A method and system for screening processed drugs of Quzhi Fructus based on data mining
Through data mining technology, the multi-classifier model is constructed and the preparation method of Qu Zhi aurantia is automatically screened, which solves the problem of lack of objectivity in the screening process in the existing technology, and improves the accuracy and quality control of drug screening.
Patent Information
- Application Number
- CN202410525528.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-29
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-04-29
AI Technical Summary
The existing screening process for the preparation of Quzhi aurantium is lacking objective and standardization, which leads to subjective influence of the screening results being subjectively affected by the operator, which may lead to the unqualified drug being retained or the qualified drug being eliminated, affecting the health of the company and patients.
Using a data mining method, we set the preparation method name set, acquisition and feature extraction, and constructed sample images, and constructed multi-classifier models for training and verification, which are used to automatically screen the preparation methods that meet the requirements and reduce manual participation.
It has achieved a more objective and accurate screening of the preparation of the drugs by Quzhi aurantium, reduced artificial errors, improved the quality control of the preparation of the drugs, and reduced the retention of unqualified drugs and the removal of qualified drugs.
Smart Images

Figure CN118380076B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of filtering based on additional data, and in particular, relates to a method and system for screening processed drugs of Quzhi Fructus based on data mining. Background Art
[0002] Quzhou Zhike processed medicine is a kind of Chinese medicine preparation, which is mainly made of Quzhou Zhike as raw material and is processed through a specific processing technology; Quzhou Zhike is a kind of Chinese medicinal material, which has the effects of regulating qi and removing dampness, strengthening the spleen and stomach, stopping diarrhea and clearing rashes; processing is a processing technology in Chinese medicine preparation, which can change the nature and taste characteristics of the medicinal material, increase the efficacy, and improve the texture and preservation of the medicine; for Qu Zhike processing, the general process includes frying, boiling, steaming and other steps, and the specific processing method will vary according to the characteristics and uses of the medicinal material; after processing, Qu Zhike is often used to prepare various Chinese herbal compound prescriptions or single Chinese medicines, which are used to treat symptoms such as spleen and stomach weakness, loss of appetite, abdominal distension and pain, diarrhea, etc., and has the effect of regulating the stomach and improving digestive function.
[0003] Chinese patent CN110851617B discloses a multi-source information drug screening method based on knowledge graph, including the following steps: fusion of multi-source heterogeneous data sources; optimization of molecular structure; optimization of scoring process architecture; molecular docking and scoring; construction of knowledge graph entities; construction of knowledge graph entity association relationships; screening and retrieval; the present invention organizes information such as optimized three-dimensional molecular structure, scored and screened molecular information into the knowledge graph, and performs loop retrieval in the knowledge graph, making drug screening more efficient and more accurate.
[0004] The existing screening of Qu Zhi Ke processed drugs is often carried out by screening personnel based on their own subjective views, and there is no objective standardized screening process. As a result, the screening results may be affected by the subjective influence of the operators, resulting in inaccurate screening results, causing processed drugs that do not meet the corresponding requirements to be retained or processed drugs that meet the corresponding requirements to be eliminated, causing losses to merchants or enterprises, and also affecting the health of patients using Qu Zhi Ke processed drugs. Summary of the invention
[0005] In response to the problems in the related technology, the present invention proposes a method and system for screening processed drugs of Quzhike based on data mining to overcome the above-mentioned technical problems existing in the existing related technology.
[0006] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:
[0007] The present invention is a method for screening processed drugs of Quzhi Fructus based on data mining, comprising the following steps:
[0008] S1. Set a name set of processing methods of Quzhike; collect the corresponding processed Quzhike sample image matrix according to the processing method name set; construct a processed Quzhike feature category set; extract features of each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix;
[0009] S2, constructing an initial multi-classifier model; the initial multi-classifier model includes a first classifier, a second classifier, a third classifier and a fourth classifier; dividing the processed Qu Zhi Ke sample image feature set matrix to obtain a processed Qu Zhi Ke training sample image feature set matrix and a processed Qu Zhi Ke verification sample image feature set matrix; using the processed Qu Zhi Ke training sample image matrix and the Qu Zhi Ke processing method name set to train the first classifier, the second classifier, the third classifier and the fourth classifier respectively; then using the processed Qu Zhi Ke verification sample image matrix and the Qu Zhi Ke processing method name set to verify the first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier once, to obtain a final multi-classifier model;
[0010] S3, collecting the sample set of Quzhike processed drugs to be screened and the corresponding processing method names; using the final multi-classifier model to classify each Quzhike processed drug sample in the sample set of Quzhike processed drugs to be screened, and obtaining the processing method name set of Quzhike processed drugs to be screened; setting the screening rules for Quzhike processed drugs; screening the sample set of Quzhike processed drugs to be screened according to the screening rules for Quzhike processed drugs, and obtaining the screened Quzhike processed drug sample set;
[0011] The first classifier, the second classifier, the third classifier and the fourth classifier are preferably random forest classifiers, SVM classifiers, KNN classifiers and naive Bayes classifiers, respectively; the name set of the processing methods of Qu Zhi Ke includes, for example, boiling, steaming, vinegar-roasting, wine-roasting, bran-frying, bran-processing and honey-processing; the feature category set of Qu Zhi Ke after processing includes, for example, color and whether there is a white core; because the external features of the processed Qu Zhi Ke medicine obtained by using different processing methods are different, in this aspect, by constructing the correspondence between the sample image features in the sample image feature set matrix of Qu Zhi Ke after processing and the processing method names in the Qu Zhi Ke processing method name set, the sample image features are used as data, and the Qu Zhi Ke processing method names are used as data labels ; By constructing a multi-classifier model, the multi-classifier model is trained and verified using these data and data labels, so that the trained and verified multi-classifier model can analyze the corresponding processing method name according to the sample image features; then the multi-classifier model is used to classify the samples of Quzhike processed drugs to be screened to obtain the corresponding processing method name, and then the processing method name is compared with the actual processing method name, so as to analyze whether the drug prepared by the processing method meets the requirements. If it does not meet the requirements, it will be eliminated, and if it meets the requirements, it will be retained, thereby achieving the purpose of screening Quzhike processed drugs. In this screening process, there is no need for too much manual participation, making the screening process more objective and accurate.
