Honeysuckle production place identification method based on first-order difference and Inception V3 model

Abstract

The invention relates to a honeysuckle production place identification method based on a first-order difference and Inception V3 model, and the method comprises the steps: obtaining honeysuckle near infrared spectrum data as sample data, constructing a sample data set, and dividing the sample data set into a training set and a verification set; preprocessing the sample data by adopting first-order difference to obtain honeysuckle flower one-dimensional near infrared spectrum data characteristics; the method comprises the following steps of: constructing a 1D-Inception V3-CNN (Convolutional Neural Network) model, wherein the 1D-Inception V3-CNN model is used for replacing a convolutional layer in a one-dimensional convolutional neural network 1D-CNN with an improved Inception V3 structure; forward and backward training verification is carried out on the constructed 1D-Inception V3-CNN model through the training set and the verification set, and a trained 1D-Inception V3-CNN model is obtained; and the trained 1D-Inception V3-CNN model is utilized to carry out honeysuckle flower production place identification. The method is beneficial to rapid, accurate and efficient identification of honeysuckle from different producing areas.

Description

Technical Field

[0001] The present invention relates to the technical field of Chinese herbal medicine origin identification, and in particular to a method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 models. Background Art

[0002] With the continuous development and expansion of the Chinese herbal medicine market, the quality and safety of Chinese herbal medicines have attracted more and more attention. As a common Chinese herbal medicine, honeysuckle has the effects of clearing away heat and detoxification, anti-inflammatory and swelling, and is widely used in the treatment of colds, fever, sore throats and other diseases. Since honeysuckle from different origins has differences in soil environment, climatic conditions, etc., resulting in differences in the content of active ingredients and efficacy, it is of great practical significance to identify the origin of honeysuckle.

[0003] There are many existing research methods for honeysuckle traceability, including trait identification, partial least squares regression analysis (PLS-DA), DNA detection, etc. However, these methods all have certain shortcomings, such as the subjectivity of trait identification, uncertainty and high sample requirements of partial least squares regression analysis, and high technical requirements and highly complex equipment of DNA detection. Therefore, there is currently no unified, rapid and efficient method for honeysuckle origin traceability.

[0004] In recent years, near-infrared spectroscopy has been widely used in the quality evaluation and origin identification of traditional Chinese medicines due to its rapid, non-destructive, and sample pretreatment-free characteristics. However, traditional near-infrared spectral analysis methods often require complex mathematical transformations and feature extraction steps, and it is difficult to fully utilize the nonlinear information in spectral data, which limits its application in complex sample analysis. In order to solve this problem, deep learning algorithms have gradually been introduced into near-infrared spectral data (NIRS) analysis.

[0005] Convolutional Neural Networks (CNN) is one of the representative algorithms of deep learning. This end-to-end data analysis feature enables the internal operation mechanism of CNN to complete feature extraction and classification at the same time, greatly simplifying the qualitative analysis process of NIRS data. Due to the powerful feature extraction ability and excellent classification performance of CNN, some scholars have applied CNN to the qualitative analysis of NIRS data. Inception is a classic deep convolutional neural network model. It can improve the feature expression ability of the model without significantly increasing the amount of calculation through the design of multi-scale convolution kernels. However, the original Inception model is mainly designed for image classification tasks, and direct application to near-infrared spectral data may cause problems such as overfitting. Summary of the invention

[0006] The purpose of the present invention is to provide a method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model, which is conducive to fast, accurate and efficient identification of honeysuckle from different origins.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is: a method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model, comprising:

[0008] Acquire honeysuckle near-infrared spectroscopy data as sample data, construct a sample data set, and divide it into a training set and a validation set;

[0009] The sample data was preprocessed by first-order difference to obtain the one-dimensional near-infrared spectrum data characteristics of honeysuckle;

[0010] Constructing a 1D-Inception V3-CNN model, wherein the 1D-InceptionV3-CNN model replaces the convolutional layer in the one-dimensional convolutional neural network 1D-CNN with an improved InceptionV3 structure; performing forward and backward training verification on the constructed 1D-Inception V3-CNN model through a training set and a verification set to obtain a trained 1D-Inception V3-CNN model;

[0011] The trained 1D-InceptionV3-CNN model was used to identify the origin of honeysuckle.

[0012] Furthermore, the specific method for obtaining the near-infrared spectrum data of honeysuckle is:

[0013] During the harvest season, i.e. from May to September, honeysuckle samples were collected in the main honeysuckle producing areas in my country, including Henan, Shandong, Hebei, and Chongqing. Each region covered multiple counties and townships. The near-infrared spectrometer was used to obtain near-infrared spectral data of the collected honeysuckle samples.

