Method for identifying radix ophiopogonis medicinal materials from different producing areas

Abstract

The invention discloses a method for identifying radix ophiopogonis medicinal materials from different producing areas. Collecting spectral data of radix ophiopogonis medicinal materials from different producing areas by adopting an infrared spectrometer; optimizing hyper-parameters of the initial one-dimensional convolutional neural network model based on the spectral data to obtain a target one-dimensional convolutional neural network; and identifying the spectral data of the to-be-identified radix ophiopogonis medicinal material based on the target one-dimensional convolutional neural network, and determining the production place corresponding to the to-be-identified radix ophiopogonis medicinal material. According to the method, the infrared spectrum technology and the target one-dimensional convolutional neural network are combined, so that the spectral characteristics of the internal components of the radix ophiopogonis medicinal materials can be accurately captured, and the effective identification of the radix ophiopogonis medicinal materials from different producing areas can be realized. Compared with a traditional microscope observation or chemical analysis method, the method has higher recognition precision. A traditional method needs to depend on experienced experts for manual identification, and the cost is high.

Description

Technical Field

[0001] The present application relates to the field of medicinal material technology, and in particular to a method for identifying Radix Ophiopogonis japonicus medicinal materials from different origins. Background Art

[0002] Ophiopogon japonicus is a traditional Chinese medicinal material widely used in the field of traditional Chinese medicine. Ophiopogon japonicus is highly valued for its effects of nourishing yin and moistening the lungs, clearing away heat and reducing fire, and moistening dryness and relieving constipation. Modern pharmacological studies have shown that Ophiopogon japonicus has potential application value in the treatment of respiratory diseases, diabetes, cardiovascular diseases, etc. These effects are mainly attributed to the unique ingredients in Ophiopogon japonicus, such as ophiopogon glycosides, mannose and other bioactive substances, which have multiple health-care effects on the body and provide a solid scientific basis for its wide application in the field of traditional Chinese medicine.

[0003] However, the efficacy and composition of Ophiopogon japonicus vary significantly due to different origins. This origin difference not only affects the medicinal value of Ophiopogon japonicus, but also poses a challenge to its quality control and clinical application. In fact, due to the uniqueness of their chemical composition, physical properties and growth habits, various types of Chinese medicinal materials have specific analysis and identification methods adapted to them. As for Ophiopogon japonicus, traditional identification methods, such as microscopic observation and chemical analysis, often rely on experienced professionals and require manual tedious analysis and processing of complex data, which not only consumes a lot of time and manpower, but also has certain limitations in terms of accuracy. It is difficult to meet the current needs for accurate identification and quality control of Ophiopogon japonicus, and more efficient, accurate and objective identification technology is urgently needed. Summary of the invention

[0004] The embodiments of the present application provide a method for identifying Ophiopogon japonicus medicinal materials from different origins, thereby solving the problem that the microscope observation or chemical analysis methods commonly used in the prior art for identifying the origins of Ophiopogon japonicus medicinal materials are time-consuming and labor-intensive, and have limited accuracy.

[0005] In a first aspect, the present application provides a method for identifying Radix Ophiopogonis from different origins, comprising: The spectral data of Radix Ophiopogonis from different origins were collected using infrared spectrometer; Preprocessing the spectral data, wherein the preprocessing method is a standard normal variable transformation method, and the standard normal variable transformation method is used to eliminate the adverse effects of interference factors on the spectral data analysis results; Optimizing the hyperparameters of the initial one-dimensional convolutional neural network model based on the spectral data to obtain a target one-dimensional convolutional neural network; Identifying the spectral data of the to-be-identified Radix Ophiopogonis based on the target one-dimensional convolutional neural network and determining the origin corresponding to the to-be-identified Radix Ophiopogonis, including: Inputting the spectral data of the to-be-identified Radix Ophiopogonis into the input layer of the target one-dimensional convolutional neural network; The convolution layer extracts characteristic information from the spectral data of the to-be-identified Radix Ophiopogonis; The pooling layer reduces the dimension of the feature information output by the convolutional layer; The feature information extracted by the convolution layer and the feature information after dimension reduction by the pooling layer are sent to the fully connected layer for feature fusion; The fused feature information is flattened and mapped to the output layer through linear transformation, and the output layer outputs the classification probability value of the Ophiopogon japonicus medicinal material to be identified to determine the origin corresponding to the Ophiopogon japonicus medicinal material to be identified; wherein the output layer is classified by the Noisy softmax function to output the classification probability value of the Ophiopogon japonicus medicinal material to be identified.