[0012] Preferably, the S1 comprises the following steps:
[0013] S11, set the name set of the processing method of Quzhi Fructus a={a1, a2, ..., a i ,...,a a′}, a i represents the name of the i-th processing method of Quzhike, a′ represents the total number of the processing methods of Quzhike in the name set of processing methods of Quzhike;
[0014] S12, according to the name set of the processing method of Quzhi Fructus a={a1, a2, ..., a i ,...,a a′} Collect the corresponding processed Quzhi Fructus sample image matrix as follows,
[0015]
[0016] in, represents the jth Quzhike sample image processed by the i-th Quzhike processing method; Indicates the total number of images of Quzhike samples collected after being processed by each processing method of Quzhike;
[0017] S13. Constructing the characteristic category set of processed Quzhi Fructus represents the characteristic category of the i-th processed Fructus Aurantii Immaturus. It represents the total number of Quzhike characteristic categories in the processing Quzhike characteristic category set;
[0018] According to the characteristic category set of Quzhi Fructus after processing The image matrix of the processed Fructus Aurantii Immaturus sample The features of each sample in the image are extracted to obtain the image feature set matrix c of the processed Quzhi Fructus samples; as follows,
[0019]
[0020] Among them, c ij represents the feature set of the jth Quzhike sample image processed by the i-th Quzhike processing method, c ij ={c ij1 ,c ij2 ,...,c iji ,...,c ijb~},c iji represents the i-th feature data in the feature set of the j-th Quzhike sample image processed by the i-th Quzhike processing method;
[0021] By determining the name set of Quzhike processing methods, a basis is provided for the subsequent collection of Quzhike sample images; then, by extracting features from Quzhike sample images, quantifiable data is provided for the subsequent training and verification of multi-classifier models based on Quzhike sample images and the corresponding Quzhike processing method names, making the training and verification results more accurate.
[0022] Preferably, S2 comprises the following steps:
[0023] S21, constructing an initial multi-classifier model, wherein the classifier set in the initial multi-classifier model is and Respectively represent the first classifier, the second classifier, the third classifier and the fourth classifier in the initial multi-classifier model;
[0024] S22, set the training set ratio to The validation set ratio is The processed Quzhike sample image feature set matrix c is divided according to the training set ratio and the verification set ratio to obtain the processed Quzhike training sample image feature set matrix d and the processed Quzhike verification sample image feature set matrix d′; they are respectively as follows:
[0025]
[0026] Among them, dij represents the feature set of the jth Quzhike training sample image processed by the i-th Quzhike processing method; d i ' j represents the feature set of the jth Quzhike verification sample image processed by the i-th Quzhike processing method; represents the total number of feature sets of the collected Quzhike training sample images processed by each Quzhike processing method;
[0027] S23, using the processed Quzhike training sample image matrix d and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} Train the first classifier, the second classifier, the third classifier and the fourth classifier respectively to obtain a trained first classifier, a trained second classifier, a trained third classifier and a trained fourth classifier;
[0028] S24, the processed Qu Zhi Ke verification sample image matrix d' and the Qu Zhi Ke processing method name set a={a1, a2, ..., a i ,...,a a′} respectively input into the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier for a verification, and obtain a verification result; according to the verification result, adjust the parameters of the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier; obtain the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0029] S25, obtaining a final multi-classifier model according to the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0030] The initial multi-classifier model was trained and verified to obtain the final multi-classifier model, which provided an analytical tool for the subsequent classification of Quzhi Fructus samples to be screened.
[0031] Preferably, the S23 comprises the following steps:
[0032] S231, setting the training error thresholds of the first classifier, the second classifier, the third classifier and the fourth classifier to e1, e2, e3 and e4 respectively;
[0033] S232, the processed Quzhike training sample image matrix d and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} are respectively input into the first classifier, the second classifier, the third classifier and the fourth classifier for training;
[0034] When the training error of the first classifier is less than or equal to e1, the training of the first classifier is completed, and the trained first classifier is obtained;
[0035] When the training error of the second classifier is less than or equal to e2, the training of the second classifier is completed, and the trained second classifier is obtained;
[0036] When the training error of the third classifier is less than or equal to e3, the training of the third classifier is completed, and the trained third classifier is obtained;
[0037] When the training error of the fourth classifier is less than or equal to e4, the training of the fourth classifier is completed, and the trained fourth classifier is obtained;
[0038] The first classifier, the second classifier, the third classifier and the fourth classifier are trained separately, so that each classifier in the multi-classifier model can be optimized, making the overall classification effect of the multi-classifier model more accurate.