[0014] Furthermore, the specific method of preprocessing sample data using first-order difference is:

[0015] The near-infrared spectral data of honeysuckle is represented as a vector X=[x1,x2,…,x n ], where x n Represents the spectral intensity at the nth wavelength; calculates the first-order difference:

[0016] Δx i =x i+1 -x i

[0017] Where Δx irepresents the change in spectral intensity between two adjacent wavelengths, x i represents the spectral intensity at the i-th wavelength, i=1,2,…,n-1;

[0018] Thus, the one-dimensional near-infrared spectral data feature of honeysuckle Δx=[Δx1,Δx2,…,Δx n-1 ], which is an n-1 dimensional vector.

[0019] Furthermore, the specific method of constructing the 1D-InceptionV3-CNN model is:

[0020] First, the original Inception structure is improved to the InceptionV2 structure by using convolution decomposition;

[0021] Secondly, the asymmetric convolution decomposition structure is used to improve the InceptionV2 structure adapted to two-dimensional images into the InceptionV3 structure adapted to one-dimensional data;

[0022] Finally, the convolutional layer in the original one-dimensional convolutional neural network is replaced with the Inception V3 structure to obtain the 1D-InceptionV3-CNN model.

[0023] Furthermore, the original Inception structure includes four components: 1×1 convolution, 3×3 convolution, 5×5 convolution, and 3×3 maximum pooling. The operation results of the four components are combined on the channel, that is, multiple convolution kernels are used to extract information of different scales of the image, and finally fused to obtain a better representation of the image;

[0024] For the original Inception structure, two layers of 3×3 convolution kernels are used to replace one layer of 5×5 convolution kernel to obtain the Inception V1 structure;

[0025] For the Inception V1 structure, delete the 1×1 convolution and 3×3 maximum pooling to obtain the Inception V2 structure;

[0026] For the Inception V2 structure, in order to adapt to one-dimensional data, all 3×3 convolutions are replaced with 1×3 convolutions to obtain the Inception V3 structure.

[0027] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides a method for identifying the origin of honeysuckle based on first-order difference and Inception V3 model, which performs first-order difference preprocessing on honeysuckle near-infrared spectrum data to enhance the difference between spectral curves and reduce the influence of factors such as time, temperature and humidity; at the same time, in order to solve the problems existing in the application of 1D-CNN model to NIRS data, a 1D-InceptionV3-CNN model is constructed so that it can be directly applied to one-dimensional NIRS data and obtain better origin identification effect; the method combines first-order difference preprocessing with improved InceptionV3 structure and is applied to the identification of the origin of Chinese herbal medicine honeysuckle. Compared with traditional identification methods, the method has high accuracy, fast and efficient in identifying honeysuckle from different origins, which can not only meet the actual needs of honeysuckle origin identification and provide technical support for the quality control of honeysuckle, but also can be further promoted to various Chinese herbal medicine fields, providing new ideas and technical means for the origin identification of other Chinese medicinal materials, and has strong practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is the near infrared spectra of honeysuckle samples from different origins in the embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of dividing a sample data set in an embodiment of the present invention;

[0030] Figure 3 It is the original Inception structure block diagram in the embodiment of the present invention;

[0031] Figure 4 is a block diagram of the improved Inception V1 structure in an embodiment of the present invention;

[0032] Figure 5 It is a block diagram of the improved Inception V2 structure in an embodiment of the present invention;

[0033] Figure 6 It is a structural block diagram of the improved Inception V3 in an embodiment of the present invention;

[0034] Figure 7 1D-Inception V3-CNN model architecture diagram according to an embodiment of the present invention;

[0035] Figure 8 It is a forward and backward training flow chart of a convolutional neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0038] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0039] In order to achieve rapid, accurate and efficient identification of honeysuckle from different origins, this embodiment provides a method for identifying the origin of honeysuckle based on first-order difference and Inception V3 model, including:

[0040] S1. Acquire the near-infrared spectral data of honeysuckle as sample data, construct a sample data set, and divide it into a training set and a validation set;

[0041] S2, preprocessing the sample data by first-order difference to obtain the one-dimensional near-infrared spectrum data characteristics of honeysuckle;

[0042] S3, constructing a 1D-Inception V3-CNN model, wherein the 1D-Inception V3-CNN model replaces the convolutional layer in the one-dimensional convolutional neural network 1D-CNN with an improved InceptionV3 structure; performing forward and backward training verification on the constructed 1D-Inception V3-CNN model through a training set and a verification set to obtain a trained 1D-InceptionV3-CNN model;

[0043] S4. Use the trained 1D-Inception V3-CNN model to identify the origin of honeysuckle.