[0006] In combination with the first aspect, in a possible implementation, the hyperparameters of the initial one-dimensional convolutional neural network model are optimized based on the spectral data to obtain a target one-dimensional convolutional neural network, including: The spectral data are grouped into categories based on a preset ratio, the spectral data in each category are randomly divided, and the training set and the test set divided into each category are combined to obtain the training set and the test set after the entire data set is randomly divided into layers; Implementing a training and testing process on the initial one-dimensional convolutional neural network model based on the training set and the test set to optimize the hyperparameters of the initial one-dimensional convolutional neural network model; The optimized initial one-dimensional convolutional neural network model is determined as the target one-dimensional convolutional neural network.

[0007] In combination with the first aspect, in a possible implementation method, when training the initial one-dimensional convolutional neural network model, the activation function used is the rectified linear unit function, the optimization model selected is the Adam optimizer, and the learning rate is any value of 0.0001, 0.001 or 0.01.

[0008] In combination with the first aspect, in a possible implementation manner, the target one-dimensional convolutional neural network includes an input layer, three convolutional layers connected in sequence, three pooling layers corresponding to the convolutional layers, a fully connected layer, and an output layer; Each of the convolutional layers uses a 1×5 convolution kernel, and the numbers of the convolutional kernels in the three convolutional layers are 64, 32, and 16, respectively, and the step size of each convolutional layer is set to 2; The three pooling layers are all processed in a maximum pooling manner.

[0009] In combination with the first aspect, in a possible implementation, when the initial one-dimensional convolutional neural network model is trained, the loss function formula is as follows: ; In the formula, Loss is the loss function, and n is the number of samples in the training set; is the output probability distribution of the i-th sample; and is the output of the i-th node and the j-th node of the model before Noisy softmax; C is the number of categories; is the variance of Gaussian distribution, and the introduced noise follows Gaussian distribution N (0, ); W is the weight matrix; λ is the regularization parameter.

[0010] In a second aspect, an embodiment of the present application provides an electronic 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 computer program, the method for identifying Ophiopogon japonicus medicinal materials from different origins as described in the first aspect or any possible method of the first aspect is implemented.

[0011] In a third aspect, an embodiment of the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying Ophiopogon japonicus medicinal materials from different origins as described in the first aspect or any possible one of the first aspects.