[0039] Preferably, the S24 comprises the following steps:
[0040] S241, setting the parameter set of the first classifier to The parameter set of the second classifier is The parameter set of the third classifier is And the parameter set of the fourth classifier is Set the parameters of the first classifier The parameter set of the second classifier The parameter set of the third classifier and the parameter set of the fourth classifier Merge to get the initial multi-classifier parameter set
[0041] S242, setting the verification accuracy thresholds of the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier to e1′, e′2, e3′ and e′4 respectively;
[0042] S243, the processed Quzhike verification sample image matrix d' and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} are respectively input into the first classifier after training, the second classifier after training, the third classifier after training and the fourth classifier after training for a verification, and a verification accuracy set is obtained. and Respectively represent the first verification accuracy rate of the trained first classifier, the first verification accuracy rate of the trained second classifier, the first verification accuracy rate of the trained third classifier, and the first verification accuracy rate of the trained fourth classifier;
[0043] S244, when There is no need to set the parameters of the first classifier Make adjustments; when The parameters of the first classifier need to be set Make adjustments;
[0044] when There is no need to set the parameters of the second classifier Make adjustments; when The parameter set for the second classifier is Make adjustments;
[0045] when There is no need to set the parameter set of the third classifier Make adjustments; when The parameter set for the third classifier is required Make adjustments; when There is no need to adjust the parameter set of the fourth classifier Make adjustments; when The parameter set for the fourth classifier needs to be Performing adjustments; obtaining the adjusted first classifier, the adjusted second classifier, the adjusted third classifier, and the adjusted fourth classifier;
[0046] The first classifier, the second classifier, the third classifier and the fourth classifier after training are verified and the parameters are adjusted according to the verification results, so that the classification effects of the first classifier, the second classifier, the third classifier and the fourth classifier are further guaranteed, thereby ensuring the accuracy of the final screening of the Quzhi Fructus samples to be screened.
[0047] Preferably, in S244, the parameter set of the first classifier is Adjust or set the parameters of the second classifier Adjust or modify the parameter set of the third classifier Adjust or modify the parameter set of the fourth classifier The adjustment process includes the following steps:
[0048] S2441, construct a chicken flock population, the size of the chicken flock population is Assuming the maximum number of iterations is f, the chicken population is expressed as follows,
[0049]
[0050] Among them, f ij ′ represents the jth chicken in the i-th search group in the chicken population; Represents the total number of search groups in the chicken population; represents the total number of chickens in the i-th search group in the chicken population;
[0051] S2442, randomly initialize the positions of all chickens in the chicken population to obtain the initial position matrix g of the chicken population; as follows,
[0052]
[0053] Among them, g ij represents the initial position of the jth chicken in the i-th search group in the chicken population;
[0054] S2443, setting the fitness calculation formula g′ of the chicken population;
[0055] When the parameter set of the first classifier When making adjustments, the fitness calculation formula is as follows:
[0056]
[0057] When the parameter set for the second classifier When making adjustments, the fitness calculation formula is as follows:
[0058]
[0059] When the parameter set of the third classifier When making adjustments, the fitness calculation formula is as follows:
[0060]
[0061] When the parameter set of the fourth classifier When making adjustments, the fitness calculation formula is as follows:
[0062]
[0063] In the formula, represents the bias coefficient;
[0064] The fitness value of each chicken in the chicken population is calculated according to the fitness calculation formula to obtain the chicken population fitness data matrix as follows,
[0065]
[0066] The initial position of each chicken in the chicken population is recorded as the optimal position of the chicken individual in the first round of iteration; The position of the corresponding individual chicken is recorded as the optimal position of the chicken population in the first round of iteration;
[0067] S2444, setting a fitness threshold h; and performing the fitness data matrix Traverse by row, set the chicken with the largest fitness value in each search group as the rooster; except for the rooster, set the chicken with a fitness value greater than or equal to h in each search group as the female chicken; set the chicken with a fitness value less than h in each search group as the chick;
[0068] S2445, start iteration, set the current iteration number to h′, update the positions of the roosters, hens and chicks in each iteration, and update the optimal position of the chicken population and the optimal position of the individual chickens; stop iteration when h′≥f, and output the optimal parameter set of the first classifier Or the optimal parameter set for the second classifier Or the optimal parameter set of the third classifier Or the optimal parameter set of the fourth classifier
[0069] S2446, substituting the optimal parameter set output in S2445 into the corresponding classifier to obtain the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0070] The chicken swarm algorithm is used to gradually optimize the parameters of the first classifier, the second classifier, the third classifier and the fourth classifier, which can avoid falling into the local optimum and failing to obtain the global optimum during the parameter optimization process, thereby improving the efficiency of optimizing the classifier parameters; and the error rate of the multi-classifier optimized by the chicken swarm algorithm is reduced, thereby improving the accuracy of the multi-classifier in classifying the Quzhi Fructus samples to be screened.
[0071] Preferably, S3 comprises the following steps:
[0072] S31. Collect the sample set of processed drugs of Quzhi Fructus to be screened And the corresponding preparation method name l, represents the i-th Quzhike processed drug sample to be screened, Indicates the total number of Quzhike processed drug samples collected for screening;
[0073] S32, using the final multi-classifier model to screen the Quzhi Fructus processed drug sample set Each Quzhike processed drug sample in the sample is classified to obtain a set of processing method names of the Quzhike processed drugs to be screened. l i ' represents the name of the processing method corresponding to the i-th Quzhike processed drug sample to be screened;
[0074] S33, according to the processing method name set of the Quzhi Fructus processed medicine to be screened And l set the screening rules for Quzhike processed drugs;
[0075] S34, screening the Quzhike processed drug sample set to be screened according to the Quzhike processed drug screening rule Screening was performed to obtain the sample set of processed drugs of Quzhi Fructus after screening It indicates the i-th Quzhike processed drug sample remaining after screening; Indicates the total number of the remaining Quzhike processed drug samples;
[0076] The final multi-classifier model is used to classify each Qu Zhi Ke processed drug sample in the Qu Zhi Ke processed drug sample set to be screened, and the processing method name information of the Qu Zhi Ke processed drug sample can be obtained. Then, by comparing the processing method name information with the actual processing method name, it can be determined whether the external characteristics of the processed drug meet the requirements, so that the drug can be screened.
