CN115861710B - Method and device for identifying wood tree species based on multi-source feature fusion - Google Patents

Abstract

The invention discloses a wood species identification method and device based on multi-source feature fusion, and belongs to the technical field of wood identification. The invention simultaneously utilizes the structural characteristics, genetic characteristics and chemical characteristics of wood for tree species identification, covers various types of tree species information of wood, and solves the problem of existing wood identification that lacks reliable feature reference data sets and relies on a single feature for tree species identification results. Unreliable and unsuitable for complex application scenarios, it achieves accurate identification of wood at the "species" level. On the one hand, it improves the identification accuracy, and on the other hand, it improves the stability and applicability of the identification results.

Description

Timber species identification method and device based on multi-source feature fusion

Technical field

The present invention relates to the technical field of wood identification, and in particular, to a wood species identification method and device based on multi-source feature fusion.

Background technique

Wood identification is to compare the sample to be tested with wood specimens containing complete tree taxonomy and collection information, and give the species name of the sample to be tested based on the differences between wood species. Traditional wood identification technology uses the human eye to observe the structural characteristics of the sample to be tested at the macro and micro scales. It mainly relies on the professional knowledge and identification experience of the appraiser. The identification results are greatly affected by human subjective factors, and traditional wood identification In most cases, technology can only identify wood to "genus" or "category" and cannot obtain identification results at the "species" level.

In order to achieve accurate identification of wood at the "species" level, new technologies for wood identification based on computer technology have been developed in recent years. However, although there are some new technologies used to solve the problem of wood species identification, due to the certain degree of intra-species variation in wood, Variability, existing new wood identification technologies all provide tree species identification results based on single characteristics such as anatomy, genetics, chemistry, etc., and the credibility of the identification results is not high. Moreover, in different complex application scenarios, due to various conditions, some features are difficult to obtain, which often makes tree species identification based on this unobtainable single feature difficult to achieve.

Contents of the invention

In order to solve the shortcomings of the existing technology, the present invention provides a method and device for identifying wood species based on multi-source feature fusion, which achieves accurate identification of wood at the "species" level. On the one hand, it improves the identification accuracy, and on the other hand, it improves the identification accuracy. Stability and applicability of results.

The technical solutions provided by the present invention are as follows:

A wood species identification method based on multi-source feature fusion, the method includes:

Establish a wood classification feature reference data set and divide it into a training set and a test set; wherein the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples;

Construct a deep learning model, and use the training set to train the deep learning model;

The key features of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples are respectively extracted through the deep learning model using an independent learning method;

Fusion of key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, using the test set to verify the fusion results, and adjusting and optimizing the deep learning model;

Obtain the wood section image, wood DNA sequence and/or wood chemical fingerprint of the wood sample to be identified, input the wood section image, wood DNA sequence and/or wood chemical fingerprint map into the deep learning model to obtain the wood sample to be identified The name of the tree species.

Further, a wood classification feature reference data set is established and divided into a training set and a test set, including:

Obtaining structural images of wood cross sections, radial sections and chord sections collected from wood specimens, and performing data enhancement processing to obtain the wood section image samples;

Obtain the DNA extracted from the wood specimen, conduct amplification, sequencing and DNA barcode evaluation, screen out effective DNA barcode sequences, and obtain the wood DNA sequence sample;

Obtain the mass spectrometry data collected by scanning the surface of the wood specimen using a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample;

Perform data cleaning on the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a reference data set of wood classification characteristics;

Using the k-fold cross-validation method, the wood classification feature reference data set was divided into the training set and the test set at a ratio of 8:2.

Further, constructing a deep learning model and using the training set to train the deep learning model includes:

Construct a deep learning model based on a convolutional neural network. The deep learning model includes three convolutional neural networks in parallel. Each convolutional neural network includes an input layer, a convolutional layer, an attention mechanism layer, a pooling layer, and Fully connected layer;

The wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set are respectively input into the three convolutional neural networks of the deep learning model for training. In the convolution layer, attention mechanism layer and The pooling layer performs feature extraction, and the fully connected layer performs feature classification.

Further, the key features of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples were respectively extracted through the deep learning model using an autonomous learning method, including:

Extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images respectively, and perform matrix representation to obtain a key feature matrix of the wood section image sample;

The base sequences of the screened effective DNA barcode sequences are represented in a matrix using the k-mer algorithm to obtain the key feature matrix of the wood DNA sequence sample;

Based on the wood chemical fingerprint sample, the structure and content information of the wood characteristic compounds are obtained, and the molecular weight of each characteristic compound is expressed in a matrix to obtain the key feature matrix of the wood chemical fingerprint sample.