[0012] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects: The embodiment of the present application proposes a method for identifying Ophiopogon japonicus medicinal materials from different origins. The method combines infrared spectroscopy technology and a target one-dimensional convolutional neural network, can learn and analyze a large amount of spectral data, and can accurately capture the overall chemical information characteristics of Ophiopogon japonicus medicinal materials, thereby realizing effective identification of Ophiopogon japonicus medicinal materials from different origins. Compared with traditional microscope observation or chemical analysis methods, this method has higher recognition accuracy. Traditional methods often rely on experienced experts for manual identification, which is costly; while this method uses spectral data and deep learning algorithms for automatic identification, significantly reducing manual intervention, thereby reducing the cost of identification. In this application, the original spectral data of Ophiopogon japonicus medicinal materials from different origins are first preprocessed by the standard normal variable transformation method (SNV, Standard Normal Variate Transformation) to eliminate the influence of interference factors in the data. This preprocessing helps the model to more accurately learn the relationship between spectral features and chemical components of Ophiopogon japonicus, thereby obtaining more accurate prediction results. In the output layer of the target one-dimensional convolutional neural network, we use the Noisy softmax function (normalized exponential function with noise) for classification to output the classification probability value of the Ophiopogon japonicus medicinal material to be identified. The introduction of Noisy softmax is mainly to solve the limitations of traditional softmax (normalized exponential function) that may occur in some cases. It increases the randomness of the target one-dimensional convolutional neural network. This randomness helps the target one-dimensional convolutional neural network explore more possibilities during the training process. In complex decision-making environments or when there is uncertainty, the output of traditional softmax may be too "certain", while Noisy softmax makes the output of the target one-dimensional convolutional neural network more flexible and reliable by introducing noise. Therefore, this method for identifying Ophiopogon japonicus medicinal materials from different origins can accurately identify the origin of Ophiopogon japonicus medicinal materials, providing a new solution for the identification of the origin of Ophiopogon japonicus and other Chinese medicinal materials. This method plays an important role in medicinal material quality control and market supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 A flow chart of a method for identifying Radix Ophiopogonis from different origins provided in an embodiment of the present application; Figure 2 and Figure 3The loss value and accuracy change diagrams of the initial one-dimensional convolutional neural network model as the number of iterations increases are respectively provided in the embodiments of the present application; Figure 4 The embodiment of the present application provides that as the number of iterations increases, the data set is changed under different hyperparameters. The initial one-dimensional convolutional neural network model loss value under .

[0015] Figure 5 A structural diagram of a target one-dimensional convolutional neural network provided in an embodiment of the present application; Figure 6 The present invention provides a flowchart for optimizing the hyperparameters of an initial one-dimensional convolutional neural network model based on spectral data to obtain a target one-dimensional convolutional neural network. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0017] In the description of the embodiments of the present application, it should be noted that the orientation or positional relationship indicated by the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the embodiments of the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. The terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance. In addition, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0018] Embodiment 1 The present application embodiment provides a method for identifying Radix Ophiopogonis from different origins, such as Figures 1 to 6 As shown, the method includes the following steps S1-S4: S1: Use infrared spectrometer to collect spectral data of Ophiopogon japonicus from different origins.

[0019] It should be noted that in order to collect and analyze Ophiopogon japonicus samples from five production areas in Zhejiang, Shandong, Guangdong, Shaanxi and Sichuan, we adopted a unified pretreatment process. First, 60 g of each sample was weighed and placed in a petri dish and dried in an oven for 6 hours to remove moisture. After drying, the sample was cooled and ground into powder with a grinder, and then sieved through a 200-mesh sieve to ensure the consistency of particle size. The treated Ophiopogon japonicus powder was mixed and packed into a sealed bag for later use. This pretreatment process has the following advantages: ensuring that the physical form (such as particle size) of Ophiopogon japonicus samples from different production areas is consistent, thereby eliminating the interference caused by morphological differences in subsequent analysis and improving the accuracy and comparability of the analysis; putting the treated Ophiopogon japonicus powder into a sealed bag is not only convenient for storage, but also effectively reduces the deterioration of samples caused by environmental factors (such as humidity, oxygen, etc.); it is convenient for the transportation and distribution of samples, which is conducive to subsequent experiments and research.

