Wheat germ processing scheme recommendation method and system combined with deep learning

Abstract

The application provides a wheat germ processing scheme recommendation method and system combined with deep learning, first, multi-dimensional basic data of wheat germ (including variety characteristics, raw material quality, microscopic structure image data) and processing scene demand parameters (including processing product application scene, processing capacity adaptation, processing resource constraint data) are acquired; then, feature gene extraction processing is performed on the above data to obtain a wheat germ feature gene set and a scene demand feature gene set; the feature gene set and the scene demand feature gene set are input into a pre-trained deep scheme evolution model to generate a processing link feature gene sequence through a gene interaction layer; a plurality of processing scheme prototypes are constructed based on the processing link feature gene sequence; finally, double-cycle adaptive optimization is performed on the processing scheme prototype and the feature gene set to obtain a target processing recommendation scheme and transmit the target processing recommendation scheme to a processing execution terminal, so that scientific and accurate wheat germ processing scheme recommendation is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, in particular to a wheat germ processing scheme recommendation method and system combining deep learning. BACKGROUND

[0002] In the field of wheat germ processing, formulating a scientific and reasonable processing scheme is crucial for improving product quality, meeting the needs of different application scenarios, and optimizing resource allocation. Traditional wheat germ processing scheme formulation mainly relies on manual experience and simple experimental methods. Workers usually design processing flow and parameters based on their own understanding of the characteristics of wheat germ and past processing experience, combined with basic conditions such as processing site and equipment. However, the above-mentioned methods have obvious limitations.

[0003] On the one hand, manual experience is subjective and limited, and different personnel may have different perceptions and judgments of the characteristics of wheat germ, resulting in uneven quality of processing schemes. Moreover, with the continuous enrichment of wheat germ varieties and the increasing diversification of processing application scenarios, manual experience is difficult to fully consider various complex factors, and cannot accurately match different varieties of wheat germ with the needs of specific application scenarios. On the other hand, simple experimental methods can only test a limited number of variables, making it difficult to comprehensively analyze and optimize the multi-dimensional characteristics of wheat germ and various constraints in the processing process. This not only increases the cost and time of testing, but also may result in the final determined processing scheme not achieving the best effect, failing to fully utilize the value of wheat germ, and failing to meet the market demand for high-quality processed products. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a wheat germ processing scheme recommendation method combining deep learning, which comprises:

[0005] Obtaining multi-dimensional basic data of wheat germ and processing scene demand parameters, the multi-dimensional basic data of wheat germ including variety characteristic data, raw material quality data, and microscopic structure image data of wheat germ, and the processing scene demand parameters including processing product application scene data, processing capacity adaptation data, and processing resource constraint data;

[0006] Performing feature gene extraction processing on the multi-dimensional basic data of wheat germ and the processing scene demand parameters to obtain a set of wheat germ feature genes and a set of scene demand feature genes;

[0007] input the wheat germ characteristic gene set and the scene demand characteristic gene set into the pre-trained deep scheme evolution model, and the wheat germ characteristic gene set and the scene demand characteristic gene set are dynamically recombined through a gene interaction layer of the deep scheme evolution model to generate a processing link characteristic gene sequence;

[0008] Based on the processing link characteristic gene sequence, a processing link unit is combined through a scheme prototype construction layer of the deep scheme evolution model to form multiple processing scheme prototypes;

[0009] The multiple processing scheme prototypes are double-circulation adapted and optimized with the wheat germ characteristic gene set and the scene demand characteristic gene set to obtain a target processing recommendation scheme, and the target processing recommendation scheme is transmitted to a processing execution terminal.

[0010] In another aspect, the embodiment of the present application also provides a wheat germ processing scheme recommendation system combined with deep learning, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for running the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0011] Based on the above aspects, the embodiment of the present application can simulate the process of biological evolution by acquiring multi-dimensional basic data of wheat germ and processing scene demand parameters, performing feature gene extraction processing on the acquired data, converting complex data into a representative characteristic gene set, dynamically recombining the wheat germ characteristic gene set and the scene demand characteristic gene set through a gene interaction layer of a pre-trained deep scheme evolution model, generating a processing link characteristic gene sequence that meets specific demands, and realizing intelligent design and optimization of the processing link. Multiple processing scheme prototypes are constructed based on the processing link characteristic gene sequence, which increases the diversity of the scheme and provides more possibilities for selecting the optimal scheme. Through double-circulation adaptation and optimization of the multiple processing scheme prototypes and the characteristic gene set, various factors can be considered to ensure that the target processing recommendation scheme obtained finally meets the processing scene demand, maximizes the value of the wheat germ, improves the processing efficiency and product quality, and reduces the processing cost. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is an execution flow schematic diagram of the wheat germ processing scheme recommendation method combined with deep learning provided by the embodiment of the present application.

[0013] Figure 2 is a hardware architecture schematic diagram of the wheat germ processing scheme recommendation system combined with deep learning provided by the embodiment of the present application. DETAILED DESCRIPTION

[0014] The application will be specifically described below with reference to the drawings of the specification, Figure 1 is a flowchart of a wheat germ processing scheme recommendation method combining deep learning provided by an embodiment of the application, and the wheat germ processing scheme recommendation method combining deep learning will be described in detail below.

[0015] Step S110: Obtain wheat germ multi-dimensional basic data and processing scene demand parameters, the wheat germ multi-dimensional basic data including wheat germ variety characteristic data, raw material quality data, and microscopic structure image data, and the processing scene demand parameters including processing product application scene data, processing capacity adaptation data, and processing resource constraint data.

[0016] In this embodiment, taking a food processing plant planning to process a batch of wheat germ for making high-end nutritional supplements as an example, the implementation process of the entire method is unfolded. When obtaining the wheat germ multi-dimensional basic data, the method combines raw material procurement records and laboratory testing. The variety characteristic data is extracted from the procurement records, including the variety name of the wheat germ, the soil type of the planting area, the climate conditions, the planting cycle, and other information. The raw material quality data is obtained through laboratory testing, covering the content of nutrients such as protein, dietary fiber, and vitamin E, as well as data such as moisture content, impurity types and proportions. The microscopic structure image data is obtained by scanning electron microscopy on randomly selected multiple wheat germ samples, obtaining image information of cell arrangement, cell wall thickness, endosperm structure, etc. at different magnifications.

[0017] The processing scene demand parameters are determined through communication with the production management department and the market department of the plant. The processing product application scene data is clear for making high-end nutritional supplements, the target user group is the health-conscious middle and high-end consumer group, and the product needs to meet the relevant food safety standards and nutrition index requirements. The processing capacity adaptation data is determined according to market demand, including daily planned output, continuous operation time of production line, processing capacity of each batch, etc. The processing resource constraint data involves the existing processing equipment of the plant, such as the type and processing capacity of cleaning equipment, the specifications of drying equipment, the precision of grinding equipment, etc., and the information of raw material supply, including the total amount of the batch of wheat germ, the supply cycle, and the energy consumption limit and environmental protection requirements in the production process.

[0018] When obtaining these data, for privacy-sensitive information involved, such as the business secrets of raw material suppliers and the personal information of detection personnel, data desensitization technology is used for processing. Specifically, the name and contact information of the supplier are anonymized, and the identity information of the detection personnel is encrypted and stored, and only authorized management personnel can access it through specific rights when necessary, to ensure the security and confidentiality of the data.

[0019] Step S120: Feature gene extraction processing is performed on the multi-dimensional basic data of wheat germ and the processing scene demand parameters to obtain a wheat germ feature gene set and a scene demand feature gene set. The wheat germ feature gene set includes variety feature genes, quality feature genes, and structure feature genes. The scene demand feature gene set includes application scene feature genes, production capacity adaptation feature genes, and resource constraint feature genes.

[0020] In this embodiment, feature gene extraction processing is performed on the obtained multi-dimensional basic data of wheat germ and the processing scene demand parameters. First, for the multi-dimensional basic data of wheat germ, key information in variety feature data, raw material quality data, and microscopic structure image data is extracted, converted into corresponding feature gene segments, and then integrated into a wheat germ feature gene set. For the processing scene demand parameters, the core content of application scene data, processing production capacity adaptation data, and processing resource constraint data is also analyzed, converted into feature gene segments, and then formed into a scene demand feature gene set.

[0021] Step S121: Variety feature data in the multi-dimensional basic data of wheat germ is extracted, and structured analysis is performed on the variety feature data to separate variety source association information, variety growth cycle association information, and variety resistance association information. The variety source association information, variety growth cycle association information, and variety resistance association information are converted into quantifiable variety feature gene segments, each of which corresponds to the core representation of a variety attribute.

[0022] In this embodiment, after extracting the variety feature data of wheat germ, structured analysis is performed. The variety source association information includes the soil pH, altitude, and annual precipitation of the planting area, which are converted into quantifiable numerical values, such as the soil pH represented by a pH value range, the altitude recorded in meters, and the annual precipitation counted in millimeters, to form a variety feature gene segment corresponding to the variety source association information. This variety feature gene segment includes numerical values corresponding to different source attributes.

[0023] The variety growth cycle association information covers the number of days from sowing to harvesting and the duration of each growth stage. These time information is converted into a numerical sequence in units of days to form a variety feature gene segment corresponding to the variety growth cycle association information. The numerical values in this variety feature gene segment correspond to the duration of specific stages in the growth cycle.

[0024] The variety resistance correlation information includes resistance to common diseases and pests, drought resistance, cold resistance, etc. By consulting the breeding data and planting records of the variety, the resistance information is quantified into different grade values, for example, the resistance is divided into several grades from strong to weak, and each grade corresponds to a specific value, forming a variety characteristic gene fragment corresponding to the variety resistance correlation information. The variety characteristic gene fragment contains the quantified values of multiple resistance indicators.

[0025] Step S122: Extracting raw material quality data in the multi-dimensional basic data of wheat germ, analyzing the ingredient composition information, freshness correlation information, and impurity content correlation information in the raw material quality data, and decomposing the ingredient composition information, freshness correlation information, and impurity content correlation information into quality characteristic gene fragments. Each quality characteristic gene fragment contains specific characterization parameters of one quality indicator.

