Method for identifying pepper variety and producing area based on electronic nose
By using an electronic nose-based manifold discriminant extreme learning machine (MDELM) classifier, combined with manifold constraints, linear discriminant constraints, and maximum variance constraints, the problem of identifying chili varieties and origins was solved. This enabled efficient and accurate chili classification and origin tracing, adapting to different data volume variations and exhibiting good anti-interference capabilities.
Patent Information
- Application Number
- CN202411023637.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-19
AI Technical Summary
Existing technologies are insufficient for quickly and accurately identifying chili varieties and their origins that are similar in appearance and smell. Traditional methods suffer from drawbacks such as high subjectivity, high cost, and complex operation. Furthermore, existing electronic nose systems have failed to achieve multi-variety classification and origin traceability.
A manifold discriminant extreme learning machine (MDELM) classifier based on an electronic nose was designed, combining manifold constraints, linear discriminant constraints, and maximum variance constraints. By mapping the features of training sample data to the ELM feature space, a manifold discriminant extreme learning machine (MDELM) classifier was designed for the identification of chili pepper varieties and origins.
It improves the efficiency and accuracy of chili pepper classification, solves the problems of chili pepper variety classification and origin traceability, has rapid identification ability and good generalization performance, adapts to different data volume changes, and has strong anti-interference ability.
Smart Images

Figure CN118981707A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of electronic nose signal processing, and in particular relates to a method for identifying pepper varieties and origins based on an electronic nose. Background Art
[0002] The variety and origin of peppers are important factors affecting the quality and taste of peppers. There are many types of peppers, and many peppers have great similarities in appearance and smell. The quality and taste of the same peppers from different origins or different peppers from the same origin are very different, so there are also large differences in selling value. Due to the above reasons, it is very common to use inferior peppers to impersonate high-quality peppers. Traditional identification methods mainly include sensory analysis and physical and chemical analysis.
[0003] Sensory analysis mainly refers to the appraiser's use of the olfactory receptors to distinguish different objects. However, due to its defects such as strong subjectivity, poor repeatability, limited sensitivity and recognition, and possible harm to the health of the appraiser, it is not applicable in many occasions.
[0004] Physical and chemical analysis mainly obtains quantitative and qualitative analysis results by measuring and analyzing the physical or chemical properties of substances. However, due to their shortcomings such as high cost, complex operation, cumbersome sample preparation, and high maintenance and calibration requirements, they are currently only used in some fields that require precise data support.
[0005] The existing technologies for classifying peppers using electronic nose systems only classify two types of peppers and only classify the types of peppers. There is no classification of multiple varieties, and no technology for tracing the origin of peppers has been proposed.
[0006] Therefore, a method for identifying pepper varieties and origins is needed. Summary of the invention
[0007] In view of this, the object of the present invention is to provide a method for identifying pepper varieties and origins based on an electronic nose.
[0008] The objective of the present invention is achieved through the following technical solutions:
[0009] A method for identifying pepper varieties and origins based on an electronic nose, comprising:
[0010] Using the electronic nose, obtaining target electronic nose data features of the pepper to be detected;
[0011] The target electronic nose data features are input into a trained manifold discriminant extreme learning machine classifier, namely, an MDELM classifier, to identify the variety and origin of the pepper to be detected. The MDELM classifier integrates manifold constraints, linear discriminant constraints and maximum variance constraints into the ELM framework.
[0012] Furthermore, the training of the MDELM classifier includes:
[0013] Get the training sample set X and the corresponding label Y, where: represents the i-th training sample data, y i represents the label corresponding to the i-th training sample data. The label corresponding to each training sample contains the variety or origin information of the training sample. D represents the feature dimension of the electronic nose data in the original space of the training sample data. N represents the number of training samples. represents the set of real numbers;
[0014] Map the training sample data X to the L-dimensional ELM feature space to generate the ELM feature matrix H, where: h i represents the feature representation of the i-th training sample in the ELM feature space;
[0015] Using the ELM feature matrix H, the manifold constraint, line type discrimination constraint and maximum variance constraint are established respectively;
[0016] According to the manifold constraint, the linear discrimination constraint and the maximum variance constraint, the objective function of the MDELM classifier training is determined, and the objective function is solved to obtain the MDELM classifier under the optimal solution, which is the trained MDELM classifier.