[0020] In the present embodiment, we used the German Bruker-VERTEX70 infrared spectrometer to ensure that the collected spectral data has a high degree of accuracy and reliability. Before formally collecting spectral data, we performed background scanning correction to further eliminate the interference that the instrument background may cause to the spectral data, thereby improving the purity of the spectral data. During the spectral data collection process, we set the spectral data scanning times of each sample to 16 times and the resolution to 4 , the spectral wave number range is 500-4000 . This wavenumber range widely covers the vibration absorption peaks of most chemical bonds and functional groups in Ophiopogon japonicus medicinal materials, and can comprehensively and deeply reflect the chemical composition and structural information of the medicinal materials. In order to further improve the repeatability and stability of spectral data, we repeatedly collected spectra for each sample 3 times, and calculated the arithmetic mean of the 3 collection results. This step effectively reduces the random errors that may be caused by a single collection, ensuring the accuracy and reliability of the spectral data. In the end, we collected spectral data of 50 groups of different samples of Ophiopogon japonicus medicinal materials from each origin, and obtained a total of 250 groups of spectral data. These rich data provide a solid foundation for subsequent analysis and modeling. Therefore, the infrared spectral data acquisition method and process used in the embodiments of the present application are scientific, rigorous and efficient, which can ensure the accuracy and reliability of spectral data and provide strong support for subsequent analysis and research.

[0021] The embodiment of the present application aims to explore the influence of the output form and number of labels of the target one-dimensional convolutional neural network (CNN) on the complexity of the model when identifying Ophiopogon japonicus medicinal materials from five origins (Shaanxi, Shandong, Guangdong, Sichuan and Zhejiang). To this end, we specially designed a 1×5 matrix vector M=[M1, M2, M3, M4, M5] as the label of the input sample. Each element in the matrix vector M represents the above five origins in order, and the value of each element can only be 1 or 0. Specifically, when the value of an element is 1, it means that the input sample belongs to the corresponding origin; when the value of an element is 0, it means that the input sample does not belong to the origin. This label design method is concise and clear, which helps to simplify the complexity of the model.

[0022] S2: Preprocess the spectral data using the standard normal variable transformation method, which is used to eliminate the adverse effects of interference factors on the spectral data analysis results.

[0023] It should be noted that the embodiment of the present application uses the standard normal variable transformation method (SNV) to preprocess the spectral data of Ophiopogon japonicus from different origins. SNV preprocessing is to standardize the spectral data of each sample so that its mean is 0 and its standard deviation is 1. After SNV preprocessing of the original spectral data of Ophiopogon japonicus from different origins, the influence of interference factors in the data is eliminated, and the characteristic information in the spectrum can be better highlighted without losing the original information of the data.

[0024] S3: Based on the spectral data, the hyperparameters of the initial one-dimensional convolutional neural network model are optimized to obtain the target one-dimensional convolutional neural network.

[0025] It should be noted that by optimizing the hyperparameters, the initial one-dimensional convolutional neural network model can better fit the spectral data, thereby improving the prediction accuracy.

[0026] S4: Identifying the spectral data of the to-be-identified Radix Ophiopogonis based on the target one-dimensional convolutional neural network, and determining the origin corresponding to the to-be-identified Radix Ophiopogonis, including the following steps S401-S405: S401: Inputting the spectral data of the Radix Ophiopogonis to be identified into the input layer of the target one-dimensional convolutional neural network.

[0027] S402: The convolution layer extracts characteristic information from the spectral data of the Ophiopogon japonicus medicinal material to be identified. The convolution layer can efficiently extract key characteristic information from the spectral data of the Ophiopogon japonicus medicinal material, and such characteristic information is crucial for distinguishing Ophiopogon japonicus medicinal materials from different origins.

[0028] S403: The pooling layer reduces the dimension of the feature information output by the convolutional layer, which helps to reduce data redundancy, improve computing efficiency, and retain important features.

[0029] S404: The feature information extracted by the convolution layer and the feature information after dimensionality reduction by the pooling layer are sent to the fully connected layer for feature fusion. This step aims to enrich the input features of the model and improve the performance of the target one-dimensional convolutional neural network by combining feature information at different levels.

[0030] S405: Flatten the fused feature information and map it to the output layer through linear transformation, and the output layer outputs the classification probability value of the Ophiopogon japonicus to be identified to determine the origin corresponding to the Ophiopogon japonicus to be identified; wherein the output layer is classified through the Noisy softmax function to output the classification probability value of the Ophiopogon japonicus to be identified.