[0026] In this embodiment, when analyzing the raw material quality data, the ingredient composition information includes the content of nutrients such as protein, dietary fiber, and vitamin E. The content of each ingredient is expressed in the form of percentage or content per 100 grams, forming a quality characteristic gene fragment corresponding to the ingredient composition information. The quality characteristic gene fragment contains quantified parameters of multiple nutrients.

[0027] The freshness correlation information involves indicators such as acid value and peroxide value. The detection results of these indicators are converted into corresponding numerical values, forming a quality characteristic gene fragment corresponding to the freshness correlation information. The numerical values in the quality characteristic gene fragment reflect the freshness degree of the wheat germ.

[0028] The impurity content correlation information includes the types of impurities, such as soil, weed seeds, and other grain particles, as well as the proportion of each impurity. After quantifying these information, a quality characteristic gene fragment corresponding to the impurity content correlation information is formed. The quality characteristic gene fragment contains the identification of impurity types and the numerical value of the corresponding proportion.

[0029] Step S123: Extracting microstructure image data in the multi-dimensional basic data of wheat germ, obtaining cell structure distribution information, tissue structure density information, and surface morphology correlation information in the microstructure image through image feature analysis technology, and converting the cell structure distribution information, tissue structure density information, and surface morphology correlation information into structure characteristic gene fragments. Each structure characteristic gene fragment corresponds to a quantized description of one microstructure attribute.

[0030] In this embodiment, image feature analysis technology is used to process the microstructure image data. First, the image is preprocessed, including noise reduction, contrast enhancement, etc., to more clearly extract features.

[0031] Step S1231: input the microstructure image data of the wheat germ into the image feature analysis module, divide the microstructure image into regions, and separate the cell region, tissue gap region, and surface coverage region.

[0032] In this embodiment, the image feature analysis module uses an edge detection algorithm and a region growing algorithm to divide the microstructure image into regions. The edge detection algorithm identifies the boundaries between different regions in the image, such as the boundaries between cells and the boundaries between cells and tissue gaps. The region growing algorithm collects pixels with the same or similar features to form different regions based on the similarity of the pixels. By combining these two algorithms, the image is separated into cell regions, tissue gap regions, and surface coverage regions, each with clear boundaries and ranges.

[0033] Step S1232: extract pixel features from the cell region, calculate the gray scale distribution information, edge profile information, and texture direction information of the pixels in the cell region, determine the arrangement of the cell structure, cell size correlation information, and cell integrity correlation information based on the gray scale distribution information, edge profile information, and texture direction information of the pixels in the cell region, and form cell structure distribution information.

[0034] In this embodiment, when extracting pixel features from the cell region, the gray scale distribution information is calculated by counting the number and proportion of pixels with different gray scale values in the cell region to obtain a numerical sequence corresponding to a gray scale distribution histogram. The edge profile information is obtained by calculating the gradient changes of the cell edge pixels, reflecting the shape and complexity of the cell outline, and forming a multi-dimensional numerical set. The texture direction information is obtained by analyzing the arrangement and direction of the pixels in the cell region, represented by multiple direction parameters.

[0035] Based on this information, the arrangement of the cell structure is determined, such as whether it is arranged regularly or not, and the direction of the arrangement, represented by corresponding codes and parameters. The cell size correlation information includes the average diameter, maximum diameter, and minimum diameter of the cells. The cell integrity correlation information is quantified based on whether the cells are damaged or deformed, and the corresponding numerical values are obtained. These information collectively constitute the cell structure distribution information.

[0036] Step S1233: perform density analysis on the tissue gap region, calculate the number of tissue gaps per unit area, gap size distribution, and gap connectivity, generate tissue structure density information based on the number of tissue gaps per unit area, gap size distribution, and gap connectivity, and the tissue structure density information can represent the tightness of the wheat germ tissue.

[0037] In this embodiment, when the density of the tissue gap region is analyzed, the number of tissue gaps in a unit area is counted in a set unit area, and a numerical value is obtained. The gap size distribution is measured by the area or volume of each tissue gap, which is classified and counted according to the size range to form a distribution sequence. The gap connectivity is obtained by analyzing whether the tissue gaps are connected to each other and the degree of connection, and is represented by a multi-dimensional parameter.

[0038] Based on the data generated by the tissue structure density information, including the number of gaps per unit area, the gap size distribution parameters, and the connectivity parameters, the tightness of the wheat germ tissue can be reflected. The fewer and smaller the tissue gaps are, and the worse the connectivity is, the tighter the tissue is.

[0039] Step S1234: Morphological feature extraction is performed on the surface coverage area to obtain the surface relief information, the smoothness correlation information, and the defect distribution information. The surface relief information, the smoothness correlation information, and the defect distribution information jointly constitute the surface morphology correlation information.

[0040] In this embodiment, when the morphological feature extraction is performed on the surface coverage area, the surface relief information is obtained by calculating the height variation of the surface pixels, and is represented by a plurality of height difference statistical parameters, such as the maximum height difference and the average height difference. The smoothness correlation information is calculated according to the change rate of the surface pixel grayscale value. The smaller the change rate is, the smoother the surface is, and a corresponding quantitative parameter is formed.

[0041] The defect distribution information includes the position, number, and size of defects such as cracks and spots on the surface. After quantization, a multi-dimensional numerical set is formed, in which each dimension corresponds to a related parameter of a defect. These information jointly constitute the surface morphology correlation information.

[0042] Step S1235: The cell structure distribution information, the tissue structure density information, and the surface morphology correlation information are respectively gene fragment coded. The cell structure distribution information corresponds to a group of structure characteristic gene fragments, the tissue structure density information corresponds to a group of structure characteristic gene fragments, and the surface morphology correlation information corresponds to a group of structure characteristic gene fragments. The quantization parameters and the correlation of the cell structure distribution information, the tissue structure density information, and the surface morphology correlation information are preserved during the coding process.

[0043] In this embodiment, when the cell structure distribution information is gene fragment coded, the arrangement mode of the cell structure, the cell size correlation information, the cell integrity correlation information and other quantitative parameters are converted into specific numerical sequences according to the preset coding rules to form a group of structure characteristic gene fragments. Each fragment in the structure characteristic gene group corresponds to a specific attribute in the cell structure distribution information, and the correlation relationship between the fragments is consistent with the attribute correlation in the original information.

[0044] When the tissue structure density information is coded, the number of gaps per unit area, the gap size distribution parameters, the connectivity parameters and the like are converted into numerical sequences according to the coding rules to form corresponding structure characteristic gene fragments, so as to ensure that the mutual relationship between these parameters is retained after coding.

[0045] When the surface morphology correlation information is coded, the relief degree information, the smoothness correlation information and the defect distribution information of the surface are converted into numerical sequences respectively to form corresponding structure characteristic gene fragments, while the correlation between the parameters is maintained.

[0046] Step S124: The variety characteristic gene fragments, the quality characteristic gene fragments and the structure characteristic gene fragments are associated and integrated, the interaction relationship between the variety characteristic gene fragments, the quality characteristic gene fragments and the structure characteristic gene fragments is marked, and a wheat germ characteristic gene set containing a gene fragment list and an interaction relationship is formed.

[0047] In this embodiment, the variety characteristic gene fragments, the quality characteristic gene fragments and the structure characteristic gene fragments are associated and integrated. First, the interaction relationship between these fragments is analyzed. For example, the growth cycle information in the variety characteristic gene fragments may affect the nutrient content in the quality characteristic gene fragments, and the cell structure information in the structure characteristic gene fragments may be correlated with the variety source in the variety characteristic gene fragments.

[0048] The interaction relationship is marked by establishing a correlation matrix, and the elements in the matrix represent the correlation strength between different gene fragments. Then all the gene fragments are arranged into a list, and together with the correlation matrix, a wheat germ characteristic gene set is formed, which completely presents the characteristics of the wheat germ and the internal relationship.

[0049] Step S125: Analyze the application scenario data, processing capacity adaptation data, and processing resource constraint data in the processing scene demand parameter, respectively convert the application scenario data into application scenario feature gene segments, the processing capacity adaptation data into capacity adaptation feature gene segments, and the processing resource constraint data into resource constraint feature gene segments, associate the application scenario feature gene segments, the capacity adaptation feature gene segments, and the resource constraint feature gene segments to form a scene demand feature gene set, and the gene segment structure of the scene demand feature gene set is adapted to the gene segment structure of the wheat germ feature gene set.

[0050] In this embodiment, the processing scene demand parameter is analyzed, each item of data is respectively converted into a corresponding feature gene segment, and is associated to form a scene demand feature gene set, and it is ensured that the gene segment structure thereof is adapted to the wheat germ feature gene set, so as to facilitate subsequent interactive processing.

[0051] Step S1251: Analyze the application scenario data in the processing scene demand parameter, determine the target use type of the processing product, the target user group association information, and the product quality level association requirement, decompose the target use type of the processing product into use feature sub-items, decompose the target user group association information into user feature sub-items, and decompose the product quality level association requirement into quality feature sub-items, the use feature sub-items correspond to a group of application scenario feature gene segments, the user feature sub-items correspond to a group of application scenario feature gene segments, and the quality feature sub-items correspond to a group of application scenario feature gene segments.

[0052] In this embodiment, when the application scenario data is analyzed, the target use type of the processing product is to make high-end nutritional supplements, which is decomposed into use feature sub-items such as the form of the supplements (powder, capsule, etc.), the eating method (direct eating, adding to other food, etc.), and each sub-item is converted into a group of application scenario feature gene segments containing specific parameters of the sub-item.

[0053] The target user group association information is a health-conscious middle and high-end consumer group, which is decomposed into user feature sub-items such as age range, consumption ability, and health demand focus, each sub-item corresponds to a group of application scenario feature gene segments, and records related description parameters.

[0054] The product quality level association requirement is to meet relevant food safety standards and nutrition indicators, which is decomposed into quality feature sub-items such as pollutant limit, minimum content of nutritional ingredients, etc., each sub-item is converted into a group of application scenario feature gene segments containing corresponding standard values and detection method identifiers.