[0017] Furthermore, the establishment of manifold constraints includes:
[0018] Perform manifold regularization on H, that is
[0019]
[0020] Where β represents the ELM output weight matrix, W ij represents the weight of the close relationship between the i-th sample and the j-th sample in the training sample, h i and h j They represent the mapping of the i-th sample and the j-th sample in the training set in the ELM space, respectively. The superscript T represents the matrix transpose;
[0021] Let L = DW, then
[0022] Among them, W represents the weight W ij The adjacency matrix composed of , D represents the degree matrix, which is a diagonal matrix. Except for the main diagonal elements, all other elements are zero. Tr() represents the trace of the matrix.
[0023] Manifold constraint is to minimize Tr(β T HLHT β), that is
[0024] The establishment of line type discrimination constraints includes:
[0025] Minimize the distance from each sample in the same class to the class center to cluster the same class, and maximize the distance between each class center and the center of the entire data set to keep different classes away. The formula is:
[0026]
[0027] Among them, S W and S B They represent the intra-class scatter matrix and inter-class scatter matrix of the training set respectively.
[0028] The establishment of the maximum variance constraint includes:
[0029] The most representative features are selected or learned by maximizing the variance of the data, i.e.
[0030]
[0031] Furthermore, the weight W ij for:
[0032]
[0033] Among them, ξ>1, t represents the parameter of the heat kernel function, and label(a) represents the label of sample a.
[0034] Furthermore, the objective function of MDELM classifier training is:
[0035]
[0036] Among them, || || F represents the Frobenius norm.
[0037] Furthermore, solving the objective function includes:
[0038] The objective function is transformed into a minimization problem containing equality constraints for solution, namely:
[0039]
[0040] stTr(β T S B β)+Tr(β T HH T β)=τ
[0041] Among them, τ>0 is a constant,
[0042] Construct the Lagrangian function:
[0043]
[0044] Where γ represents the Lagrange multiplier;
[0045] Will The partial derivative of β is set to zero, and the optimal solution of β is obtained, that is,
[0046]
[0047] The optimal solution of β is β=[(1-γ)HH T +HLH T +I L +S W -γS B ] -1 HY,
[0048] Among them, I L represents the L-dimensional identity matrix.
[0049] The present invention also provides a processor, which implements the steps of the above method when executing a program.
[0050] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.
[0051] The present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.
[0052] The beneficial effects of the present invention are:
[0053] The method of the present invention integrates manifold learning, linear discriminant analysis and variance maximum theory into a unified ELM framework, and designs a manifold discriminant extreme learning machine MDELM classifier. The classifier only needs to input the data to be classified into the classifier, in which the geometric structure information of the data will be retained, the data dimension will be reduced, and the discriminant information between the data will be enhanced, and finally the classification accuracy is output by the classifier. The use of the classifier is extremely simple, and the two traditional steps of first dimensionality reduction and then classification are combined to effectively improve the efficiency and accuracy of pepper classification; moreover, the proposed MDELM model is a classifier learning framework that can quickly identify peppers, which well solves the classification of pepper varieties and the tracing of origin problems, which involves two steps of random feature mapping and output weight training. The model proposed by the present invention has the advantages of ELM and ensures the generalization performance in the application of electronic noses; compared with the existing classification task of peppers using electronic noses, the method proposed in the present invention is more in line with the application requirements in actual scenarios.
[0054] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:
[0056] Figure 1 It is a schematic diagram of the MDELM model;
[0057] Figure 2 It is the comparison result of classification accuracy of different methods under data set A;
[0058] Figure 3 It is the comparison result of different methods for data set A with different numbers of training samples;
[0059] Figure 4 (a) is the comparison result of classification accuracy of different methods under dataset B;
[0060] Figure 4 (b) is the comparison result of classification accuracy of different methods under dataset C;
[0061] Figure 5 It is the comparison result of various methods for data set B with different numbers of training samples;
[0062] Figure 6It is the comparison result of various methods for data set C with different numbers of training samples. DETAILED DESCRIPTION
[0063] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.