[0077] Preferably, the screening rules for the processed drugs of Quzhi Fructus are as follows:
[0078] When l i When ′≠l, the Quzhike processed medicine sample is judged to be unqualified, and the Quzhike processed medicine sample is discarded;
[0079] When l i When ′=1, the Quzhike processed drug sample is determined to be qualified, and the Quzhike processed drug sample is retained;
[0080] The screening rules for processed drugs of Quzhike can be used to specifically and quantitatively judge and screen samples of processed drugs of Quzhike, providing a specific basis for screening.
[0081] A Quzhike processed drug screening system based on data mining, comprising a Quzhike sample image acquisition module after processing, a Quzhike sample image feature extraction module, a multi-classifier model construction optimization module and a screening module;
[0082] The processed Quzhike sample image acquisition module is used to acquire the corresponding processed Quzhike sample image matrix according to the Quzhike processing method name set;
[0083] The Quzhike sample image feature extraction module is used to extract features from each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix;
[0084] The multi-classifier model construction optimization module is used to construct an initial multi-classifier model, and then train and verify the initial multi-classifier model to obtain a final multi-classifier model;
[0085] The screening module is used to screen the Quzhi Fructus processed medicine sample set to be screened according to the final multi-classifier model.
[0086] The present invention has the following beneficial effects:
[0087] 1. In the present invention, a processed Quzhike sample image acquisition module, a Quzhike sample image feature extraction module, a multi-classifier model construction optimization module and a screening module are set; the first classifier, the second classifier, the third classifier and the fourth classifier are preferably random forest classifiers, SVM classifiers, KNN classifiers and naive Bayes classifiers, respectively; the name set of the processing methods of Quzhike includes, for example, boiling, steaming, vinegar roasting, wine roasting, bran frying, bran and honey processing; the feature category set of processed Quzhike includes, for example, color and whether there is a white core; because the external features of the processed Quzhike medicine obtained by different processing methods are different, in this aspect, the corresponding relationship between the sample image features in the processed Quzhike sample image feature set matrix and the processing method names in the Quzhike processing method name set is constructed , taking the sample image features as data, and the name of the processing method of Quzhike as data labels; by constructing a multi-classifier model, the multi-classifier model is trained and verified using these data and data labels, so that the multi-classifier model after training and verification can analyze the corresponding processing method name according to the sample image features; then the multi-classifier model is used to classify the Quzhike processed drug samples to be screened to obtain the corresponding processing method name, and then the processing method name is compared with the actual processing method name, so as to analyze whether the drug prepared by the processing method meets the requirements. If it does not meet the requirements, it will be eliminated, and if it meets the requirements, it will be retained, thereby achieving the purpose of screening Quzhike processed drugs. In this screening process, there is no need for too much manual participation, making the screening process more objective and accurate.
[0088] 2. In the present invention, by determining the name set of the processing methods of Quzhike, a basis is provided for the subsequent collection of Quzhike sample images; and by extracting features from the Quzhike sample images, quantifiable data is provided for the subsequent training and verification of the multi-classifier model according to the Quzhike sample images and the corresponding Quzhike processing method names, so that the training and verification results are more accurate.
[0089] 3. In the present invention, the initial multi-classifier model is trained and verified to obtain the final multi-classifier model; an analysis tool is provided for the subsequent classification of the Quzhike samples to be screened; the first classifier, the second classifier, the third classifier and the fourth classifier after training are verified and the parameters are adjusted according to the verification results, so that the classification effects of the first classifier, the second classifier, the third classifier and the fourth classifier are further guaranteed, and the accuracy of the final screening of the Quzhike samples to be screened is guaranteed; wherein, the chicken swarm algorithm is used to gradually optimize the parameters of the first classifier, the second classifier, the third classifier and the fourth classifier, which can avoid falling into the local optimum and failing to obtain the global optimum in the process of optimizing the parameters, so that the efficiency of optimizing the classifier parameters is improved; and the error rate of the multi-classifier optimized by the chicken swarm algorithm is reduced, thereby improving the accuracy of the multi-classifier in classifying the Quzhike samples to be screened.
[0090] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0091] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.
[0092] Figure 1 The present invention is a schematic diagram of the process of screening Quzhike processed drugs by a Quzhike processed drug screening system based on data mining. DETAILED DESCRIPTION
[0093] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.
[0094] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.