[0055] Step S1252: Analyze the processing capacity adaptation data in the processing scene requirement parameter, extract the unit time yield requirement, continuous production time length requirement, and batch production scale requirement of the processing process, convert the unit time yield requirement of the processing process into a yield characteristic subitem, convert the continuous production time length requirement into a time length characteristic subitem, and convert the batch production scale requirement into a scale characteristic subitem, the yield characteristic subitem corresponds to a group of capacity adaptation characteristic gene segments, the time length characteristic subitem corresponds to a group of capacity adaptation characteristic gene segments, and the scale characteristic subitem corresponds to a group of capacity adaptation characteristic gene segments.

[0056] In this embodiment, the processing capacity adaptation data is analyzed, the unit time yield requirement is converted into a yield characteristic subitem, such as hourly yield, minute yield, etc., each subitem corresponds to a group of capacity adaptation characteristic gene segments, and contains a specific yield numerical range.

[0057] The continuous production time length requirement is converted into a time length characteristic subitem, such as the number of continuous production hours per day, the maximum length of each continuous production, etc., each subitem corresponds to a group of capacity adaptation characteristic gene segments, and records the time length related parameters.

[0058] The batch production scale requirement is converted into a scale characteristic subitem, such as the raw material input amount of each batch, the product output amount of each batch, etc., each subitem corresponds to a group of capacity adaptation characteristic gene segments, and contains a specific numerical value of the scale.

[0059] Step S1253: Analyze the processing resource constraint data in the processing scene requirement parameter, obtain the type limitation of the processing required equipment, the stability requirement of the raw material supply, and the energy consumption control requirement, decompose the type limitation of the processing required equipment into an equipment characteristic subitem, decompose the stability requirement of the raw material supply into a raw material characteristic subitem, and decompose the energy consumption control requirement into an energy consumption characteristic subitem, the equipment characteristic subitem corresponds to a group of resource constraint characteristic gene segments, the raw material characteristic subitem corresponds to a group of resource constraint characteristic gene segments, and the energy consumption characteristic subitem corresponds to a group of resource constraint characteristic gene segments.

[0060] In this embodiment, the processing resource constraint data is analyzed, the type limitation of the processing required equipment is decomposed into an equipment characteristic subitem, such as the type of cleaning equipment, the heating mode of drying equipment, the accuracy level of grinding equipment, etc., each subitem corresponds to a group of resource constraint characteristic gene segments, and contains specific parameters and model requirements of the equipment.

[0061] The stability requirement of the raw material supply is decomposed into a raw material characteristic subitem, such as the supply cycle of the raw material, the quantity fluctuation range of each supply, etc., each subitem corresponds to a group of resource constraint characteristic gene segments, and records the related time and quantity parameters.

[0062] The energy consumption control requirement is disassembled into energy consumption characteristic sub-items, such as the upper limit of the power consumption per product, the upper limit of the water consumption per product, and the like, each sub-item corresponds to a group of resource constraint characteristic gene segments, and contains a limiting value of energy consumption.

[0063] Step S1254: The application scenario characteristic gene segment, the production capacity adaptation characteristic gene segment, and the resource constraint characteristic gene segment are marked for relevance, the mutual influence relationship between the application scenario characteristic gene segment and the production capacity adaptation characteristic gene segment, the mutual influence relationship between the application scenario characteristic gene segment and the resource constraint characteristic gene segment, and the mutual influence relationship between the production capacity adaptation characteristic gene segment and the resource constraint characteristic gene segment are determined, for example, the correlation between the quality characteristic sub-item in the application scenario characteristic gene segment and the equipment characteristic sub-item in the resource constraint characteristic gene segment.

[0064] In this embodiment, the three types of characteristic gene segments are marked for relevance. The mutual influence between the application scenario characteristic gene segment and the production capacity adaptation characteristic gene segment is analyzed. When the product quality grade requirement in the application scenario is high, it may affect the production capacity per unit time, thereby establishing a correlation between the two.

[0065] The relationship between the application scenario characteristic gene segment and the resource constraint characteristic gene segment is determined, for example, the form requirement of the product in the application scenario may have a specific limitation on the type of processing equipment, thereby forming a correlation.

[0066] The mutual influence between the production capacity adaptation characteristic gene segment and the resource constraint characteristic gene segment is determined, for example, the continuous production time requirement may be limited by the energy consumption of the equipment, and the corresponding correlation is established.

[0067] Step S1255: Based on the relevance marking result, the application scenario characteristic gene segment, the production capacity adaptation characteristic gene segment, and the resource constraint characteristic gene segment are sorted and integrated according to the correlation relationship, to form a scenario demand characteristic gene set, and each gene segment in the scenario demand characteristic gene set contains corresponding demand sub-item information and associated gene segment identification.

[0068] In this embodiment, according to the relevance marking result, the application scenario characteristic gene segment, the production capacity adaptation characteristic gene segment, and the resource constraint characteristic gene segment are sorted and integrated according to the correlation closeness. The segments with greater mutual influence are placed in adjacent positions to form an ordered scenario demand characteristic gene set.

[0069] Each gene segment contains corresponding demand sub-item information, such as the quality characteristic sub-item specific parameter in the application scenario characteristic gene segment, and simultaneously contains the identification of the associated gene segment, such as which production capacity adaptation or resource constraint characteristic gene segment is associated with the segment, thereby facilitating quick identification and calling in the subsequent processing process.

[0070] Step S130: input the wheat germ feature gene set and the scenario demand feature gene set into the pre-trained deep scheme evolution model, and perform dynamic recombination on the two types of feature genes in the wheat germ feature gene set and the scenario demand feature gene set through the gene interaction layer of the deep scheme evolution model to generate a processing link feature gene sequence.

[0071] In this embodiment, the previously processed wheat germ feature gene set and the scenario demand feature gene set are input into the pre-trained deep scheme evolution model. The deep scheme evolution model is trained by a large amount of wheat germ processing case data and can understand the internal correlation and interaction mechanism of the two types of gene sets. The gene interaction layer, as an important component of the model, is responsible for dynamic recombination of the two types of input feature genes. By analyzing the matching degree and correlation between gene fragments, a processing link feature gene sequence that conforms to the processing logic is generated.

[0072] Step S131: input the wheat germ feature gene set and the scenario demand feature gene set into the gene interaction layer of the deep scheme evolution model. The gene interaction layer first performs similarity matching on the gene fragments in the wheat germ feature gene set and the gene fragments in the scenario demand feature gene set to determine the gene fragment pairs with corresponding correlations.

[0073] In this embodiment, after the gene interaction layer receives the wheat germ feature gene set and the scenario demand feature gene set, it starts the similarity matching program. The similarity matching program determines the corresponding relationship between gene fragments by calculating the coincidence and similarity of attribute parameters, function descriptions, and other information of each gene fragment in the two types of gene sets. For example, the quality feature gene fragment related to protein content in the wheat germ feature gene set can be matched with the application scenario feature gene fragment related to the protein index of the nutritional supplement in the scenario demand feature gene set. When the similarity of the two reaches a certain threshold, it is determined as a gene fragment pair with corresponding correlation.

[0074] Step S132: based on the correlation strength value of the gene fragment pair, the gene interaction layer calls the pre-stored gene recombination rules to perform information fusion on the gene fragment pair with corresponding correlation to generate a fusion gene fragment. The fusion gene fragment contains the core information of the gene fragment corresponding to the wheat germ feature gene set and the core information of the gene fragment corresponding to the scenario demand feature gene set in the gene fragment pair.

[0075] In this embodiment, after determining the gene fragment pairs, the correlation strength value of each gene fragment pair is calculated. According to the correlation strength value, the gene interaction layer calls the corresponding rules from the pre-stored gene recombination rule library for information fusion. In the fusion process, the core information of two types of gene fragments can be retained, such as the core parameters of the nutritional components of wheat germ and the core requirements of the nutritional indicators in the scene demand, and these core information is integrated into the new fusion gene fragment, so that the fusion gene fragment can reflect the inherent characteristics of the wheat germ and meet the scene demand.

[0076] Step S1321: The gene interaction layer calculates the correlation strength value of each gene fragment pair, which is determined based on the attribute similarity, functional matching degree, and demand adaptation degree of the gene fragments corresponding to the wheat germ characteristic gene set and the gene fragments corresponding to the scene demand characteristic gene set in the gene fragment pair.

[0077] In this embodiment, when calculating the correlation strength value, the attribute similarity is to compare the coincidence degree of the attribute parameter types and ranges of the two in the gene fragment pair, such as the coincidence of the parameter range of the vitamin E content in the wheat germ and the parameter range of the vitamin E index in the scene demand. The functional matching degree is to analyze the correlation degree of the roles played by the two in the processing process, such as the matching of the cell structure characteristics of the wheat germ and the demand of the grinding link in processing. The demand adaptation degree is to judge the degree to which the wheat germ characteristics can meet the scene demand, such as whether the impurity content of the wheat germ meets the requirement of the product purity of the scene. Through the comprehensive calculation of the values of the three dimensions, the correlation strength value of each gene fragment pair is obtained.

[0078] Step S1322: According to the preset range of the correlation strength value, the gene interaction layer calls the corresponding gene recombination rule. The gene fragment pair with the correlation strength value meeting the first preset range calls the deep fusion rule, the gene fragment pair with the correlation strength value meeting the second preset range calls the moderate fusion rule, and the gene fragment pair with the correlation strength value meeting the third preset range calls the basic fusion rule.

[0079] In this embodiment, three correlation strength value ranges are preset. The first preset range is the interval of higher correlation strength value, which is suitable for the deep fusion rule; the second preset range is the interval of medium correlation strength value, which is suitable for the moderate fusion rule; and the third preset range is the interval of lower correlation strength value, which is suitable for the basic fusion rule. The gene interaction layer automatically calls the corresponding recombination rule according to the interval to which the calculated correlation strength value belongs. For example, when the correlation strength value of the gene fragment of the protein content in the wheat germ and the gene fragment of the protein index in the scene demand is in the first preset range, the deep fusion rule is called.