[0031] It should be noted that this method combines infrared spectroscopy technology and a target one-dimensional convolutional neural network, which can learn and analyze a large amount of spectral data, accurately capture the overall chemical information characteristics of Ophiopogon japonicus medicinal materials, and thus effectively identify Ophiopogon japonicus medicinal materials from different origins. Compared with traditional microscopic observation or chemical analysis methods, this method has higher recognition accuracy. Traditional methods often rely on experienced experts for manual identification, which is costly; while this method uses spectral data and deep learning algorithms for automatic identification, which significantly reduces manual intervention and thus reduces the recognition cost. In this application, the original spectral data of Ophiopogon japonicus medicinal materials from different origins were first preprocessed by standard normal variable transformation (SNV) to eliminate the influence of interference factors in the data. This preprocessing helps the model to more accurately learn the relationship between spectral features and chemical components of Ophiopogon japonicus, thereby obtaining more accurate prediction results. In the output layer of the target one-dimensional convolutional neural network, we use the Noisy softmax function for classification to output the classification probability value of the Ophiopogon japonicus medicinal material to be identified. The introduction of Noisysoftmax is mainly to solve the limitations that may occur in traditional softmax in some cases, which increases the randomness of the target one-dimensional convolutional neural network. This randomness helps the target one-dimensional convolutional neural network explore more possibilities during the training process. In complex decision-making environments or when there is uncertainty, the output of traditional softmax may be too "certain", while Noisy softmax introduces noise to make the output of the target one-dimensional convolutional neural network more flexible and reliable. Therefore, this method for identifying Ophiopogon japonicus from different origins can accurately identify the origin of Ophiopogon japonicus, providing a new solution for the identification of the origin of Ophiopogon japonicus and other Chinese medicinal materials. This method plays an important role in the quality control and market supervision of medicinal materials.

[0032] In the present application embodiment, Figure 6As shown, the hyperparameters of the initial one-dimensional convolutional neural network model are optimized based on the spectral data to obtain the target one-dimensional convolutional neural network, including the following steps: S301: grouping the spectral data into categories based on a preset ratio, randomly dividing the spectral data in each category, merging the training set and the test set divided into each category, and obtaining the training set and the test set after the entire data set is randomly divided into layers.

[0033] Specifically, the present application groups the spectral data into categories according to a preset ratio of 3:2, randomly divides the spectral data in each category using the randperm function (random permutation function), merges the training set and test set divided into each category, and obtains the entire data set that is randomly divided into training set and test set after stratification.

[0034] It should be noted that the random partitioning method helps to reduce the dependence of the initial one-dimensional convolutional neural network model on a specific data order, thereby improving the robustness of the initial one-dimensional convolutional neural network model. Even if the data undergoes certain changes or disturbances during input, the initial one-dimensional convolutional neural network model can maintain stable performance.

[0035] S302: Implementing a training and testing process on the initial one-dimensional convolutional neural network model based on the training set and the test set to optimize the hyperparameters of the initial one-dimensional convolutional neural network model.

[0036] It should be noted that this application uses the training set to train the initial one-dimensional convolutional neural network model, and uses the test set to evaluate the performance of the initial one-dimensional convolutional neural network model. According to the evaluation results, the hyperparameters of the initial one-dimensional convolutional neural network model (such as convolution kernel size, number of convolution layers, learning rate, etc.) are adjusted to optimize the performance of the initial one-dimensional convolutional neural network model.

[0037] S303: Determine the optimized initial one-dimensional convolutional neural network model as the target one-dimensional convolutional neural network.

[0038] It should be noted that after multiple iterations and optimizations, the model with the best performance was determined as the target one-dimensional convolutional neural network for subsequent spectral data analysis or prediction tasks.