[0044] 1. Collection and division of near infrared spectral data

[0045] During the peak harvest season (May to September), honeysuckle samples were collected from major honeysuckle producing areas in my country, including Shandong, Chongqing, Henan and Hebei. The sample coverage is relatively wide, with each region covering multiple counties and townships.

[0046] Then, a micro near-infrared spectrometer (model: MicroNIR 1700) was used to obtain the near-infrared spectral data of the collected honeysuckle samples, and finally a total of 628 honeysuckle NIRS data were obtained as sample data. Figure 1As shown. A sample data set is constructed using the sample data obtained.

[0047] In the absence of a large amount of sample data, the sample data set is divided into a training set and a validation set at a ratio of 7:3. The sample data set used in this embodiment has 628 sample data from 4 production areas. The specific data division is as follows: Figure 2 shown.

[0048] 2. Preprocessing of sample data

[0049] The near-infrared spectral data of honeysuckle may be affected by trends (such as time trends) and seasonality (such as changes in temperature, humidity, etc.). In order to enhance the differences between spectral curves and remove the trend and seasonal influences of sample data, this embodiment uses first-order differences to preprocess the near-infrared spectral data of honeysuckle.

[0050] The near-infrared spectral data of honeysuckle is represented as a vector X=[x1,x2,…,x n ], where x n Represents the spectral intensity at the nth wavelength. Calculate the first-order difference:

[0051] Δx i =x i+1 -x i

[0052] Where Δx i represents the change in spectral intensity between two adjacent wavelengths, x i represents the spectral intensity at the i-th wavelength, i=1,2,…,n-1. Thus, the one-dimensional near-infrared spectral data feature of honeysuckle is obtained. Δx=[Δx1,Δx2,…,Δx n-1 ], which is an n-1 dimensional vector.

[0053] 3. Construction of 1D-InceptionV3-CNN model

[0054] First, the convolution decomposition is used to improve the original Inception structure to the InceptionV2 structure.

[0055] Secondly, in order to further reduce parameters and streamline the model, the asymmetric convolution decomposition structure is adopted to improve the InceptionV2 structure adapted to two-dimensional images into the InceptionV3 structure adapted to one-dimensional data.

[0056] Finally, the convolutional layer in the original one-dimensional convolutional neural network (1D-CNN) is replaced with the InceptionV3 structure to obtain the 1D-InceptionV3-CNN model, as shown in Figure 7 shown.

[0057] The following is a further explanation of how to obtain the asymmetric convolution improved InceptionV3 structure adapted to one-dimensional data.

[0058] First, the original Inception basic structure includes four components: 1×1 convolution, 3×3 convolution, 5×5 convolution, and 3×3 maximum pooling. The operation results of the four components are combined on the channel, that is, multiple convolution kernels are used to extract information of different scales of the image, and finally fused to obtain a better representation of the image. Figure 3 shown.

[0059] Secondly, the larger the size of the convolution kernel, the more it can expand the receptive field, but at the same time it generates more parameters. However, a convolution kernel that is too small may reduce the feature extraction ability of the model. To solve this problem, a continuous 3×3 convolution kernel is used to replace a single layer of 5×5 convolution kernel. Replacing a layer of 5×5 convolution kernel with three layers of 3×3 convolution kernels reduces the number of parameters by about 0.55 times; replacing a layer of 5×5 convolution kernel with two layers of 3×3 convolution kernels reduces the number of parameters by about 0.72 times. Therefore, this embodiment replaces a layer of 5×5 convolution kernel with two layers of 3×3 convolution kernels. The improved InceptionV1 structure is as follows: Figure 4 shown.

[0060] Again, in the InceptionV1 structure, the main functions of the 1×1 convolution kernel are: 1) to fuse the information of multiple channels, which is suitable for multi-channel image data; 2) to reduce the number of channels of the feature map, thereby reducing the computational complexity. However, since NIRS data has only a single channel, the 1×1 convolution loses its effectiveness in this case. Therefore, in order to further simplify the structure, the Inception structure in this embodiment also deletes the two branches of 1×1 convolution and 3×3 pooling to obtain the InceptionV2 structure. The improved InceptionV2 structure is shown in Figure 1. Figure 5 shown.