[0080] Step S1323: For the gene fragment pair calling the deep fusion rule, the gene interaction layer fuses all attribute parameters of the gene fragment corresponding to the wheat germ characteristic gene set in the gene fragment pair with all attribute parameters of the gene fragment corresponding to the scene demand characteristic gene set parameter by parameter, calculates the average value of the parameters after parameter-by-parameter fusion and retains the parameter fluctuation range before parameter-by-parameter fusion, and generates a deep fusion gene fragment containing complete parameter information.

[0081] In this embodiment, for the gene fragment pair calling the deep fusion rule, such as the dietary fiber content gene fragment in the wheat germ.

[0082] In this embodiment, for the gene fragment pair calling the deep fusion rule, such as the gene fragment of the dietary fiber content in the wheat germ and the gene fragment of the dietary fiber index in the scene demand, all attribute parameters of the two are fused by parameter. For example, the content numerical range of the dietary fiber in the wheat germ is fused with the required numerical range of the dietary fiber in the scene demand, the average value of each corresponding parameter is calculated, and the original parameter fluctuation range of each is retained, and finally the deep fusion gene fragment generated contains the fusion average value and the fluctuation range information, and the parameter characteristics of the two are completely presented.

[0083] Step S1324: For the gene fragment pair calling the moderate fusion rule, the gene interaction layer selects the core attribute parameters of the gene fragment corresponding to the wheat germ characteristic gene set in the gene fragment pair and the core attribute parameters of the gene fragment corresponding to the scene demand characteristic gene set for fusion, retains the original values of the non-core parameters of the gene fragment corresponding to the wheat germ characteristic gene set in the gene fragment pair and the original values of the non-core parameters of the gene fragment corresponding to the scene demand characteristic gene set, and generates a moderate fusion gene fragment containing core fusion parameters and original non-core parameters.

[0084] In this embodiment, when the moderate fusion rule is called, the core attribute parameters of the two, such as the actual moisture content value of the wheat germ and the moisture content value required after drying in the scene, are selected for fusion, and the fused core parameters are calculated. For non-core parameters, such as the environmental temperature during the wheat germ moisture detection and the model of the drying equipment in the scene demand, the original values are retained and not fused, thereby generating a moderate fusion gene fragment containing core fusion parameters and original non-core parameters.

[0085] Step S1325: For the gene fragment pair calling the basic fusion rule, the gene interaction layer only establishes the association identifier between the gene fragment corresponding to the wheat germ feature gene set and the gene fragment corresponding to the scene demand feature gene set in the gene fragment pair, retains the complete parameter information of the gene fragment corresponding to the wheat germ feature gene set and the complete parameter information of the gene fragment corresponding to the scene demand feature gene set, and generates a basic fusion gene fragment with an association identifier.

[0086] In this embodiment, when the association strength value of the gene fragment pair is in the third preset range, such as the planting area soil pH value gene fragment of the wheat germ and the environmental protection requirement gene fragment of the packaging material in the scene demand, the gene interaction layer only establishes the association identifier between the two, indicating that they have certain indirect association. At the same time, all information such as the soil pH value parameter range of the wheat germ feature gene fragment and the packaging material environmental protection index parameter of the scene demand gene fragment is completely retained, and a basic fusion gene fragment with an association identifier is generated.

[0087] Step S133: For the single gene fragment without forming a corresponding association, the gene interaction layer supplements the association information through the feature expansion technology, so that the single gene fragment can form a new association relationship with other fusion gene fragments, and a supplementary gene fragment is generated.

[0088] In this embodiment, there are some single gene fragments without forming a corresponding association with other gene fragments, such as a certain trace mineral gene fragment in the wheat germ or a special requirement gene fragment for the cleanliness of the processing workshop in the scene demand. The gene interaction layer adopts the feature expansion technology to supplement relevant association information for these single gene fragments, for example, supplements the product stability information that may be affected in the processing process for the trace mineral gene fragment, and supplements the hygiene standard association information with other processing links for the cleanliness requirement gene fragment of the workshop, so that these single gene fragments can establish a new association with the generated fusion gene fragments, and further generate a supplementary gene fragment.

[0089] Step S134: The fusion gene fragment and the supplementary gene fragment are input into the gene sorting unit, and the gene sorting unit arranges the fusion gene fragment and the supplementary gene fragment in sequence based on the conventional execution order of the wheat germ processing link and the functional attributes of the fusion gene fragment and the supplementary gene fragment.

[0090] In this embodiment, after the fusion gene fragment and the supplementary gene fragment are generated, they are input into the gene sorting unit. The gene sorting unit first refers to the conventional execution order of the wheat germ processing link, such as the order of the cleaning, drying, grinding and other links, and combines the functional attributes of each gene fragment, such as which fragments are related to the cleaning link and which fragments are related to the drying link, to preliminarily arrange all the gene fragments in sequence, so as to ensure that the arrangement result conforms to the basic processing flow logic.

[0091] Step S1341: Establish a conventional execution sequence library of wheat germ processing links, which includes the standard execution sequence of the cleaning link, drying link, grinding link, extraction link, and refining link commonly used in wheat germ processing.

[0092] In this embodiment, the conventional execution sequence library is established according to the commonly used wheat germ processing flow in the industry. Among them, the cleaning link is located at the front end, which is used to remove impurities; then the drying link, which reduces the moisture content of the wheat germ; then the grinding link, which breaks it into a certain particle size; then the extraction link, which obtains the target nutrient ingredients; and finally the refining link, which improves the product purity. Each link has a clear sequence definition as a basic reference for gene sorting.

[0093] Step S1342: Use the gene sorting unit to label the functional attributes of the fusion gene fragments and the supplementary gene fragments, determine the type of wheat germ processing link corresponding to each fusion gene fragment and the type of wheat germ processing link corresponding to each supplementary gene fragment, for example, a fusion gene fragment related to the cleaning link is labeled as a gene fragment related to the cleaning link, and a supplementary gene fragment related to the drying link is labeled as a gene fragment related to the drying link.

[0094] In this embodiment, the gene sorting unit analyzes the functional attributes of the fusion gene fragments and the supplementary gene fragments. For example, a certain fusion gene fragment contains parameters and requirements for impurity removal, which is labeled as related to the cleaning link; a certain supplementary gene fragment involves temperature control and moisture evaporation information, which is labeled as related to the drying link. Through the above labeling, the type of processing link corresponding to each gene fragment is determined.

[0095] Step S1343: Based on the conventional execution sequence library, the labeled fusion gene fragments and supplementary gene fragments are preliminarily arranged according to the conventional sequence of the corresponding wheat germ processing link to form a preliminary sorting sequence.

[0096] In this embodiment, according to the order of cleaning, drying, grinding, extraction, and refining in the conventional execution sequence library, the gene fragments labeled as the cleaning link are placed at the front, followed by the gene fragments labeled as the drying link, and then the gene fragments corresponding to the grinding, extraction, and refining links are arranged in turn to form a preliminary sorting sequence, so that the order of the gene fragments is consistent with the basic processing flow.

[0097] Step S1344: The supplementary gene fragments of the same wheat germ processing link type in the preliminary sorting sequence are secondarily sorted according to the functional subdivision attributes of the supplementary gene fragments of the same wheat germ processing link type. In the secondary sorting process, when the supplementary gene fragments of the same wheat germ processing link type have functional conflicts, the application scenario characteristic genes in the scene demand characteristic gene set are used to determine the priority sorting rule.

[0098] In this embodiment, for the gene fragments of the same processing link in the preliminary sorting sequence, such as multiple gene fragments related to the cleaning link, the gene fragments are secondarily sorted according to the functional subdivision attributes. For example, the gene fragments related to screening impurities are arranged first, and then the gene fragments related to air separation are arranged. If the gene fragments of the same link have functional conflicts, such as one fragment requiring rapid cleaning to improve efficiency and another fragment requiring fine cleaning to ensure purity, the application scenario characteristic genes in the scene demand characteristic gene set are referred to. If the application scenario emphasizes product purity, the gene fragments related to fine cleaning are preferentially arranged.

[0099] Step S135: In the sorting process, the gene sorting unit introduces a dynamic adjustment mechanism, and adjusts the sorting result according to the production capacity adaptation characteristic genes and resource constraint characteristic genes in the scene demand characteristic gene set, to finally form a processing link characteristic gene sequence containing ordered gene fragments.

[0100] In this embodiment, the gene sorting unit introduces a dynamic adjustment mechanism on the basis of the preliminary sorting and secondary sorting. According to the production capacity adaptation characteristic genes in the scene demand characteristic gene set, such as the unit time yield requirement, if the yield is required to be high, the order of some link gene fragments can be adjusted to preferentially ensure the gene fragments for efficient processing. According to the resource constraint characteristic genes, such as the device type limitation, if a certain type of device can only be used in a specific link, the positions of the related gene fragments can be adjusted accordingly to ensure that the sorting result meets the actual production conditions, and finally a processing link characteristic gene sequence is formed.

[0101] Step S140: Based on the processing link characteristic gene sequence, the processing link units are combined by the scheme prototype construction layer of the deep scheme evolution model to form multiple processing scheme prototypes.

[0102] In this embodiment, after receiving the processing link characteristic gene sequence, the scheme prototype construction layer of the deep scheme evolution model analyzes the sequence, determines the types of processing link units that need to be combined, and calls the corresponding units from the processing link unit library for adjustment and combination to supplement the connection process, and then multiple processing scheme prototypes with differences are formed.

[0103] Step S141: Analyze the processing link characteristic gene sequence by using the scheme prototype construction layer of the deep scheme evolution model, extract the processing link type identifier in the processing link characteristic gene sequence, determine the processing link unit types that need to be combined, such as cleaning unit, drying unit, and grinding unit.

[0104] In this embodiment, the scheme prototype construction layer analyzes the processing link characteristic gene sequence, identifies the processing link type identifier therein, such as the multiple occurrences of the gene fragment identifier related to cleaning in the sequence, which indicates that the cleaning unit is needed; the occurrence of the identifier related to drying indicates that the drying unit is needed, and so on, to determine all the processing link unit types that need to be combined.