Further, the wood cross-section, radial section and chord section structure images all include macro-structure images and micro-structure images, and the data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping;

When obtaining the characteristic compounds of wood, the wood chemical fingerprint sample is subjected to peak alignment processing by traversing the charge-to-mass ratio data, and the characteristic compounds are retrieved by comparing the charge-to-mass ratio.

Further, the wood anatomical structure feature information includes cross-section feature information extracted from the wood cross-section structure image, radial section feature information extracted from the wood radial section structure image, and radial section feature information extracted from the wood chord section structure image. Chord section feature information extracted from the image;

The cross-section characteristic information includes pore frequency characteristics, pore diameter characteristics and/or axial parenchyma frequency characteristics, and the radial section characteristic information includes wood ray cell type characteristics and/or duct-ray pit type characteristics. , the chord section feature information includes wood ray width features, wood ray height features and/or wood ray frequency features.

Further, the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples were fused, and the test set was used to verify the fusion results, and the deep learning model was tested. Adjustments and optimizations include:

Normalize the key feature matrices of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples;

Fusion matrix of key feature matrices of the normalized wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample to obtain a fusion matrix;

Based on the fusion matrix, the identification result is output, the test set is used to test and adjust parameters of the deep learning model, and each parameter of the deep learning model is adjusted and optimized according to the identification result, so that the deep learning model The classification accuracy reaches more than 99%.

Among them, the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples are fused, including:

Superpose the eigenvalues with the same number of rows and columns in the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix;

or;

The key feature matrix of the normalized wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample is subjected to a matrix splicing operation to obtain the fusion matrix.

Further, the step of obtaining the wood section image, wood DNA sequence and/or wood chemical fingerprint of the wood sample to be identified, and inputting the wood section image, wood DNA sequence and/or wood chemical fingerprint into the deep learning model, Obtain the tree species name of the wood sample to be identified, including:

According to the actual application scenario and the data availability of the wood sample to be identified, obtain one or more of the wood cross-section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified;

One or more of the obtained wood section images, wood DNA sequences and wood chemical fingerprints of the wood sample to be identified are input into the deep learning model, and the tree species name of the wood sample to be identified is output.

A wood species identification device based on multi-source feature fusion, the device includes:

A data set creation module is used to establish a wood classification feature reference data set and divide it into a training set and a test set; wherein the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples. ;

A model building module, used to build a deep learning model and use the training set to train the deep learning model;

A feature extraction module, configured to extract key features of the wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample through the deep learning model using an independent learning method;

The model optimization module is used to fuse the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, use the test set to verify the fusion results, and verify the deep learning model Make adjustments and optimizations;

The tree species identification module is used to obtain the wood section image, wood DNA sequence and/or wood chemical fingerprint of the wood sample to be identified, and input the wood section image, wood DNA sequence and/or wood chemical fingerprint into the deep learning model. , get the name of the tree species of the wood sample to be identified.

Further, the data set creation module includes:

A wood section image acquisition unit is used to obtain the wood cross section, radial section and chord section structural images collected from the wood specimen, and perform data enhancement processing to obtain the wood section image sample;

A wood DNA sequence acquisition unit is used to obtain DNA extracted from the wood specimen, perform amplification, sequencing and DNA barcode evaluation, screen out effective DNA barcode sequences, and obtain the wood DNA sequence sample;

A wood chemical fingerprint acquisition unit is used to obtain mass spectrometry data collected by scanning the surface of the wood specimen using a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample;

A data cleaning unit, used to perform data cleaning on the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a reference data set for the wood classification characteristics;

A data set dividing unit is used to divide the wood classification feature reference data set into the training set and the test set in a ratio of 8:2 using a k-fold cross-validation method.

Further, the model building modules include:

A model building unit used to build a deep learning model based on a convolutional neural network. The deep learning model includes three convolutional neural networks in parallel. Each convolutional neural network includes an input layer, a convolutional layer, and an attention mechanism. layer, pooling layer and fully connected layer;

The model training unit is used to input the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set into three convolutional neural networks of the deep learning model for training. In the convolution layer , the attention mechanism layer and the pooling layer are used for feature extraction, and the fully connected layer is used for feature classification.

Further, the feature extraction module includes:

The first extraction unit is used to respectively extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images, and perform matrix representation to obtain a key feature matrix of the wood section image sample;

The second extraction unit is used to matrix-represent the base sequences of the screened effective DNA barcode sequences using the k-mer algorithm to obtain the key feature matrix of the wood DNA sequence sample;

The third extraction unit is used to obtain the structure and content information of wood characteristic compounds based on the wood chemical fingerprint sample, and perform a matrix representation of the molecular weight of each characteristic compound to obtain the key feature matrix of the wood chemical fingerprint sample. .