[0095] The present invention is a method for screening processed drugs of Quzhi Fructus based on data mining, comprising the following steps:
[0096] S1. Set a name set of processing methods of Quzhike; collect the corresponding processed Quzhike sample image matrix according to the processing method name set; construct a processed Quzhike feature category set; extract features of each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix;
[0097] The S1 comprises the following steps:
[0098] S11, set the name set of the processing method of Quzhi Fructus a={a1, a2, ..., a i ,...,a a′}, a i represents the name of the i-th processing method of Quzhike, a′ represents the total number of the processing methods of Quzhike in the name set of processing methods of Quzhike;
[0099] S12, according to the name set of the processing method of Quzhi Fructus a={a1, a2, ..., a i ,...,a a′} Collect the corresponding processed Quzhi Fructus sample image matrix as follows,
[0100]
[0101] in, represents the jth Quzhike sample image processed by the i-th Quzhike processing method; a~ represents the total number of Quzhike sample images collected after being processed by each Quzhike processing method;
[0102] S13. Constructing the characteristic category set of processed Quzhi Fructus represents the characteristic category of the i-th processed Fructus Aurantii Immaturus. It represents the total number of Quzhike characteristic categories in the processing Quzhike characteristic category set;
[0103] According to the characteristic category set of Quzhi Fructus after processing The image matrix of the processed Fructus Aurantii Immaturus sample The features of each sample in the image are extracted to obtain the image feature set matrix c of the processed Quzhi Fructus samples; as follows,
[0104]
[0105] Among them, c ij represents the feature set of the jth Quzhike sample image processed by the i-th Quzhike processing method, c iji represents the i-th feature data in the feature set of the j-th Quzhike sample image processed by the i-th Quzhike processing method;
[0106] S2, constructing an initial multi-classifier model; the initial multi-classifier model includes a first classifier, a second classifier, a third classifier and a fourth classifier; dividing the processed Qu Zhi Ke sample image feature set matrix to obtain a processed Qu Zhi Ke training sample image feature set matrix and a processed Qu Zhi Ke verification sample image feature set matrix; using the processed Qu Zhi Ke training sample image matrix and the Qu Zhi Ke processing method name set to train the first classifier, the second classifier, the third classifier and the fourth classifier respectively; then using the processed Qu Zhi Ke verification sample image matrix and the Qu Zhi Ke processing method name set to verify the first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier once, to obtain a final multi-classifier model;
[0107] The S2 comprises the following steps:
[0108] S21, constructing an initial multi-classifier model, wherein the classifier set in the initial multi-classifier model is and Respectively represent the first classifier, the second classifier, the third classifier and the fourth classifier in the initial multi-classifier model;
[0109] S22, setting the training set ratio to c~%, and the validation set ratio to 1-c~%; dividing the processed Quzhike sample image feature set matrix c according to the training set ratio and the validation set ratio, to obtain the processed Quzhike training sample image feature set matrix d and the processed Quzhike validation sample image feature set matrix d′; respectively as follows,
[0110]
[0111] Among them, d ij represents the feature set of the jth Quzhike training sample image processed by the i-th Quzhike processing method; d i ' j represents the feature set of the jth Quzhike verification sample image processed by the i-th Quzhike processing method; represents the total number of feature sets of the collected Quzhike training sample images processed by each Quzhike processing method;
[0112] S23, using the processed Quzhike training sample image matrix d and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} Train the first classifier, the second classifier, the third classifier and the fourth classifier respectively to obtain a trained first classifier, a trained second classifier, a trained third classifier and a trained fourth classifier;
[0113] The S23 comprises the following steps:
[0114] S231, setting the training error thresholds of the first classifier, the second classifier, the third classifier and the fourth classifier to e1, e2, e3 and e4 respectively;
[0115] S232, the processed Quzhike training sample image matrix d and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} are respectively input into the first classifier, the second classifier, the third classifier and the fourth classifier for training;
[0116] When the training error of the first classifier is less than or equal to e1, the training of the first classifier is completed, and the trained first classifier is obtained;
[0117] When the training error of the second classifier is less than or equal to e2, the training of the second classifier is completed, and the trained second classifier is obtained;
[0118] When the training error of the third classifier is less than or equal to e3, the training of the third classifier is completed, and the trained third classifier is obtained;
[0119] When the training error of the fourth classifier is less than or equal to e4, the training of the fourth classifier is completed, and the trained fourth classifier is obtained;
[0120] S24, the processed Qu Zhi Ke verification sample image matrix d' and the Qu Zhi Ke processing method name set a={a1, a2, ..., a i ,...,a a′} respectively input into the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier for a verification, and obtain a verification result; according to the verification result, adjust the parameters of the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier; obtain the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0121] The S24 comprises the following steps:
[0122] S241, setting the parameter set of the first classifier to The parameter set of the second classifier is The parameter set of the third classifier is And the parameter set of the fourth classifier is Set the parameters of the first classifier The parameter set of the second classifier The parameter set of the third classifier and the parameter set of the fourth classifier Merge to get the initial multi-classifier parameter set
[0123] S242, setting the verification accuracy thresholds of the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier to e1′, e′2, e3′ and e′4 respectively;
[0124] S243, the processed Quzhike verification sample image matrix d' and the Quzhike processing method name set a={a1, a2, ..., a i ,...,a a′} are respectively input into the first classifier after training, the second classifier after training, the third classifier after training and the fourth classifier after training for a verification, and a verification accuracy set is obtained. and Respectively represent the first verification accuracy rate of the trained first classifier, the first verification accuracy rate of the trained second classifier, the first verification accuracy rate of the trained third classifier, and the first verification accuracy rate of the trained fourth classifier;