[0105] Step S142: According to the processing link type identifier, call the processing link unit library, and extract the standard processing link unit of the corresponding type, each of which contains the basic operation process and the conventional parameter range.

[0106] In this embodiment, the processing link unit library stores various standard processing link units. According to the determined processing link unit types, such as cleaning unit and drying unit, the corresponding standard units are extracted from the library. Each standard cleaning unit contains the basic operation processes such as screening, air separation, and magnetic separation, as well as the conventional parameter ranges such as screening mesh size and air separation wind speed; the standard drying unit contains the basic process of hot air drying and the conventional parameter ranges such as temperature and time.

[0107] Step S143: Map the gene fragment information in the processing link characteristic gene sequence to the corresponding standard processing link unit, adjust the basic operation process and the conventional parameter range of the corresponding standard processing link unit, make the attributes of the adjusted processing link unit consistent with the gene fragment information in the processing link characteristic gene sequence, and generate a customized processing link unit.

[0108] In this embodiment, the gene fragment information in the processing link characteristic gene sequence is mapped to the corresponding standard processing link unit. For example, the information about the impurity type and removal requirement in the gene fragment related to the cleaning link is mapped to the standard cleaning unit, the cleaning operation step for specific impurities is added, and the parameter range of the screening mesh is adjusted; the temperature requirement and moisture control information in the gene fragment related to the drying link is mapped to the standard drying unit, the parameter range of the drying temperature and the holding time in the operation process are modified, and a customized processing link unit that meets the gene fragment information is generated.

[0109] For example, step S1431: The scheme prototype construction layer of the deep scheme evolution model establishes a mapping relationship table between the gene fragment information in the processing link characteristic gene sequence and the processing link unit attributes, which is used to limit the processing link unit attribute category corresponding to each gene fragment category.

[0110] In the present embodiment, the mapping relationship table explicitly shows the correspondence between different gene fragment categories and the unit attribute categories of the processing link. For example, the gene fragment category related to impurities corresponds to the impurity removal attribute category of the cleaning unit; the gene fragment category related to moisture corresponds to the moisture control attribute category of the drying unit; the gene fragment category related to particle size corresponds to the particle size adjustment attribute category of the grinding unit, etc. Through the table, it is ensured that the gene fragment information can be accurately mapped to the corresponding attributes of the processing link unit.

[0111] Step S1432: Based on the mapping relationship table, each gene fragment information in the processing link characteristic gene sequence is assigned to the attribute field of the corresponding processing link unit.

[0112] In the present embodiment, according to the mapping relationship table, each gene fragment information in the processing link characteristic gene sequence is assigned to the attribute field of the corresponding processing link unit. For example, the gene fragment information “need to remove impurities with a diameter greater than a certain value” is assigned to the “impurity size screening” attribute field of the cleaning unit; the gene fragment information “the moisture content after drying needs to be lower than a certain value” is assigned to the “final moisture content” attribute field of the drying unit.

[0113] Step S1433: The basic operation process of the processing link unit is adjusted, and the operation steps in the basic operation process are added or deleted according to the functional requirements in the gene fragment information in the processing link characteristic gene sequence.

[0114] In the present embodiment, according to the functional requirements of the gene fragment information assigned to the attribute field, the basic operation process of the processing link unit is adjusted. For example, if the gene fragment information of the cleaning unit requires the removal of magnetic impurities, and the standard cleaning unit does not have a magnetic separation step in the basic operation process, a magnetic separation step is added to the basic operation process; if the gene fragment information of the drying unit indicates that no secondary drying is required, and the standard process has a secondary drying step, the step is deleted.

[0115] Step S1434: The range of the conventional parameters of the processing link unit is adjusted, the quantitative parameters in the gene fragment information in the processing link characteristic gene sequence are taken as the reference values of the adjusted conventional parameter range, and the upper and lower limits of the adjusted conventional parameter range are determined according to the fluctuation information in the gene fragment information in the processing link characteristic gene sequence.

[0116] In the present embodiment, when adjusting the range of the conventional parameters of the processing link unit, the quantitative parameters in the gene fragment information are taken as the reference values. For example, the quantitative parameter in the gene fragment information of the grinding unit is a certain particle size value, which is taken as the reference value of the grinding particle size; the fluctuation information in the gene fragment information is to allow a certain range of fluctuation, and according to the fluctuation information, the upper and lower limits of the adjusted grinding particle size parameter range are determined, so that the parameter range can cover a reasonable fluctuation interval.

[0117] Step S1435: After the adjustment is completed, the adjusted operation process of the processing link unit is associated with the adjusted parameter range for verification. After the verification is passed, the customized processing link unit is generated.

[0118] In this embodiment, the association verification is to check whether the adjusted operation process and the parameter range match each other without contradiction. For example, in the cleaning unit, if a magnetic separation step is added, it needs to be verified whether there is a corresponding magnetic separation intensity parameter range; in the drying unit, if the drying temperature range is adjusted, it needs to be verified whether the temperature range is suitable for the parameter range of the drying time to ensure that the expected drying effect can be achieved within the set time and in the temperature range. After the verification is passed, the customized processing link unit is generated.

[0119] Step S144: According to the sorting order of the processing link characteristic gene sequence, the customized processing link units are serially combined, and the connection processes between adjacent customized processing link units are supplemented to form a preliminary scheme framework.

[0120] In this embodiment, according to the sorting order of cleaning, drying, grinding, extraction, and refining in the processing link characteristic gene sequence, the corresponding customized processing link units are serially connected. The connection processes between adjacent units are supplemented, such as the connection process between the cleaning unit and the drying unit, which includes the transportation mode and speed control of the cleaned wheat germ to the drying equipment; the connection process between the drying unit and the grinding unit, which includes temperature detection and moisture content sampling of the dried wheat germ to ensure that its state meets the feeding requirements of the grinding unit, and the anti-caking treatment measures during the transportation process are set.

[0121] The connection process between the grinding unit and the extraction unit covers the particle size screening of the grinding product, which is filtered through a screen with a specific aperture to remove large particles that do not meet the extraction requirements, and the material amount after screening is recorded as the feeding reference of the extraction unit.

[0122] The connection process between the extraction unit and the refining unit includes preliminary detection of the purity of the extraction liquid, qualitative analysis of the impurity content in the extraction liquid by using a rapid test paper, and adjustment of the pressure and flow of the conveying pipeline to ensure that the extraction liquid enters the refining unit stably.

[0123] By supplementing these connection processes, each customized processing link unit forms a continuous and smooth whole to constitute a preliminary scheme framework, which clearly defines the input-output relationship of each link and the operation requirements of the intermediate transition.

[0124] Step S145: Adjust the order of the customized processing units in the preliminary scheme framework and the operation parameters of the customized processing units within a limited range to generate multiple different scheme frameworks. Each different scheme framework forms a processing scheme prototype after the complete operation instructions are supplemented.

[0125] In this embodiment, when adjusting the preliminary scheme framework, firstly, the order of the customized processing units is adjusted. The order of some non-critical units is fine-tuned while keeping the main processing unit type unchanged. For example, some refining pretreatment steps after the extraction unit are moved to the extraction unit to form an adjusted scheme framework. Or a short cooling unit is added after the drying unit to enter the grinding unit to form another adjusted scheme framework.

[0126] For the operation parameters of the customized processing units, the parameters are adjusted within a limited range based on the original parameters. For example, the screening frequency of the cleaning unit is set to fluctuate within a certain percentage of the original frequency, the temperature of the drying unit is adjusted within a small range, and the speed of the grinding unit is fine-tuned according to the characteristics of the material.

[0127] Through these adjustments, multiple different scheme frameworks are generated. For each scheme framework, complete operation instructions are supplemented, including specific operation steps of each unit, qualification requirements of the operator, maintenance points of the equipment, and handling plans for abnormal situations, etc. For example, in the scheme framework containing a cooling unit, the start-up procedure of the cooling equipment, the control method of the cooling time, and the detection standard of the cooling effect are specified. After the supplementary instructions, each scheme framework forms an independent processing scheme prototype.

[0128] Step S150: Double-cycle adaptive optimization is performed on the multiple processing scheme prototypes and the two types of characteristic gene sets, i.e., the wheat germ characteristic gene set and the scene demand characteristic gene set, to obtain a target processing recommendation scheme, which is transmitted to the processing execution terminal.

[0129] In this embodiment, the multiple generated processing scheme prototypes are subjected to double-cycle adaptive optimization. The first optimization cycle focuses on the adaptability of the processing scheme prototype to the wheat germ characteristic gene set, and the second optimization cycle focuses on the adaptability to the scene demand characteristic gene set. Through repeated cycle adjustment, the adaptability of the scheme is improved, and finally the target processing recommendation scheme is determined and transmitted to the processing execution terminal.

[0130] Step S151: In the first round of optimization cycles, each processing scheme prototype is subjected to fitness analysis with the wheat germ characteristic gene set, and the processing links that do not match the gene fragments in the wheat germ characteristic gene set are extracted from each processing scheme prototype. The operation parameters and operation processes of the processing links that do not match the gene fragments in the wheat germ characteristic gene set are adjusted, so that the fitness of each processing scheme prototype with the wheat germ characteristic gene set is improved.

[0131] In this embodiment, in the first round of optimization cycles, each processing scheme prototype is subjected to fitness analysis with the wheat germ characteristic gene set. Each processing link in the processing scheme prototype is compared with the gene fragments in the wheat germ characteristic gene set one by one, and the unmatched parts are identified. For example, the temperature parameter of the drying link in a certain processing scheme prototype does not match the heat resistance information of the variety characteristic gene fragment in the wheat germ characteristic gene set. The heat resistance of this variety is relatively low, and the current drying temperature is too high.

[0132] For such unmatched processing links, their operation parameters and operation processes are adjusted. For the above drying link, the drying temperature is reduced, and the drying time is prolonged to ensure that the drying effect is achieved without damaging the nutritional ingredients of the wheat germ. In addition, the air flow speed of the drying equipment is adjusted to match the microstructure characteristics of the wheat germ, so as to avoid material loss caused by excessive air flow. Through these adjustments, the fitness of each processing scheme prototype with the wheat germ characteristic gene set is improved.