[0064] The present invention aims to solve the problem that peppers and pepper powders with similar appearance and smell cannot be quickly and accurately identified by humans. The efficiency and accuracy of pepper identification are improved, and the problem of inferior products being passed off as good ones and fake products being passed off as real ones in the pepper market is alleviated. Based on the above purpose, the present invention designs a manifold discriminant extreme learning machine (MDELM) classifier by integrating manifold learning, linear discriminant analysis and variance maximum theory into a unified extreme learning machine framework. Figure 1 is a schematic diagram of the MDELM classifier. Figure 1 As shown, the MDELM classifier integrates manifold constraints, line type discrimination constraints and variance maximization constraints into the ELM framework.
[0065] Using the MDELM classifier, the present invention proposes a method for identifying pepper varieties and origins based on an electronic nose, comprising:
[0066] Using the electronic nose, obtaining target electronic nose data features of the pepper to be detected;
[0067] The target electronic nose data features are input into the trained manifold discriminant extreme learning machine classifier, namely, the MDELM classifier, to identify the variety and origin of the pepper to be detected.
[0068] The training of the MDELM classifier includes:
[0069] Get the training sample set X and the corresponding label Y, where: represents the i-th training sample data, y i represents the label corresponding to the i-th training sample data. The label corresponding to each training sample is the variety or origin information of the training sample. D represents the feature dimension of the electronic nose data in the original space of the training sample data. N represents the number of training samples. Represents the set of real numbers.
[0070] In some embodiments, different classification tasks have different label files. For origin identification, each y i corresponds to the origin of each pepper sample; for variety identification, each y iIt corresponds to the variety information of each pepper sample. When training the MDELM classifier, different training samples and corresponding label files can be selected for training according to different classification tasks. For the two classification tasks of origin and variety, two classifiers (i.e., origin classifier and variety classifier) are used for classification respectively. Both classifiers are MDELM classifiers, and the two have the same structure and the same training method, but the training data is different. In some embodiments, the number of output categories of the two classifiers may also be different. For example, for variety classification, the training sample may contain N1 (N1 can be any suitable number) varieties of peppers. The variety classifier is trained using these N1 varieties of peppers, and the classifier generated by the training can classify N1 varieties of peppers, that is, output N1 categories; for origin classification, the training sample may contain N2 (N2 can be any suitable number) peppers from origins, and the origin classifier is trained using these N2 peppers from origins, and the classifier generated by the training can classify N2 peppers from origins, that is, output N2 categories.
[0071] Next, the training sample data X is mapped to the L-dimensional ELM feature space to generate the ELM feature matrix H, where h i represents the feature representation of the i-th training sample in the ELM feature space;
[0072] Using the ELM feature matrix H, the manifold constraint, line type discrimination constraint and maximum variance constraint are established respectively;
[0073] According to the manifold constraint, the linear discrimination constraint and the maximum variance constraint, the objective function of the MDELM classifier training is determined, and the objective function is solved to obtain the MDELM classifier under the optimal solution, which is the trained MDELM classifier.
[0074] Traditional linear dimensionality reduction methods may not be able to effectively capture the nonlinear structure of electronic nose data. The goal of manifold learning is to achieve dimensionality reduction and feature extraction of data by learning the potential manifold representation of the data. According to manifold theory, two samples that are close in the sample space are very likely to belong to the same category. Therefore, when using an electronic nose to process classification tasks, it is hoped that more reliable classification information can be provided to the model by learning the position information between samples in the original data space. The present invention selects the Laplace mapping for learning nonlinear structures in high-dimensional data. It is based on the Laplace matrix and optimizes the mapping of data points in the dimensionality reduction space to maintain the geometric structure relationship of the data. The establishment of manifold constraints includes:
[0075] Perform manifold regularization on H, that is
[0076]
[0077] Where β represents the ELM output weight matrix, W ij represents the weight of the close relationship between the i-th sample and the j-th sample in the training sample, h i and h j They represent the mapping of the i-th sample and the j-th sample in the training set in the ELM space, respectively. The superscript T represents the matrix transpose;
[0078] Let L = DW, then
[0079] Among them, W represents the weight W ij The adjacency matrix composed of , D represents the degree matrix, which is a diagonal matrix. Except for the main diagonal elements, all other elements are zero, that is, Tr() represents the trace of the matrix and L represents the Laplacian matrix.