[0039] Specifically, reasonable division of training and test sets, as well as optimization of hyperparameters, help reduce the risk of overfitting of the initial one-dimensional convolutional neural network model on the training set, so that the initial one-dimensional convolutional neural network model can also perform well on unseen data.

[0040] In the embodiment of the present application, when the initial one-dimensional convolutional neural network model is trained, the activation function used is the Rectified Linear Unit (ReLU) function. The optimization model selected is the Adam (Adaptive Moment Estimation) optimizer, and the learning rate is any value of 0.0001, 0.001 or 0.01.

[0041] It should be noted that the working principle of the ReLU function is: when the input is greater than 0, the output is equal to the input; when the input is less than or equal to 0, the output is 0. This unique nonlinear characteristic enables the initial one-dimensional convolutional neural network model to approximate any nonlinear function, thereby significantly enhancing the expressive power of the model. At the same time, the Adam optimizer demonstrated its excellent performance during the training process. It can adaptively adjust the learning rate of each parameter based on historical gradient information to ensure that the minimum value of the loss function is found more efficiently during the training process. This adaptive learning rate adjustment mechanism makes the training process more efficient and improves the performance of the model. As one of the most important hyperparameters, the learning rate has a direct impact on the learning speed and training effect of the initial one-dimensional convolutional neural network model. Figure 2 and Figure 3 As shown, this application sets three different learning rate optimization schemes in the experiment, and as the number of iterations increases, 0.0001 is finally selected as the optimized learning rate. This choice was made after multiple experiments and comparisons, aiming to achieve the best training effect and model performance. Therefore, this application improves the training efficiency and performance of the initial one-dimensional convolutional neural network model by adopting the ReLU activation function, Adam optimizer, and reasonable learning rate setting. These technical choices not only help the model to better fit the data during training and reduce the risk of overfitting, but also enable the model to show good generalization ability on unseen data.

[0042] In an embodiment of the present application, the target one-dimensional convolutional neural network includes an input layer, three convolutional layers connected in sequence, three pooling layers corresponding to the convolutional layers, a fully connected layer and an output layer.

[0043] Each convolutional layer uses a 1×5 convolution kernel, and the number of convolution kernels in the three convolutional layers is 64, 32, and 16, respectively, and the step size of each convolutional layer is set to 2.

[0044] The three pooling layers are all processed in the maximum pooling mode.

[0045] It should be noted that if Figure 5As shown in the figure, in the analysis process of this application, each convolution layer operation is followed by batch normalization, which is a step designed to enhance the stability of the target one-dimensional convolutional neural network and improve the training efficiency. Finally, after a series of network layer processing, the Noisy softmax function outputs the classification probability value of the Ophiopogon japonicus to be identified. This implicit data enhancement method improves the generalization ability of the target one-dimensional convolutional neural network.

[0046] In the embodiment of the present application, when the initial one-dimensional convolutional neural network model is trained, the loss function formula is as follows: ; In the formula, Loss is the loss function, and n is the number of samples in the training set; is the output probability distribution of the i-th sample; and is the output of the i-th node and the j-th node of the model before Noisy softmax; C is the number of categories; is the variance of Gaussian distribution, and the introduced noise follows Gaussian distribution N (0, ); W is the weight matrix; λ is the regularization parameter.

[0047] It should be noted that if Figure 4 As shown in the figure, after optimizing the hyperparameter σ, when σ=0.1, Noisy softmax achieved better fitting results than ordinary softmax on all data sets. While keeping other hyperparameters constant, we optimized the standard deviation σ to determine the optimal noise intensity. By adding noise to the initial one-dimensional convolutional neural network model, we take into account the uncertainty of the initial one-dimensional convolutional neural network model, so that the initial one-dimensional convolutional neural network model can take into account a certain degree of randomness when predicting, thereby providing more flexible output when facing uncertain factors. In addition, the use of weight regularization terms helps prevent the initial one-dimensional convolutional neural network model from overfitting during training, further improving the generalization ability of the initial one-dimensional convolutional neural network model.