[0133] Step S152: After completing the first round of optimization cycles, enter the second round of optimization cycles. The optimized processing scheme prototype obtained after the first round of optimization cycles is subjected to fitness analysis with the scene demand characteristic gene set, and the processing links in the optimized processing scheme prototype that do not meet the requirements of the scene demand characteristic gene set are identified. The order and operation rules of the processing links that do not meet the requirements of the scene demand characteristic gene set are adjusted based on the gene fragments in the scene demand characteristic gene set.

[0134] In this embodiment, after completing the first round of optimization cycles, enter the second round of optimization cycles. The optimized processing scheme prototype is subjected to fitness analysis with the scene demand characteristic gene set, and the processing links that do not meet the scene demand are found out. For example, the unit time yield of a certain optimized processing scheme prototype is lower than the requirement of the production capacity adaptation characteristic gene fragment in the scene demand characteristic gene set, and cannot meet the daily planned production capacity.

[0135] Based on the gene fragments in the scene demand characteristic gene set, the order and operation rules of the processing link that does not meet the requirements are adjusted. For the above insufficient production situation, it is found that the low efficiency of the grinding link leads to it, so the connection order of the grinding link and the previous drying link is adjusted to reduce the intermediate waiting time, and the operation rules of the grinding equipment are modified to improve the continuity of grinding. In addition, according to the requirement of the product purity in the application scene characteristic gene fragment, the filtering operation rules of the extraction link are strengthened, and the filtering times are increased to ensure that the product quality meets the standard of high-end nutritional supplements.

[0136] Step S153: Between the first round of optimization cycles and the second round of optimization cycles, the parameter information of the adjusted processing link in the first round of optimization cycles is fed back to the analysis process of the scene demand characteristic gene set to correct the correlation between the gene fragments in the scene demand characteristic gene set.

[0137] In this embodiment, between the first round of optimization cycles and the second round of optimization cycles, the parameter information of the processing link adjusted in the first round of optimization cycles is collected, such as the adjusted drying temperature, grinding speed, etc. These parameter information is fed back to the analysis process of the scene demand characteristic gene set to re-examine the correlation between the gene fragments in the scene demand characteristic gene set.

[0138] For example, the drying temperature is reduced in the first round of optimization cycles, which may affect the time arrangement of the subsequent link and thus affect the production capacity. Therefore, according to the adjusted drying time, the correlation between the production capacity adaptation characteristic gene fragment and the device running time in the resource constraint characteristic gene fragment in the scene demand characteristic gene set is corrected to ensure that the scene demand characteristic gene set can more accurately reflect the constraints and demands in the actual processing process.

[0139] Step S154: Repeat the first round of optimization cycles and the second round of optimization cycles, and calculate the comprehensive adaptation value of each processing scheme prototype and the two types of characteristic gene sets of the wheat germ characteristic gene set and the scene demand characteristic gene set after each execution of the first round of optimization cycles and the second round of optimization cycles; the comprehensive adaptation value is determined by the following way: first, normalize the adaptation degree of each processing scheme prototype to the same evaluation scale with the wheat germ characteristic gene set and the scene demand characteristic gene set, and then perform weighted calculation based on the normalized results.

[0140] In this embodiment, the first and second rounds of optimization cycles are repeatedly executed. After each cycle, the comprehensive adaptation value of each processing scheme prototype is calculated. When calculating, the adaptation degree to the wheat germ characteristic gene set and the adaptation degree to the scene demand characteristic gene set are normalized respectively to the same evaluation scale, such as converted to values between 0 and 1.

[0141] Then, weights are set according to the importance of the two, for example, for the scene of making high-end nutritional supplements, the weight of the adaptation degree of the scene requirement can be slightly higher than that of the wheat germ characteristics. The two normalized adaptation degrees are multiplied by the corresponding weights, and the results are spliced to form a comprehensive adaptation value, which contains information of two dimensions, reflecting the adaptation of the two sets of characteristic genes.

[0142] Step S155: When the comprehensive adaptation value reaches the preset adaptation standard and the change of the comprehensive adaptation value after the first and second optimization cycles are performed for two consecutive times is less than the preset threshold, the optimization cycle is stopped, and the processing scheme prototype with the highest comprehensive adaptation value is selected as the target processing recommendation scheme.

[0143] In this embodiment, the preset adaptation standard is that the values of the two dimensions in the comprehensive adaptation value are not less than 0.85, and the preset threshold is 0.03. During the repeated cycle process, the change of the comprehensive adaptation value is continuously monitored. When the comprehensive adaptation value of a certain processing scheme prototype reaches the preset adaptation standard, and the change of the comprehensive adaptation value after two consecutive cycles is less than 0.03, the optimization cycle is stopped.

[0144] The target processing recommendation scheme is selected from all processing scheme prototypes, which has the highest comprehensive adaptation value. The target processing recommendation scheme can meet the characteristics of the wheat germ itself and well fit the requirements of the processing scene, such as meeting the yield standard, product quality meeting the standard, reasonable utilization of equipment resources, etc.

[0145] Step S156: The target processing recommendation scheme is transmitted to the processing execution terminal.

[0146] In this embodiment, the determined target processing recommendation scheme is format-converted to meet the data receiving requirements of the processing execution terminal. The converted scheme contains detailed parameters, operation processes, equipment numbers, time nodes, etc. of each processing link. The scheme data is sent to the processing execution terminal through an encrypted transmission protocol, and a data verification mechanism is used in the transmission process to ensure the integrity and accuracy of the data. After receiving, the processing execution terminal decrypts and analyzes the scheme to provide visual operation guidance for the operator, guiding the actual wheat germ processing process.

[0147] The method can further include the step of pre-training the deep scheme evolution model.

[0148] In this embodiment, the deep scheme evolution model needs to be pre-trained before being applied. The pre-training process uses historical processing data, including multi-dimensional basic data of past processed wheat germ, corresponding processing scene requirement parameters, and final adopted processing scheme and scheme implementation effect data, etc.

[0149] Step S211: Collect historical processing data, clean and preprocess the historical processing data, remove abnormal data and duplicate data, and supplement missing data.

[0150] In this embodiment, the historical data of wheat germ processing in the past three years is collected, which is stored in the database of the enterprise. The data is cleaned, and abnormal data is identified through statistical analysis, such as processing parameters that are obviously beyond the reasonable range, yield records that do not match the actual situation, etc. These abnormal data are excluded.

[0151] For duplicate data, one complete record can be retained. For missing data, reasonable inference and supplementation are made based on data under the same batch or similar conditions, for example, part of the quality detection data of a batch of wheat germ is missing, which can be supplemented by referring to other batches of data under the same variety and planting conditions. The preprocessed historical data forms a structured data set, which is convenient for model training and use.

[0152] Step S212: Divide the preprocessed historical processing data into training set, validation set and test set, wherein the training set is used for learning model parameters, the validation set is used for adjusting the hyperparameters of the model, and the test set is used for evaluating the generalization ability of the model.

[0153] In this embodiment, the preprocessed historical processing data is divided into training set, validation set and test set according to the ratio of 7:2:1. The training set contains a large amount of historical data of different varieties and different processing scenes, which is used to learn the correlation between features and the generation rule of processing scheme.

[0154] The validation set is used to evaluate the performance of the model during the training process, and the hyperparameters of the model, such as learning rate, network layer number, hidden unit number, etc. are adjusted according to the validation result to optimize the fitting effect of the model. The test set is used to evaluate the performance of the model on unseen data after the training of the model is completed, to test the generalization ability of the model, and to ensure that the model can be applied to new processing scenes.

[0155] Step S213: Construct the network structure of the deep scheme evolution model, including the gene interaction layer and the scheme prototype construction layer, the gene interaction layer adopts the bidirectional long short-term memory network structure, and the scheme prototype construction layer adopts the convolutional neural network structure.

[0156] In this embodiment, the network structure of the deep scheme evolution model is constructed. The gene interaction layer adopts the bidirectional long short-term memory network structure, which can process sequence data and capture the long-term dependence and bidirectional association information of gene fragments in the feature gene set of wheat germ and the feature gene set of scene demand.

[0157] The bidirectional long short-term memory network comprises two hidden layers in forward and backward directions, each of which is composed of a plurality of memory units, and the input, forgetting and output of information are controlled through a gating mechanism, effectively solving the long sequence dependency problem. The scheme prototype construction layer adopts a convolutional neural network structure, which extracts local features in the process link gene sequence through a convolution kernel, and performs feature dimension reduction and abstraction through a pooling operation, and can efficiently learn the combination mode of the process link unit from the sequence.

[0158] Step S214: setting the hyperparameters of model training, including learning rate, training batch size, training iteration number, regularization coefficient, etc.

[0159] In this embodiment, the hyperparameters of model training are set. The learning rate is initially set to a small value so that the model can learn stably. As the training progresses, the learning rate is gradually reduced using a learning rate decay strategy to improve the convergence accuracy of the model.

[0160] The training batch size is determined according to the computer hardware performance and the size of the data set to ensure efficient use of computing resources in each training. The training iteration number is set to a large value, and an early stopping mechanism is set. When the loss function value on the validation set does not decrease continuously for multiple iterations, the training is stopped to avoid model overfitting. The regularization coefficient is used to control the complexity of the model to prevent overfitting, which is realized by adding a regularization term to the loss function.

[0161] Step S215: input the training set into the deep scheme evolution model, use the back propagation algorithm for model training, and adjust the hyperparameters through the validation set until the loss function value of the model on the validation set reaches the preset threshold.

[0162] In this embodiment, the training set is input into the deep scheme evolution model for training. In each training batch, the model processes the input gene set to generate the predicted result of the process scheme prototype. The predicted result is compared with the actual process scheme in the historical data to calculate the loss function value, which reflects the difference between the predicted scheme and the actual scheme.