Further, the wood cross-section, radial section and chord section structure images all include macro-structure images and micro-structure images, and the data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping;

When obtaining the characteristic compounds of wood, the wood chemical fingerprint sample is subjected to peak alignment processing by traversing the charge-to-mass ratio data, and the characteristic compounds are retrieved by comparing the charge-to-mass ratio.

Further, the wood anatomical structure feature information includes cross-section feature information extracted from the wood cross-section structure image, radial section feature information extracted from the wood radial section structure image, and radial section feature information extracted from the wood chord section structure image. Chord section feature information extracted from the image;

The cross-section characteristic information includes pore frequency characteristics, pore diameter characteristics and/or axial parenchyma frequency characteristics, and the radial section characteristic information includes wood ray cell type characteristics and/or duct-ray pit type characteristics. , the chord section feature information includes wood ray width features, wood ray height features and/or wood ray frequency features.

Further, the model optimization module includes:

A normalization unit, used to normalize the key feature matrices of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples;

A feature fusion unit is used to fuse the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix;

A model optimization unit, configured to output identification results based on the fusion matrix, use the test set to test and adjust parameters of the deep learning model, and adjust and optimize each parameter of the deep learning model according to the identification results, so that The classification accuracy of the deep learning model reaches more than 99%.

Further, the feature fusion unit is used for:

Superpose the eigenvalues with the same number of rows and columns in the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix;

or;

The key feature matrix of the normalized wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample is subjected to a matrix splicing operation to obtain the fusion matrix.

Further, the tree species identification module includes:

The data acquisition unit is used to obtain one or more of the wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified based on the actual application scenario and the data availability of the wood sample to be identified;

The tree species identification unit is used to input one or more of the obtained wood section images, wood DNA sequences and wood chemical fingerprints of the wood sample to be identified into the deep learning model, and output the tree species name of the wood sample to be identified.

The invention has the following beneficial effects:

This invention collects biological information such as structural characteristics, genetic characteristics, and chemical characteristics of wood specimens to establish a wood classification feature reference data set of wood section image samples, wood DNA sequence samples, and wood chemical fingerprint samples; builds a deep learning model, and integrates wood Anatomy, genetics and chemical classification feature data information, using three types of data in the wood classification feature reference data set as the training set, extract the key features in the three data types through independent learning methods, and train to be able to independently identify wood tree species deep learning model. On this basis, any one or more combinations of the wood cross-section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified can be input into the trained deep learning model to perform tree species identification. The model The tree species name of the sample to be identified is automatically given.

The invention simultaneously utilizes the structural characteristics, genetic characteristics and chemical characteristics of wood for tree species identification, covers various types of tree species information of wood, and solves the problem of existing wood identification that lacks reliable feature reference data sets and relies on a single feature for tree species identification results. Unreliable and unsuitable for complex application scenarios, it achieves accurate identification of wood at the "species" level. On the one hand, it improves the identification accuracy, and on the other hand, it improves the stability and applicability of the identification results.

Description of the drawings

Figure 1 is a flow chart of the wood species identification method based on multi-source feature fusion of the present invention;

Figure 2 is a schematic diagram of the wood species identification device based on multi-source feature fusion of the present invention.

Detailed ways

In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some of the embodiments of the present invention, but not all of the embodiments. The components of the embodiments of the invention generally described and illustrated in the figures herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the appended drawings is not intended to limit the scope of the claimed invention, but rather to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without any creative work fall within the scope of protection of the present invention.

Embodiments of the present invention provide a wood species identification method based on multi-source feature fusion. As shown in Figure 1, the method includes:

S100: Establish a wood classification feature reference data set and divide it into a training set and a test set; the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples.

This step uses wood specimens as database construction samples to establish a wood classification feature reference data set that contains a series of wood section image samples, wood DNA sequence samples, and wood chemical fingerprint samples. The three types of sample data in the wood classification feature reference data set are all from Accurately named wood specimens with accurate species markings.

S200: Build a deep learning model and use the training set to train the deep learning model.

The deep learning model can be a model based on a convolutional neural network and has an attention mechanism layer, and the deep learning model is trained through a training set.

S300: Use the deep learning model to extract key features of wood cut image samples, wood DNA sequence samples and wood chemical fingerprint samples using independent learning methods.

The key features of wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples are wood structural characteristics, wood genetic characteristics and wood chemical characteristics respectively.

S400: Fusion of the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, using the test set to verify the fusion results, and adjust and optimize the deep learning model.

The key features can be expressed in a matrix form, and can also be fused in a matrix form. The test set can be used for verification to obtain a deep learning model that can independently identify wood species.

S500: Obtain the wood section image, wood DNA sequence and/or wood chemical fingerprint of the wood sample to be identified, input the wood section image, wood DNA sequence and/or wood chemical fingerprint map into the deep learning model to obtain the tree species of the wood sample to be identified. name.