[0125] S244, when There is no need to set the parameters of the first classifier Make adjustments; when The parameters of the first classifier need to be set Make adjustments;
[0126] when There is no need to set the parameters of the second classifier Make adjustments; when The parameter set for the second classifier is Make adjustments;
[0127] when There is no need to set the parameter set of the third classifier Make adjustments; when The parameter set for the third classifier is required Make adjustments; when There is no need to adjust the parameter set of the fourth classifier Make adjustments; when The parameter set for the fourth classifier needs to be Performing adjustments; obtaining the adjusted first classifier, the adjusted second classifier, the adjusted third classifier, and the adjusted fourth classifier;
[0128] In S244, the parameter set of the first classifier Adjust or set the parameters of the second classifier Adjust or modify the parameter set of the third classifier Adjust or modify the parameter set of the fourth classifier The adjustment process includes the following steps:
[0129] S2441, construct a chicken flock population, the size of the chicken flock population is Assuming the maximum number of iterations is f, the chicken population is expressed as follows,
[0130]
[0131] Among them, f ij ′ represents the jth chicken in the i-th search group in the chicken population; Represents the total number of search groups in the chicken population; represents the total number of chickens in the i-th search group in the chicken population;
[0132] S2442, randomly initialize the positions of all chickens in the chicken population to obtain the initial position matrix g of the chicken population; as follows,
[0133]
[0134] Among them, g ij represents the initial position of the jth chicken in the i-th search group in the chicken population;
[0135] S2443, setting the fitness calculation formula g′ of the chicken population;
[0136] When the parameter set of the first classifier When making adjustments, the fitness calculation formula is as follows:
[0137]
[0138] When the parameter set for the second classifier When making adjustments, the fitness calculation formula is as follows:
[0139]
[0140] When the parameter set of the third classifier When making adjustments, the fitness calculation formula is as follows:
[0141]
[0142] When the parameter set of the fourth classifier When making adjustments, the fitness calculation formula is as follows:
[0143]
[0144] In the formula, represents the bias coefficient;
[0145] The fitness value of each chicken in the chicken population is calculated according to the fitness calculation formula to obtain the chicken population fitness data matrix as follows,
[0146]
[0147] The initial position of each chicken in the chicken population is recorded as the optimal position of the chicken individual in the first round of iteration; The position of the corresponding individual chicken is recorded as the optimal position of the chicken population in the first round of iteration;
[0148] S2444, setting a fitness threshold h; and performing the fitness data matrix Traverse by row, set the chicken with the largest fitness value in each search group as the rooster; except for the rooster, set the chicken with a fitness value greater than or equal to h in each search group as the female chicken; set the chicken with a fitness value less than h in each search group as the chick;
[0149] S2445, start iteration, set the current iteration number to h′, update the positions of the roosters, hens and chicks in each iteration, and update the optimal position of the chicken population and the optimal position of the individual chickens; stop iteration when h′≥f, and output the optimal parameter set of the first classifier Or the optimal parameter set for the second classifier Or the optimal parameter set of the third classifier Or the optimal parameter set of the fourth classifier
[0150] S2446, substituting the optimal parameter set output in S2445 into the corresponding classifier to obtain the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0151] S25, obtaining a final multi-classifier model according to the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier;
[0152] S3, collecting the sample set of Quzhike processed drugs to be screened and the corresponding processing method names; using the final multi-classifier model to classify each Quzhike processed drug sample in the sample set of Quzhike processed drugs to be screened, and obtaining the processing method name set of Quzhike processed drugs to be screened; setting the screening rules for Quzhike processed drugs; screening the sample set of Quzhike processed drugs to be screened according to the screening rules for Quzhike processed drugs, and obtaining the screened Quzhike processed drug sample set;
[0153] The S3 comprises the following steps:
[0154] S31. Collect the sample set of processed drugs of Quzhi Fructus to be screened And the corresponding preparation method name l, represents the i-th Quzhike processed drug sample to be screened, Indicates the total number of Quzhike processed drug samples collected for screening;
[0155] S32, using the final multi-classifier model to screen the Quzhi Fructus processed drug sample set Each Quzhike processed drug sample in the sample is classified to obtain a set of processing method names of the Quzhike processed drugs to be screened. l i ' represents the name of the processing method corresponding to the i-th Quzhike processed drug sample to be screened;
[0156] S33, according to the processing method name set of the Quzhi Fructus processed medicine to be screened And l set the screening rules for Quzhike processed drugs;
[0157] The screening rules for the processed drugs of Quzhi Fructus are as follows:
[0158] When l i When ′≠l, the Quzhike processed medicine sample is judged to be unqualified, and the Quzhike processed medicine sample is discarded;
[0159] When l i When ′=1, the Quzhike processed drug sample is determined to be qualified, and the Quzhike processed drug sample is retained;
[0160] S34, screening the Quzhike processed drug sample set to be screened according to the Quzhike processed drug screening rule Screening was performed to obtain the sample set of processed drugs of Quzhi Fructus after screening It indicates the i-th Quzhike processed drug sample remaining after screening; Represents the total number of remaining Quzhike processed drug samples.
[0161] A Quzhike processed drug screening system based on data mining, comprising a Quzhike sample image acquisition module after processing, a Quzhike sample image feature extraction module, a multi-classifier model construction optimization module and a screening module;
[0162] The processed Quzhike sample image acquisition module is used to acquire the corresponding processed Quzhike sample image matrix according to the Quzhike processing method name set;
[0163] The Quzhike sample image feature extraction module is used to extract features from each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix;
[0164] The multi-classifier model construction optimization module is used to construct an initial multi-classifier model, and then train and verify the initial multi-classifier model to obtain a final multi-classifier model;
[0165] The screening module is used to screen the Quzhi Fructus processed medicine sample set to be screened according to the final multi-classifier model.
[0166] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0167] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.