[0163] The back propagation algorithm is used to adjust the weight parameters of each layer of the model according to the loss function value, and the error is propagated from the output layer to the input layer to continuously optimize the parameters. During the training process, the model performance is evaluated regularly using the validation set, and the hyperparameters are adjusted according to the validation results, such as appropriately reducing the learning rate or increasing the regularization coefficient when the validation loss increases. Continue training until the loss function value of the model on the validation set reaches the preset threshold, at which point the prediction effect of the model is relatively ideal.

[0164] Step S216: using the test set to evaluate the trained deep scheme evolution model, calculating the evaluation indexes such as the accuracy, recall rate and F1 value of the model, if the evaluation indexes reach the preset standard, the model training is completed; otherwise, the model structure and hyperparameters are adjusted, and the training is performed again.

[0165] In this embodiment, the trained model is evaluated using the test set. The wheat germ feature gene set and the scene demand feature gene set in the test set are input into the model to obtain the processing scheme prototypes generated by the model. These scheme prototypes are compared with the actual processing schemes in the test set to calculate evaluation indexes such as the accuracy, recall rate and F1 value.

[0166] The accuracy reflects the proportion of correct schemes generated by the model in all generated schemes, the recall rate reflects the proportion of actual correct schemes generated by the model, and the F1 value is a comprehensive index of the accuracy and the recall rate. If the evaluation indexes reach the preset standard, for example, the accuracy and the recall rate are not less than 0.8, and the F1 value is not less than 0.8, the model training is completed. If the standard is not reached, the reasons are analyzed, the model structure is adjusted, for example, the network layer is increased or the convolution kernel size is changed, and the hyperparameters are adjusted, and the training is performed again until the model performance reaches the standard.

[0167] Figure 2 The hardware structure of the deep learning combined wheat germ processing scheme recommendation system 100 for implementing the deep learning combined wheat germ processing scheme recommendation method is shown in the figure, which is used to implement the deep learning combined wheat germ processing scheme recommendation method. Figure 2 As shown in the figure, the deep learning combined wheat germ processing scheme recommendation system 100 can include a processor 110, a machine readable storage medium 120, a bus 130 and a communication unit 140.

[0168] In the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine readable storage medium 120, so that the processor 110 can execute the deep learning combined wheat germ processing scheme recommendation method of the above method embodiment, the processor 110, the machine readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiving action of the communication unit 140.

[0169] The specific implementation process of the processor 110 can refer to the above various method embodiments executed by the deep learning combined wheat germ processing scheme recommendation system 100, which has similar implementation principles and technical effects, and will not be described here.

[0170] In addition, the embodiment of the present application also provides a readable storage medium, wherein computer executable instructions are set in the readable storage medium, and when the processor runs the computer executable instructions, the deep learning combined wheat germ processing scheme recommendation method is realized.

[0171] It should be noted that the foregoing description of embodiments of the application has been presented for the purposes of simplicity and clarity. It is not intended to be an exhaustive description of all aspects of the embodiments of the application.

Claims

1. A method for recommending wheat germ processing schemes using deep learning, characterized in that, The method includes: The process involves obtaining multi-dimensional basic data on wheat germ and processing scenario requirements. The multi-dimensional basic data on wheat germ includes varietal characteristic data, raw material quality data, and microstructure image data. The processing scenario requirements include application scenario data of processed products, processing capacity matching data, and processing resource constraint data. Feature gene extraction was performed on multi-dimensional basic data of wheat germ and processing scenario requirement parameters to obtain a set of wheat germ feature genes and a set of scenario requirement feature genes. The set of wheat germ characteristic genes and the set of scenario requirement characteristic genes are input into a pre-trained deep solution evolution model. The gene interaction layer of the deep solution evolution model dynamically recombines the set of wheat germ characteristic genes and the set of scenario requirement characteristic genes to generate the characteristic gene sequence of the processing link. Based on the characteristic gene sequences of the processing links, multiple processing link units are combined through the scheme prototype construction layer of the deep scheme evolution model to form multiple processing scheme prototypes. A dual-loop adaptation optimization is performed on multiple processing scheme prototypes and wheat germ characteristic gene sets and scenario requirement characteristic gene sets to obtain the target processing recommendation scheme, which is then transmitted to the processing execution terminal. The process involves inputting the wheat germ characteristic gene set and the scenario requirement characteristic gene set into a pre-trained deep solution evolution model. Through the gene interaction layer of the deep solution evolution model, the wheat germ characteristic gene set and the scenario requirement characteristic gene set are dynamically recombinated to generate processing stage characteristic gene sequences, including: The set of wheat germ characteristic genes and the set of scenario requirement characteristic genes are input into the gene interaction layer of the deep solution evolution model. The gene interaction layer first performs similarity matching on gene fragments in the wheat germ characteristic gene set and gene fragments in the scenario requirement characteristic gene set to determine gene fragment pairs with corresponding associations. Based on the association strength value of gene fragment pairs, the gene interaction layer calls the pre-stored gene recombination rules to perform information fusion on gene fragment pairs with corresponding associations, generating fused gene fragments. The fused gene fragments contain the core information of gene fragments corresponding to the wheat germ feature gene set and the core information of gene fragments corresponding to the scenario requirement feature gene set in the gene fragment pair. For individual gene fragments that have not formed corresponding associations, the gene interaction layer supplements the association information through feature expansion technology, enabling individual gene fragments to form new associations with other fused gene fragments and generate supplementary gene fragments; The fusion gene fragment and the supplementary gene fragment are input into the gene sorting unit. The gene sorting unit arranges the fusion gene fragment and the supplementary gene fragment in sequence based on the conventional execution order of wheat germ processing and the functional attributes of the fusion gene fragment and the supplementary gene fragment. During the sorting process, the gene sorting unit introduces a dynamic adjustment mechanism to adjust the sorting results according to the capacity adaptation feature genes and resource constraint feature genes in the feature gene set of scenario requirements, and finally form a processing link feature gene sequence containing ordered gene fragments. The process, based on the characteristic gene sequences of processing steps, combines processing step units through a deep scheme evolution model's scheme prototype construction layer to form multiple processing scheme prototypes, including: The feature gene sequences of processing links are analyzed by constructing the prototype layer of the deep scheme evolution model, and the processing link type identifiers in the feature gene sequences of processing links are extracted to determine the types of processing link units that need to be combined. Based on the processing step type identifier, call the processing step unit library and extract the standard processing step unit of the corresponding type. Each standard processing step unit contains the basic operation process and the range of common parameters. The gene fragment information in the characteristic gene sequence of the processing link is mapped to the corresponding standard processing link unit. The basic operation process and conventional parameter range of the corresponding standard processing link unit are adjusted so that the attributes of the adjusted processing link unit are consistent with the gene fragment information in the characteristic gene sequence of the processing link, thereby generating a customized processing link unit. Based on the sorting order of characteristic gene sequences in the processing steps, the customized processing steps are connected in series and combined to supplement the connection process between adjacent customized processing steps, forming a preliminary scheme framework. The sequence of customized processing units and the operation parameters of customized processing units in the initial scheme framework are adjusted to a limited extent to generate multiple scheme frameworks with differences. Each scheme framework with differences is supplemented with complete operation instructions to form a processing scheme prototype.

2. The method for recommending wheat germ processing schemes using deep learning as described in claim 1, characterized in that, The process involves extracting feature genes from multi-dimensional basic data of wheat germ and processing scenario requirement parameters to obtain a set of wheat germ feature genes and a set of scenario requirement feature genes, including: Variety characteristic data were extracted from multi-dimensional basic data of wheat germ. The variety characteristic data were structured and analyzed to separate variety origin association information, variety growth cycle association information, and variety resistance association information. The variety origin association information, variety growth cycle association information, and variety resistance association information were converted into quantifiable variety characteristic gene fragments. Each variety characteristic gene fragment corresponds to the core representation of a variety attribute. Raw material quality data is extracted from multi-dimensional basic data of wheat germ. The composition information, freshness correlation information, and impurity content correlation information in the raw material quality data are analyzed. The composition information, freshness correlation information, and impurity content correlation information are decomposed into quality characteristic gene fragments. Each quality characteristic gene fragment contains a specific characterization parameter of a quality indicator. Microstructure image data is extracted from multidimensional basic data of wheat germ. Cell structure distribution information, tissue structure density information, and surface morphology correlation information in the microstructure image are obtained through image feature analysis technology. Cell structure distribution information, tissue structure density information, and surface morphology correlation information are converted into structural feature gene fragments. Each structural feature gene fragment corresponds to a quantitative description of a microstructure attribute. The varietal characteristic gene fragments, quality characteristic gene fragments, and structural characteristic gene fragments are associated and integrated, and the interaction relationships between the varietal characteristic gene fragments, quality characteristic gene fragments, and structural characteristic gene fragments are marked to form a set of wheat germ characteristic genes containing a list of gene fragments and their interaction relationships. The application scenario data, processing capacity adaptation data, and processing resource constraint data in the processing scenario demand parameters are analyzed. The application scenario data is converted into application scenario feature gene fragments, the processing capacity adaptation data is converted into capacity adaptation feature gene fragments, and the processing resource constraint data is converted into resource constraint feature gene fragments. The application scenario feature gene fragments, capacity adaptation feature gene fragments, and resource constraint feature gene fragments are associated to form a scenario demand feature gene set. The gene fragment structure of the scenario demand feature gene set is adapted to the gene fragment structure of the wheat germ feature gene set.