[0080] Weight W ij for:
[0081]
[0082] Among them, ξ>1, t represents the parameter of the heat kernel function, and label(a) represents the label of sample a.
[0083] Manifold constraint is to minimize Tr(β T HLH T β), that is
[0084] Linear discriminant analysis is a classic pattern recognition and classification method. Its goal is to map high-dimensional data to a low-dimensional space while keeping the distance between categories as large as possible and samples of the same category closer together. When classifying samples, it is often found that samples that do not belong to the same category are very close, while samples that belong to the same category may be far apart, which will seriously affect the judgment of the model during classification and lead to misclassification. Therefore, considering the importance of discriminant information between samples in the data set, it is hoped that samples of the same category are as close as possible, and samples of different categories are far away from each other.
[0085] In the present invention, the establishment of line type discrimination constraints includes:
[0086] Minimize the distance from each sample in the same class to the class center to cluster the same class, and maximize the distance between each class center and the center of the entire data set to keep different classes away. The formula is:
[0087]
[0088] Among them, μ cand μ represent the center of the c-th sample in the training sample set and the center of the entire training sample set respectively. W and S B They represent the intra-class scatter matrix and inter-class scatter matrix of the training set respectively, class c represents the set of samples of class c, and C represents the total number of classes.
[0089] The present invention also selects or learns the most representative features by maximizing the variance of the data, that is, establishing a maximum variance constraint, i.e.
[0090]
[0091] Extreme learning machine is a single hidden layer feedforward neural network with the characteristics of strong generalization ability, fast learning speed and less manual intervention. Inspired by the ELM classifier, this paper proposes a unified classifier learning framework, which integrates manifold constraints, linear discriminant constraints and maximum variance constraints into a unified framework, and finally obtains a complete classifier model.
[0092] The objective function of MDELM classifier training is:
[0093]
[0094] Among them, Y represents the prediction result obtained using the training set sample data, || || F represents the Frobenius norm.
[0095] Since the solution of the above objective function is not unique, the objective function can be transformed into a minimization problem containing equality constraints to obtain a unique solution.
[0096] The transformed problem can be written as:
[0097]
[0098] stTr(β T S B β)+Tr(β T HH T β)=τ
[0099] Among them, τ>0 is a constant,
[0100] In order to solve the above equation, we can construct the Lagrangian function:
[0101]
[0102] Where γ represents the Lagrange multiplier;
[0103] Then, The partial derivative of β is set to zero, and the optimal solution of β is obtained, that is,
[0104]
[0105] The optimal solution of β is β=[(1-γ)HH T +HLH T +I L +S W -γS B ] -1 HY,
[0106] Among them, I L represents the L-dimensional identity matrix.
[0107] The present invention also provides a processor, which implements the steps of the above method when executing a program.
[0108] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.
[0109] The present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.
[0110] In order to verify the effectiveness of the proposed method in processing various pepper classification tasks, a pepper odor information collection experiment was designed, and three pepper odor data sets were collected in the Intelligent Sensory Laboratory of the Pepper Research Institute of Guizhou Academy of Agricultural Sciences. A large number of experiments were carried out on the three data sets to verify the proposed method.
[0111] Dataset A collected odor data of 13 types of peppers. The final number of samples for each type of pepper is shown in Table 1.
[0112] Table 1 Dataset A
[0113]
[0114] Next, under data set A, different methods are used to classify the samples in data set A. The results are as follows: Figure 2 As shown, Figure 2The horizontal axis represents different methods, namely, Support Vector Machine (SVM), Extreme Learning Machine (ELM), Kernel ELM (KELM), Multilayer Perceptron (MLP), Decision Tree (DT), Random Forest (RF) and MDELM (i.e., the method of the present invention); the vertical axis represents the classification accuracy. According to Figure 2 It can be seen that the classification accuracy of the method of the present invention (i.e., MDELM) under data set A is higher than that of other comparison algorithms. The classification accuracy of the method of the present invention under data set A reaches 90.18%, which is 13.10% higher than the lowest SVM algorithm and 3.57% higher than the second highest random forest algorithm. Such experimental results can preliminarily reflect that the classification performance of the MDELM model under data set A is better than other algorithms.