Claims
1. A method for screening processed drugs of Quzhi Fructus based on data mining, characterized in that: The following steps are involved: S1. Set a name set of processing methods of Quzhike; collect the corresponding processed Quzhike sample image matrix according to the processing method name set; construct a processed Quzhike feature category set; extract features of each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix; S2, constructing an initial multi-classifier model; the initial multi-classifier model includes a first classifier, a second classifier, a third classifier and a fourth classifier; dividing the processed Qu Zhi Ke sample image feature set matrix to obtain a processed Qu Zhi Ke training sample image feature set matrix and a processed Qu Zhi Ke verification sample image feature set matrix; using the processed Qu Zhi Ke training sample image matrix and the Qu Zhi Ke processing method name set to train the first classifier, the second classifier, the third classifier and the fourth classifier respectively; then using the processed Qu Zhi Ke verification sample image matrix and the Qu Zhi Ke processing method name set to verify the first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier once, to obtain a final multi-classifier model; S3. Collect the sample set of processed drugs of Quzhike to be screened and the corresponding processing method names; The final multi-classifier model is used to classify each Quzhike processed drug sample in the Quzhike processed drug sample set to be screened, so as to obtain a set of processing method names of the Quzhike processed drugs to be screened; Setting a screening rule for processed drugs of Quzhike; screening a sample set of processed drugs of Quzhike to be screened according to the screening rule for processed drugs of Quzhike to obtain a screened sample set of processed drugs of Quzhike; The S1 comprises the following steps: S11. Set the name set of the processing method of Quzhi Ke , Indicates Name of the processing method of Citrus aurantium, Indicates the total number of the names of the processing methods of Quzhike in the name set of the processing methods of Quzhike; S12, according to the name collection of the processing methods of Quzhi Fructus Collect the corresponding processed Quzhi Fructus sample image matrix ;as follows, ; in, Indicates the use of The first 100 samples of Quyu Fructus aurantii; Indicates the total number of images of Quzhike samples collected after being processed by each processing method of Quzhike; S13. Constructing the characteristic category set of processed Quzhi Fructus , Indicates The characteristic categories of Quzhike after processing, It represents the total number of Quzhike characteristic categories in the processing Quzhike characteristic category set; According to the characteristic category set of Quzhi Fructus after processing The image matrix of the processed Fructus Aurantii Immaturus sample The features of each sample in the image were extracted to obtain the image feature set matrix of the processed Quzhi Fructus samples. ;as follows, ; in, Indicates the use of The first The feature set of the image of the Quzhi Fructus sample, , Indicates the use of The first The feature set of the image of Quzhi Fructus sample Characteristic data; The S2 comprises the following steps: S21, constructing an initial multi-classifier model, wherein the classifier set in the initial multi-classifier model is , , , and Respectively represent the first classifier, the second classifier, the third classifier and the fourth classifier in the initial multi-classifier model; S22, set the training set ratio to , the validation set ratio is ; According to the training set ratio and the validation set ratio, the image feature set matrix of the processed Quzhi Fructus sample is Divide and obtain the image feature set matrix of the processed Quzhi Fructus training samples Image feature set matrix of processed Quzhi Fructus validation samples ; They are as follows, ; ; in, Indicates the use of The first A feature set of 100 training sample images of Qufuzhi Fructus; Indicates the use of The first A feature set of the Quyu Fructus validation sample images; represents the total number of feature sets of the collected Quzhike training sample images processed by each Quzhike processing method; S23, using the processed Fructus Aurantii Immaturus to train sample image matrix And the name collection of the processing methods of Quzhi Ke The first classifier, the second classifier, the third classifier and the fourth classifier are trained respectively to obtain a trained first classifier, a trained second classifier, a trained third classifier and a trained fourth classifier; S24, verifying the processed Fructus Aurantii Immaturus sample image matrix And the name collection of the processing methods of Quzhi Ke Inputting the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier respectively for a verification to obtain a verification result; adjusting the parameters of the trained first classifier, the trained second classifier, the trained third classifier and the trained fourth classifier according to the verification result; obtaining the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier; S25, obtaining a final multi-classifier model according to the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier; The S23 comprises the following steps: S231, setting the training error thresholds of the first classifier, the second classifier, the third classifier and the fourth classifier to be , , and ; S232, the processed Fructus Aurantii Immaturus training sample image matrix And the name collection of the processing methods of Quzhi Ke Input into the first classifier, the second classifier, the third classifier and the fourth classifier respectively for training; When the training error of the first classifier is less than or equal to When , the training of the first classifier is completed, and the trained first classifier is obtained; When the training error of the second classifier is less than or equal to When , the training of the second classifier is completed, and the trained second classifier is obtained; When the training error of the third classifier is less than or equal to When , the training of the third classifier is completed, and the trained third classifier is obtained; When the training error of the fourth classifier is less than or equal to When , the training of the fourth classifier is completed, and the trained fourth classifier is obtained; The S24 comprises the following steps: S241, setting the parameter set of the first classifier to , the parameter set of the second classifier is , the parameter set of the third classifier is And the parameter set of the fourth classifier is , set the parameter set of the first classifier , the parameter set of the second classifier , the parameter set of the third classifier and the parameter set of the fourth classifier Merge to get the initial multi-classifier parameter set ; S242, setting the verification accuracy thresholds of the trained first classifier, the trained second classifier, the trained third classifier, and the trained fourth classifier to be , , and ; S243, verifying the processed Fructus Aurantii Immaturus sample image matrix And the name collection of the processing methods of Quzhi Ke The trained first classifier, the trained second classifier, the trained third classifier, and the trained fourth classifier are inputted respectively for verification to obtain a verification accuracy set. , , , and Respectively represent the first verification accuracy rate of the trained first classifier, the first verification accuracy rate of the trained second classifier, the first verification accuracy rate of the trained third classifier, and the first verification accuracy rate of the trained fourth classifier; S244, when There is no need to set the parameters of the first classifier Make adjustments; when , the parameter set for the first classifier is required Make adjustments; when There is no need to set the parameters of the second classifier Make adjustments; when , the parameter set for the second classifier is required Make adjustments; when There is no need to set the parameter set of the third classifier Make adjustments; when , the parameter set of the third classifier is required Make adjustments; when There is no need to adjust the parameter set of the fourth classifier Make adjustments; when , the parameter set of the fourth classifier is required Adjustment is performed; and the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier are obtained.