[0061] Finally, the asymmetric convolution decomposition is used to improve the InceptionV2 structure. The idea of ​​asymmetric convolution decomposition is to decompose a large symmetric convolution kernel into two or more asymmetric convolution kernels. These asymmetric convolution kernels have different weights in the horizontal and vertical directions, so that the original symmetric convolution operation can be approximated with fewer parameters and computational complexity. Specifically, an n×n symmetric convolution kernel can be decomposed into a combination of two n×1 and 1×n asymmetric convolution kernels. In this embodiment, the 3×3 convolution kernel is replaced by a 1×3 front-to-back connected structure to obtain the Inception V3 structure. The improved InceptionV3 structure is shown in the figure. Figure 6 shown.

[0062] 4. Training and testing of 1D-Inception V3-CNN model

[0063] After building the 1D-InceptionV3-CNN model, the model is trained and verified by forward and backward training, and the results are analyzed. The specific training process is as follows: Figure 8 shown.

[0064] Finally, the trained 1D-Inception V3-CNN model was used to identify the origin of honeysuckle.

[0065] This embodiment also provides a honeysuckle origin identification system based on first-order difference and Inception V3 model, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above method can be implemented.

[0066] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0067] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0068] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0069] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0070] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model, characterized in that: include: Acquire honeysuckle near-infrared spectroscopy data as sample data, construct a sample data set, and divide it into a training set and a validation set; The sample data was preprocessed by first-order difference to obtain the one-dimensional near-infrared spectrum data characteristics of honeysuckle; Constructing a 1D-Inception V3-CNN model, wherein the 1D-InceptionV3-CNN model replaces the convolutional layer in the one-dimensional convolutional neural network 1D-CNN with an improved InceptionV3 structure; performing forward and backward training verification on the constructed 1D-Inception V3-CNN model through a training set and a verification set to obtain a trained 1D-Inception V3-CNN model; The trained 1D-InceptionV3-CNN model was used to identify the origin of honeysuckle.

2. The method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model according to claim 1, characterized in that: The specific method for obtaining the near-infrared spectrum data of honeysuckle is: During the harvest season, i.e. from May to September, honeysuckle samples were collected in the main honeysuckle producing areas in my country, including Henan, Shandong, Hebei, and Chongqing. Each region covered multiple counties and townships. The near-infrared spectrometer was used to obtain near-infrared spectral data of the collected honeysuckle samples.

3. The method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model according to claim 1, characterized in that: The specific method of preprocessing sample data using first-order difference is: The near-infrared spectral data of honeysuckle is represented as a vector X=[x1,x2,…,x n ], where x n Represents the spectral intensity at the nth wavelength; calculates the first-order difference: Δx i =x i+1 -x i Where Δx i represents the change in spectral intensity between two adjacent wavelengths, x i represents the spectral intensity at the i-th wavelength, i=1,2,…,n-1; Thus, the one-dimensional near-infrared spectral data feature of honeysuckle Δx=[Δx1,Δx2,…,Δx n-1 ], which is an n-1 dimensional vector.

4. The method for identifying the origin of honeysuckle based on first-order difference and InceptionV3 model according to claim 1, characterized in that: The specific method of building the 1D-InceptionV3-CNN model is: First, the original Inception structure is improved to the InceptionV2 structure by using convolution decomposition; Secondly, the asymmetric convolution decomposition structure is used to improve the InceptionV2 structure adapted to two-dimensional images into the Inception V3 structure adapted to one-dimensional data; Finally, the convolutional layer in the original one-dimensional convolutional neural network is replaced with the Inception V3 structure to obtain the 1D-InceptionV3-CNN model.

5. The method for identifying the origin of honeysuckle based on first-order difference and Inception V3 model according to claim 4 is characterized in that: The original Inception structure consists of four components: 1×1 convolution, 3×3 convolution, 5×5 convolution, and 3×3 maximum pooling. The operation results of the four components are combined on the channel, that is, multiple convolution kernels are used to extract information of different scales of the image, and finally fused to obtain a better representation of the image; For the original Inception structure, two layers of 3×3 convolution kernels are used to replace one layer of 5×5 convolution kernel to obtain the InceptionV1 structure; For the Inception V1 structure, delete the 1×1 convolution and 3×3 maximum pooling to obtain the Inception V2 structure; for the Inception V2 structure, in order to adapt to one-dimensional data, all 3×3 convolutions are replaced with 1×3 convolutions to obtain the InceptionV3 structure.

Images