[0048] An embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for identifying Ophiopogon japonicus medicinal materials from different origins is implemented.

[0049] It should be noted that when the processor executes the computer program stored in the memory, it can make full use of its powerful computing power to quickly analyze and identify the characteristics of Ophiopogon japonicus from different origins, which not only improves the recognition accuracy, but also shortens the recognition time, meeting the efficiency requirements in practical applications.

[0050] An embodiment of the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned method for identifying Ophiopogon japonicus medicinal materials from different origins is implemented.

[0051] It should be noted that the non-transitory computer-readable storage medium has high data security, can prevent data from being maliciously tampered with or leaked, and ensures the security and reliability of the computer program on the storage medium and the method for identifying Ophiopogon japonicus medicinal materials from different origins implemented by it. The computer program on the storage medium can be executed by different computers or processors as long as these computers or processors have the corresponding reading and execution capabilities, which improves the portability and compatibility of the method for identifying Ophiopogon japonicus medicinal materials from different origins.

[0052] according to Figure 5 We built the corresponding target one-dimensional convolutional neural network (CNN) based on the model structure shown in the figure. To evaluate the performance of the model, we compared the softmax ( = 0) and Noisy softmax ( = 0.1) output layer function, a CNN classification model based on the original spectral data and a SNV-CNN classification model based on the spectral data preprocessed by the standard normal variable transformation (SNV) were constructed. The classification accuracy of these four models is shown in Table 1 for intuitive comparison and analysis.

[0053] Table 1 - Classification results accuracy

[0054] According to Table 1, we can see that: The accuracy of the CNN model test set after SNV preprocessing has improved, which proves the role of SNV preprocessing method in improving the performance of CNN model.

[0055] The SNV-CNN model with the Noisy softmax function as the output layer outperforms other classification models, and its test set accuracy reaches 99.69%.

[0056] This embodiment uses a method combining infrared spectroscopy technology with a target one-dimensional convolutional neural network to successfully achieve effective identification of Radix Ophiopogonis from different origins. In the test set, the classification accuracy of the model established by this method is as high as 99.69%. This result fully proves that the combination of infrared spectroscopy technology and a target one-dimensional convolutional neural network is highly feasible and accurate in identifying Radix Ophiopogonis from different origins.

[0057] Embodiment 2 Different from the first embodiment, in step S301, the spectral data are grouped into categories according to a preset ratio of 4:1, and the spectral data in each category are randomly divided using the train_test_split function (training-test set division function), and the training set and test set divided into each category are merged to obtain the training set and test set after the entire data set is randomly divided into layers.

[0058] The remaining steps are the same as those in the first embodiment.

[0059] according to Figure 5 To evaluate the performance of the model, we compared the softmax ( = 0) and Noisy softmax ( =0.1) output layer function, a CNN classification model based on the original spectral data and a SNV-CNN classification model based on the spectral data preprocessed by the standard normal variable transformation (SNV) were constructed respectively. The classification accuracy of these four models is shown in Table 2.

[0060] Table 2 - Classification results accuracy

[0061] According to Table 2, we can see that: The accuracy of the CNN model test set after SNV preprocessing has improved. The SNV-CNN model with the Noisy softmax function as the output layer outperforms other classification models, and its test set accuracy reaches 99.65%.

[0062] In this experiment, the accuracy classification results of the four models on the test set showed a high degree of consistency and small deviation compared with Example 1. This phenomenon fully demonstrates that the use of randomly divided data sets does not significantly interfere with or improve the performance of the model, and further strongly confirms that the combination of infrared spectroscopy technology and target one-dimensional convolutional neural network has practical feasibility and high accuracy in identifying Ophiopogon japonicus from different origins, providing reliable technical support and methodological basis for the accurate identification of Ophiopogon japonicus origin, showing broad application prospects and potential promotion value.