3. The method for recommending wheat germ processing schemes using deep learning according to claim 2, characterized in that, The process involves extracting microstructural image data from multidimensional basic data of wheat germ, and using image feature analysis technology to obtain cell structure distribution information, tissue structure density information, and surface morphology correlation information from the microstructural images. This cell structure distribution information, tissue structure density information, and surface morphology correlation information are then converted into structural feature gene fragments, including: The image data of wheat germ microstructure is input into the image feature analysis module to divide the microstructure image into regions, separating the cellular region, interstitial region, and surface coverage region. Pixel features are extracted from the cell region, and the gray-level distribution information, edge contour information, and texture direction information of the pixels within the cell region are calculated. Based on the gray-level distribution information, edge contour information, and texture direction information of the pixels within the cell region, the arrangement of cell structures, cell size correlation information, and cell integrity correlation information are determined to form cell structure distribution information. Density analysis is performed on the interstitial region to calculate the number of interstitial spaces, the size distribution of interstitial spaces, and the connectivity of interstitial spaces per unit area. Based on the number of interstitial spaces, the size distribution of interstitial spaces, and the connectivity of interstitial spaces per unit area, tissue structure density information is generated. The tissue structure density information can characterize the compactness of wheat germ tissue. Morphological features are extracted from the surface coverage area to obtain information on surface undulation, smoothness correlation, and defect distribution. These three information together constitute surface morphological correlation information. Cell structure distribution information, tissue structure density information, and surface morphology association information are encoded into gene fragments respectively. Cell structure distribution information corresponds to a set of structural feature gene fragments, tissue structure density information corresponds to a set of structural feature gene fragments, and surface morphology association information corresponds to a set of structural feature gene fragments. During the encoding process, the quantitative parameters and association relationships of cell structure distribution information, tissue structure density information, and surface morphology association information are preserved.

4. The method for recommending wheat germ processing schemes using deep learning according to claim 2, characterized in that, The analysis of application scenario data, processing capacity adaptation data, and processing resource constraint data in the processing scenario demand parameters involves converting the application scenario data into application scenario feature gene fragments, the processing capacity adaptation data into capacity adaptation feature gene fragments, and the processing resource constraint data into resource constraint feature gene fragments. These application scenario feature gene fragments, capacity adaptation feature gene fragments, and resource constraint feature gene fragments are then associated to form a scenario demand feature gene set, including: The application scenario data in the processing scenario requirement parameters are analyzed to determine the target use type of the processed product, the target user group association information, and the product quality level association requirements. The target use type of the processed product is decomposed into use feature sub-items, the target user group association information is decomposed into user feature sub-items, and the product quality level association requirements are decomposed into quality feature sub-items. The use feature sub-items correspond to a set of application scenario feature gene fragments, the user feature sub-items correspond to a set of application scenario feature gene fragments, and the quality feature sub-items correspond to a set of application scenario feature gene fragments. The processing capacity adaptation data in the processing scenario demand parameters is analyzed, and the unit time output requirement, continuous production duration requirement, and batch production scale requirement of the processing process are extracted. The unit time output requirement of the processing process is converted into output feature sub-items, the continuous production duration requirement is converted into duration feature sub-items, and the batch production scale requirement is converted into scale feature sub-items. The output feature sub-items correspond to a set of capacity adaptation feature gene fragments, the duration feature sub-items correspond to a set of capacity adaptation feature gene fragments, and the scale feature sub-items correspond to a set of capacity adaptation feature gene fragments. The processing resource constraint data in the processing scenario requirement parameters are analyzed to obtain the type restrictions of the equipment required for processing, the stability requirements of raw material supply, and the energy consumption control requirements. The type restrictions of the equipment required for processing are broken down into equipment feature sub-items, the stability requirements of raw material supply are broken down into raw material feature sub-items, and the energy consumption control requirements are broken down into energy consumption feature sub-items. Each equipment feature sub-item corresponds to a set of resource constraint feature gene fragments, each raw material feature sub-item corresponds to a set of resource constraint feature gene fragments, and each energy consumption feature sub-item corresponds to a set of resource constraint feature gene fragments. Associative markers are applied to application scenario characteristic gene fragments, capacity matching characteristic gene fragments, and resource constraint characteristic gene fragments to determine the mutual influence relationships between application scenario characteristic gene fragments and capacity matching characteristic gene fragments, between application scenario characteristic gene fragments and resource constraint characteristic gene fragments, and between capacity matching characteristic gene fragments and resource constraint characteristic gene fragments. Based on the association labeling results, the application scenario feature gene fragments, capacity matching feature gene fragments, and resource constraint feature gene fragments are sorted and integrated according to their association relationships to form a scenario demand feature gene set. Each gene fragment in the scenario demand feature gene set contains corresponding demand sub-item information and associated gene fragment identifiers.

5. The method for recommending wheat germ processing schemes using deep learning according to claim 1, characterized in that, Based on the association strength value of gene fragment pairs, the gene interaction layer calls pre-stored gene recombination rules to perform information fusion on gene fragment pairs with corresponding associations, generating fused gene fragments, including: The association strength value of each gene fragment pair is calculated using the gene interaction layer. The association strength value is determined based on the attribute similarity, functional matching degree, and demand adaptability of the gene fragment corresponding to the wheat germ feature gene set and the gene fragment corresponding to the scene demand feature gene set in the gene fragment pair. Based on the preset range of association strength values, the gene interaction layer calls the corresponding gene recombination rules. Gene fragment pairs whose association strength values ​​meet the first preset range call the deep fusion rule, gene fragment pairs whose association strength values ​​meet the second preset range call the medium fusion rule, and gene fragment pairs whose association strength values ​​meet the third preset range call the basic fusion rule. For gene fragment pairs that invoke deep fusion rules, the gene interaction layer performs parameter-by-parameter fusion of all attribute parameters of the gene fragment corresponding to the wheat germ feature gene set in the gene fragment pair with all attribute parameters of the gene fragment corresponding to the scene requirement feature gene set, calculates the average value of the parameters after parameter-by-parameter fusion and retains the parameter fluctuation range before parameter-by-parameter fusion, and generates a deep fusion gene fragment containing complete parameter information. For gene fragment pairs that invoke the moderate fusion rule, the gene interaction layer selects the core attribute parameters of the gene fragment corresponding to the wheat germ feature gene set and the core attribute parameters of the gene fragment corresponding to the scene requirement feature gene set in the gene fragment pair and fuses them. It retains the original values ​​of the non-core parameters of the gene fragment corresponding to the wheat germ feature gene set and the original values ​​of the non-core parameters of the gene fragment corresponding to the scene requirement feature gene set in the gene fragment pair, and generates a moderately fused gene fragment containing the core fusion parameters and the original non-core parameters. For gene fragment pairs that invoke the basic fusion rules, the gene interaction layer only establishes the association identifier between the gene fragment corresponding to the wheat germ feature gene set and the gene fragment corresponding to the scene requirement feature gene set in the gene fragment pair, retains the complete parameter information of the gene fragment corresponding to the wheat germ feature gene set and the complete parameter information of the gene fragment corresponding to the scene requirement feature gene set in the gene fragment pair, and generates a basic fusion gene fragment with the association identifier.

6. The method for recommending wheat germ processing schemes using deep learning according to claim 1, characterized in that, The process involves inputting the fusion gene fragment and the supplementary gene fragment into a gene sequencing unit. The gene sequencing unit, based on the conventional execution sequence of wheat germ processing and the functional attributes of the fusion gene fragment and the supplementary gene fragment, sequentially arranges the fusion gene fragment and the supplementary gene fragment, including: Establish a standard execution sequence library for wheat germ processing, which includes the standard execution sequences for common wheat germ processing steps such as cleaning, drying, grinding, extraction, and refining. The gene sequencing unit is used to label the functional attributes of the fusion gene fragment and the supplementary gene fragment to determine the wheat germ processing stage type corresponding to each fusion gene fragment and the wheat germ processing stage type corresponding to each supplementary gene fragment. Based on the aforementioned conventional execution sequence library, the labeled fusion gene fragments and supplementary gene fragments are initially arranged according to the conventional order of the corresponding wheat germ processing steps to form a preliminary sorting sequence; For fusion gene fragments and supplementary gene fragments of the same wheat germ processing stage type in the preliminary sorting sequence, a secondary sorting is performed based on the functional subdivision attributes of the fusion gene fragments and supplementary gene fragments of the same wheat germ processing stage type. During the secondary sorting process, when there is a functional conflict between the fusion gene fragments and supplementary gene fragments of the same wheat germ processing stage type, the priority sorting rule is determined based on the application scenario feature genes in the scenario requirement feature gene set.

7. The method for recommending wheat germ processing schemes using deep learning according to claim 1, characterized in that, The process involves a double-loop adaptation optimization of multiple processing scheme prototypes with wheat germ characteristic gene sets and scenario requirement characteristic gene sets to obtain a target processing recommendation scheme, including: In the first round of optimization, each processing scheme prototype is matched with the wheat germ characteristic gene set. Processing steps that do not match the gene fragments in the wheat germ characteristic gene set are extracted from each processing scheme prototype. The operation parameters and operation process of the processing steps that do not match the gene fragments in the wheat germ characteristic gene set are adjusted to improve the matching degree between each processing scheme prototype and the wheat germ characteristic gene set. After completing the first round of optimization, the second round of optimization is entered. The optimized processing scheme prototype obtained from the first round of optimization is matched with the set of scene requirement feature genes. The processing links in the optimized processing scheme prototype obtained from the first round of optimization that do not meet the requirements of the set of scene requirement feature genes are identified. The order and operation rules of the processing links that do not meet the requirements of the set of scene requirement feature genes are adjusted based on the gene fragments in the set of scene requirement feature genes. Between the first and second optimization cycles, the parameter information of the processing steps adjusted in the first optimization cycle is fed back to the parsing process of the scene requirement feature gene set to correct the correlation between gene fragments in the scene requirement feature gene set. The first and second optimization cycles are repeated. After each cycle, the comprehensive fit value of each processing scheme prototype with the wheat germ characteristic gene set and the scenario requirement characteristic gene set is calculated. The comprehensive fit value is determined as follows: First, the fit degree of each processing scheme prototype with the wheat germ characteristic gene set and with the scenario requirement characteristic gene set is normalized to the same evaluation scale. Then, a weighted calculation is performed based on the normalized results. When the overall fit value reaches the preset fit standard and the change in the overall fit value after two consecutive executions of the first and second optimization cycles is less than the preset threshold, the optimization cycle is stopped, and the processing scheme prototype with the highest overall fit value is selected as the target processing recommendation scheme.

8. A wheat germ processing scheme recommendation system combining deep learning, characterized in that, The wheat germ processing scheme recommendation system combining deep learning includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to run the programs, instructions or code in the memory to implement the wheat germ processing scheme recommendation method combining deep learning as described in any one of claims 1-7.

Images