2. The method for screening processed drugs of Quzhi Fructus based on data mining according to claim 1, characterized in that: In S244, the parameter set of the first classifier Adjust or set the parameters of the second classifier Adjust or modify the parameter set of the third classifier Adjust or modify the parameter set of the fourth classifier The adjustment process includes the following steps: S2441, construct a chicken flock population, the size of the chicken flock population is , set the maximum number of iterations to , the chicken population is represented as follows, ; in, Indicates the first of the search groups Chicken; represents the total number of search groups in the chicken population; Indicates the first The total number of chickens in the search group; S2442, randomly initialize the positions of all chickens in the chicken population to obtain an initial position matrix of the chicken population ;as follows, ; in, Indicates the first of the search groups The initial position of the chickens; S2443, setting the fitness calculation formula of the chicken population ; When the parameter set of the first classifier When making adjustments, the fitness calculation formula is as follows: ; When the parameter set for the second classifier When making adjustments, the fitness calculation formula is as follows: ; When the parameter set of the third classifier When making adjustments, the fitness calculation formula is as follows: ; When the parameter set of the fourth classifier When making adjustments, the fitness calculation formula is as follows: ; In the formula, represents the bias coefficient; The fitness value of each chicken in the chicken population is calculated according to the fitness calculation formula to obtain the chicken population fitness data matrix ;as follows, ; The initial position of each chicken in the chicken population is recorded as the optimal position of the chicken individual in the first round of iteration; The position of the corresponding individual chicken is recorded as the optimal position of the chicken population in the first round of iteration; S2444, set fitness threshold ; The chicken flock fitness data matrix Traverse by row, set the chicken with the largest fitness value in each search group as the rooster; except for the rooster, set the fitness value of each search group greater than or equal to As female chickens; set the fitness value of each search group to be less than of chicken individuals as chicks; S2445, start iteration, set the current number of iterations to , in each round of iteration, the positions of the roosters, hens and chicks are updated, and the optimal positions of the chicken population and the individual chickens are updated at the same time; Stop the iteration when Or the optimal parameter set for the second classifier Or the optimal parameter set of the third classifier Or the optimal parameter set of the fourth classifier ; S2446. Substitute the optimal parameter set output in S2445 into the corresponding classifier to obtain the adjusted first classifier, the adjusted second classifier, the adjusted third classifier and the adjusted fourth classifier.
3. The method for screening processed drugs of Quzhi Fructus based on data mining according to claim 1, characterized in that: The S3 comprises the following steps: S31. Collect the sample set of processed drugs of Quzhi Fructus to be screened And the corresponding preparation method name , Indicates There are samples of Quzhike processed medicines to be screened. Indicates the total number of Quzhike processed drug samples collected for screening; S32, using the final multi-classifier model to screen the Quzhi Fructus processed medicine sample set Each Quzhike processed drug sample in the sample is classified to obtain a set of processing method names of the Quzhike processed drugs to be screened. , Indicates The name of the processing method corresponding to each of the processed Quzhike drug samples to be screened; S33, according to the processing method name set of the Quzhi Fructus processed medicine to be screened as well as Set up the screening rules for Quzhike processed drugs; S34, screening the Quzhike processed drug sample set to be screened according to the Quzhike processed drug screening rule Screening was performed to obtain the sample set of processed drugs of Quzhi Fructus after screening. , It means that after screening, the remaining 10 samples of processed medicines of Quzhi Ke; Represents the total number of remaining Quzhike processed drug samples.
4. The method for screening processed drugs of Quzhi Fructus based on data mining according to claim 3, characterized in that: The screening rules for the processed drugs of Quzhi Fructus are as follows: when When the sample of the processed drug of Quzhike is determined to be unqualified, the sample of the processed drug of Quzhike is discarded; when At that time, the Quzhike processed medicine sample was determined to be qualified and the Quzhike processed medicine sample was retained.
5. A system for implementing the method for screening processed drugs of Quzhi Fructus based on data mining as claimed in any one of claims 1 to 4, characterized in that: It includes an image acquisition module for processed Quzhike samples, a feature extraction module for Quzhike sample images, a multi-classifier model building optimization module and a screening module; The processed Quzhike sample image acquisition module is used to acquire the corresponding processed Quzhike sample image matrix according to the Quzhike processing method name set; The Quzhike sample image feature extraction module is used to extract features from each sample in the processed Quzhike sample image matrix according to the processed Quzhike feature category set to obtain the processed Quzhike sample image feature set matrix; The multi-classifier model construction optimization module is used to construct an initial multi-classifier model, and then train and verify the initial multi-classifier model to obtain a final multi-classifier model; The screening module is used to screen the Quzhi Fructus processed medicine sample set to be screened according to the final multi-classifier model.
Citation Information
Patent Citations
A multi-source information drug screening method based on knowledge graphs
CN110851617B
Traditional Chinese medicine decoction piece quality inspection method, device and equipment and storage medium
CN114266740A
Data classification method based on machine learning
CN117197591A
Transformer fault diagnosis method and system for optimizing LS-TSVM and DGA based on chicken flock algorithm
CN117435881A
Methods for analyzing plant material, for determining plant material components and for detecting plant diseases in plant material
WO2022175309A1