[0063] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from other embodiments.

[0064] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present application.

Claims

1. A method for identifying Radix Ophiopogonis from different origins, characterized in that: include: The spectral data of Radix Ophiopogonis from different origins were collected using infrared spectrometer; Preprocessing the spectral data, wherein the preprocessing method is a standard normal variable transformation method, and the standard normal variable transformation method is used to eliminate the adverse effects of interference factors on the spectral data analysis results; Optimizing the hyperparameters of the initial one-dimensional convolutional neural network model based on the spectral data to obtain a target one-dimensional convolutional neural network; Identifying the spectral data of the to-be-identified Radix Ophiopogonis based on the target one-dimensional convolutional neural network and determining the origin corresponding to the to-be-identified Radix Ophiopogonis, including: Inputting the spectral data of the to-be-identified Radix Ophiopogonis into the input layer of the target one-dimensional convolutional neural network; The convolution layer extracts characteristic information from the spectral data of the to-be-identified Radix Ophiopogonis; The pooling layer reduces the dimension of the feature information output by the convolutional layer; The feature information extracted by the convolution layer and the feature information after dimension reduction by the pooling layer are sent to the fully connected layer for feature fusion; The fused feature information is flattened and mapped to the output layer through linear transformation, and the output layer outputs the classification probability value of the Ophiopogon japonicus medicinal material to be identified to determine the origin corresponding to the Ophiopogon japonicus medicinal material to be identified; wherein the output layer is classified by the Noisy softmax function to output the classification probability value of the Ophiopogon japonicus medicinal material to be identified.

2. The method for identifying Radix Ophiopogonis from different origins according to claim 1, characterized in that: Optimizing the hyperparameters of the initial one-dimensional convolutional neural network model based on the spectral data to obtain a target one-dimensional convolutional neural network, including: The spectral data are grouped into categories based on a preset ratio, the spectral data in each category are randomly divided, and the training set and the test set divided into each category are combined to obtain the training set and the test set after the entire data set is randomly divided into layers; Implementing a training and testing process on the initial one-dimensional convolutional neural network model based on the training set and the test set to optimize the hyperparameters of the initial one-dimensional convolutional neural network model; The optimized initial one-dimensional convolutional neural network model is determined as the target one-dimensional convolutional neural network.

3. The method for identifying Radix Ophiopogonis from different origins according to claim 2, characterized in that: When training the initial one-dimensional convolutional neural network model, the activation function used is the rectified linear unit function, the optimization model selected is the Adam optimizer, and the learning rate is any value of 0.0001, 0.001 or 0.

01.

4. The method for identifying Radix Ophiopogonis from different origins according to claim 1, characterized in that: The target one-dimensional convolutional neural network includes an input layer, three convolutional layers connected in sequence, three pooling layers corresponding to the convolutional layers, a fully connected layer and an output layer; Each of the convolutional layers uses a 1×5 convolution kernel, and the numbers of the convolutional kernels in the three convolutional layers are 64, 32, and 16, respectively, and the step size of each convolutional layer is set to 2; The three pooling layers are all processed in a maximum pooling manner.

5. The method for identifying Radix Ophiopogonis from different origins according to claim 3, characterized in that: When the initial one-dimensional convolutional neural network model is trained, the loss function formula is as follows: ； In the formula, Loss is the loss function, and n is the number of samples in the training set; is the output probability distribution of the i-th sample; and is the output of the i-th node and the j-th node of the model before Noisy softmax; C is the number of categories; is the variance of Gaussian distribution, and the introduced noise follows Gaussian distribution N (0, ); W is the weight matrix; λ is the regularization parameter.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for identifying Ophiopogon japonicus from different origins as described in any one of claims 1 to 5 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying Ophiopogon japonicus medicinal materials from different origins as claimed in any one of claims 1 to 5 is implemented.

Images