeds to obtain a corrected predicted value for rice starch; and the screening and identification of the rice to be screened is completed based on the corrected predicted value for rice starch.

Images