[0115] Figure 3 It is the comparison result of different methods for data set A with different numbers of training samples. Figure 3 The horizontal axis represents the number of samples, and the vertical axis represents the classification accuracy. Figure 3 As shown in the figure, the classification accuracy of the method of the present invention under different numbers of training samples is better than that of other comparison algorithms. As the number of training samples decreases, the performance of the method of the present invention degrades slightly, while the performance of other methods decreases significantly. This shows that compared with other classification algorithms, the method of the present invention has better adaptability to different amounts of data.
[0116] Dataset B and Dataset C contain the odor data of Jiaoyang No. 6 and Line Pepper 301 from 7 different origins. The origin and sample number of each pepper are shown in Table 2.
[0117] Table 2 Dataset B and Dataset C
[0118]
[0119] Figure 4 (a) is the comparison result of classification accuracy of different methods under dataset B; Figure 4 (b) is the comparison result of classification accuracy of different methods under dataset C. Figure 4 The horizontal axis represents different methods; the vertical axis represents the classification accuracy. Figure 4As shown in Figures 4(a) and 4(b), the method of the present invention also exhibits good classification effect when identifying the origin of the two peppers. The classification accuracy rate in data set B reaches 85.71%, and the classification accuracy rate in data set C reaches 98.57%, reflecting that the method of the present invention is also better than other algorithms in identifying the origin of Jiaoyang No. 6 and Line Pepper 301.
[0120] Figure 5 and Figure 6 The comparison results of various methods for dataset B and dataset C with different numbers of training samples are shown below. Figure 5 and Figure 6 The horizontal axis represents the number of samples, and the vertical axis represents the classification accuracy. Figure 5 and Figure 6 It can be seen that the classification accuracy of the method of the present invention is better than that of other comparison algorithms under the vast majority of sample numbers within the experimental range. When the number of training samples gradually increases, the classification accuracy of various algorithms can reach a high level. However, when the number of training samples gradually decreases, the method of the present invention only shows a slight performance decline, while other methods show serious performance deterioration. When the number of training samples is much less than the number of test samples, the classification accuracy of the method of the present invention is much higher than that of other algorithms. It can be seen from the red broken line (i.e., the method of the present invention) that the method of the present invention shows good adaptability under the experiment of the number of training samples within the entire experimental range, and is least affected by the change in the number of samples.
[0121] In order to more clearly understand the positive contribution of each module of the method of the present invention to the final classification, an ablation experiment was also conducted to selectively disable certain modules in the model (i.e., modules corresponding to manifold constraints, modules corresponding to linear discriminant constraints, and modules corresponding to maximum variance constraints) in order to evaluate the impact of these modules on the performance of the entire model. Finally, ablation experiments of four scenarios were designed on three data sets. Tables 3, 4, and 5 show the ablation experiment results under data sets A, B, and C, respectively. Among them, MDELM in the table means that three constraints are included; MDELM-1 means that it does not contain manifold constraints, only linear discriminant constraints and maximum variance constraints; MDELM-2 means that it does not contain linear discriminant constraints, only contains manifold constraints and maximum variance constraints; MDELM-3 means that it does not contain maximum variance constraints, only contains linear discriminant constraints and manifold constraints.
[0122] Based on the results of all ablation experiments, the classification accuracy of the MDELM model decreases after removing any module (i.e., constraint), which shows that each module of the MDELM model contributes to the overall performance of the MDELM. In addition, each module improves the model's adaptability to different data sets from different perspectives. After removing a module, the model's adaptability to different data may decrease.
[0123] Table 3 Ablation experiments on dataset A
[0124]
[0125] Table 4 Ablation experiments on dataset B
[0126]
[0127] Table 5 Ablation experiments on dataset C
[0128]
[0129] The present invention also verifies whether the method of the present invention can also be competent for the task of identifying peppers by using the early response curve of odor detection through an early recognition experiment. The specific experiment setting is: extracting the sample points of the 10th second, 20th second, 40th second, and 60th second in the response curve as features, and all samples are allocated according to the ratio of 7:3 between the number of training samples and the number of test samples.
[0130] Table 6 gives all the early recognition experimental results under the three data sets. It can be seen that the MDELM model can also achieve good classification results when using the data in the early response stage, which shows that the MDELM model has excellent anti-interference ability. Even if the features in the early response curve are selected, good classification results can still be obtained. This further shows that the MDELM has good model robustness and can show good performance in tasks with certain difficulties in data acquisition or tasks with emergencies.
[0131] Table 6 Early recognition experimental results under three data sets (%)
[0132]
[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.
Claims
1. A method for identifying pepper varieties and origins based on an electronic nose, characterized in that: include: Using the electronic nose, obtaining target electronic nose data features of the pepper to be detected; The target electronic nose data features are input into a trained manifold discriminant extreme learning machine classifier, namely, an MDELM classifier, to identify the variety and origin of the pepper to be detected. The MDELM classifier integrates manifold constraints, linear discriminant constraints and maximum variance constraints into the ELM framework.
2. The method for identifying pepper varieties and origins based on an electronic nose according to claim 1, characterized in that: The training of the MDELM classifier includes: Get the training sample set X and the corresponding label Y, where: represents the i-th training sample data, y i represents the label corresponding to the i-th training sample data. The label corresponding to each training sample is the variety or origin information of the training sample. D represents the feature dimension of the electronic nose data in the original space of the training sample data. N represents the number of training samples. represents the set of real numbers; Map the training sample data X to the L-dimensional ELM feature space to generate the ELM feature matrix H, where: h i represents the feature representation of the i-th training sample in the ELM feature space; Using the ELM feature matrix H, the manifold constraint, line type discrimination constraint and maximum variance constraint are established respectively; According to the manifold constraint, the linear discrimination constraint and the maximum variance constraint, the objective function of the MDELM classifier training is determined, and the objective function is solved to obtain the MDELM classifier under the optimal solution, which is the trained MDELM classifier.
3. The method for identifying pepper varieties and origins based on an electronic nose according to claim 2, characterized in that: The establishment of manifold constraints includes: Perform manifold regularization on H, that is Where β represents the ELM output weight matrix, W ij represents the weight of the close relationship between the i-th sample and the j-th sample in the training sample, h i and h j They represent the mapping of the i-th sample and the j-th sample in the training set in the ELM space, respectively. The superscript T represents the matrix transpose; Let L = DW, then Among them, W represents the weight W ij The adjacency matrix composed of , D represents the degree matrix, which is a diagonal matrix. Except for the main diagonal elements, all other elements are zero. Tr() represents the trace of the matrix. Manifold constraint is to minimize Tr(β T HLH T β), that is The establishment of line type discrimination constraints includes: Minimize the distance from each sample in the same class to the class center to cluster the same class, and maximize the distance from each class center to the center of the entire data set to keep different classes away. The formula is: Among them, S W and S B They represent the intra-class scatter matrix and inter-class scatter matrix of the training set respectively. The establishment of the maximum variance constraint includes: The most representative features are selected or learned by maximizing the variance of the data, i.e.
4. The method for identifying pepper varieties and origins based on an electronic nose according to claim 3, characterized in that: Weight W ij for: Among them, ξ>1, t represents the parameter of the heat kernel function, and label(a) represents the label of sample a.
5. The method for identifying pepper varieties and origins based on electronic nose according to claim 3, characterized in that: The objective function of MDELM classifier training is: Among them, || || F represents the Frobenius norm.
6. The method for identifying pepper varieties and origins based on an electronic nose according to claim 5, characterized in that: Solving the objective function includes: The objective function is transformed into a minimization problem containing equality constraints for solution, namely: stTr(β T S B b)+Tr(b T HH T b)=τ Among them, τ>0 is a constant, Construct the Lagrangian function: Where γ represents the Lagrange multiplier; Will The partial derivative of β is set to zero, and the optimal solution of β is obtained, that is, The optimal solution of β is β=[(1-γ)HH T +HLH T +I L +S W -γS B ] -1 HY, Among them, I L represents the L-dimensional identity matrix.
7. A processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.
8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.