After the deep learning model is trained and optimized, the name of the tree species can be identified based on the deep learning model. For a wood sample to be identified, one or more of the wood cross-section image, wood DNA sequence, and wood chemical fingerprint can be obtained according to the situation. species, input into the deep learning model, and the tree species name of the wood sample to be identified can be automatically output.

The existing technology has not established a reliable reference data set for multi-source classification feature fusion, and cannot integrate comprehensive and complete classification feature information of wood. The credibility of the identification results is not high, and it is not suitable for diversified and complex application scenarios.

This invention collects biological information such as structural characteristics, genetic characteristics, and chemical characteristics of wood specimens to establish a wood classification feature reference data set of wood section image samples, wood DNA sequence samples, and wood chemical fingerprint samples; builds a deep learning model, and integrates wood Anatomy, genetics and chemical classification feature data information, using three types of data in the wood classification feature reference data set as the training set, extract the key features in the three data types through independent learning methods, and train to be able to independently identify wood tree species deep learning model. On this basis, any one or more combinations of the wood cross-section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified can be input into the trained deep learning model to perform tree species identification. The model The tree species name of the sample to be identified is automatically given.

The invention simultaneously utilizes the structural characteristics, genetic characteristics and chemical characteristics of wood for tree species identification, covers various types of tree species information of wood, and solves the problem of existing wood identification that lacks reliable feature reference data sets and relies on a single feature for tree species identification results. Unreliable and unsuitable for complex application scenarios, it achieves accurate identification of wood at the "species" level. On the one hand, it improves the identification accuracy, and on the other hand, it improves the stability and applicability of the identification results.

As an improvement of the embodiment of the present invention, the aforementioned S100 includes:

S110: Obtain the wood cross-section, radial section and chord section structure images collected from the wood specimens, and perform data enhancement processing to obtain wood section image samples.

For example, the iWood wood image acquisition device can be used to collect structural images of transverse, radial, and chordal sections from wood specimens, or an optical microscope can be used to collect structural images of transverse, radial, and chordal sections from wood specimens. The obtained wood The cross-section, radial section and chordal section structural images all include macro-structural images and micro-structural images.

Data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping operations to obtain more image information and enrich the sample diversity of the wood classification feature reference data set.

S120: Obtain the DNA extracted from the wood specimen, conduct amplification, sequencing and DNA barcode evaluation, screen out the effective DNA barcode sequence (i.e. DNA base sequence), and obtain the wood DNA sequence sample.

S130: Obtain the mass spectrum data collected by scanning the surface of the wood specimen with a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample.

The mass spectrometer can use a high-resolution mass spectrometer, such as direct analysis in real time-Fourier transform ion cyclotron resonance mass spectrometer (DART-FTICR-MS). The method of normalization processing can be alignment processing or other processing methods.

S140: Perform data cleaning on wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a wood classification feature reference data set.

S150: Using the k-fold cross-validation method, the wood classification feature reference data set is divided into a training set and a test set in a ratio of 8:2.

K-fold cross-validation first randomly divides the reference data set into k mutually exclusive subsets of the same size. Each time, k-1 parts are randomly selected as the training set, and the remaining 1 part is used as the test set for a round of training. After this round of training is completed, k-1 parts are randomly selected as the training set, and the remaining 1 part is used as the test set for a round of training. After several rounds of training, a loss function is selected to evaluate the optimal model and parameters.

Further, the S200 may include:

S210: Construct a deep learning model based on a convolutional neural network. The deep learning model includes three parallel convolutional neural networks. Each convolutional neural network includes an input layer, a convolutional layer, an attention mechanism layer, a pooling layer, and Fully connected layer.

The three parallel convolutional neural networks are based on images, genetic information and chemical fingerprints.

S220: Input the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set into the three convolutional neural networks of the deep learning model for training, and perform the training on the convolution layer, attention mechanism layer and pooling layer. Feature extraction, feature classification is performed through the fully connected layer.

As a preference, the aforementioned S300 includes:

S310: Extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images respectively, and perform matrix representation to obtain the key feature matrix of the wood section image sample.

The key feature of the wood section image sample is the matrix representation of the wood anatomical structure feature information, which is the wood structure feature matrix, which can also be called the wood image feature. Methods for obtaining information on wood anatomy characteristics include:

1. Extract cross-section feature information from the wood cross-section structure image. The cross-section feature information includes tube hole frequency characteristics, tube hole diameter characteristics, and/or axial parenchyma frequency characteristics, etc.

2. Extract radial section feature information from the wood radial section structure image. The radial section feature information includes wood ray cell type characteristics and/or vessel-ray pit characteristics, etc.

3. Extract chord section feature information from the wood chord section structure image. The chord section feature information includes wood ray width features, wood ray height features, and/or wood ray frequency features.

S320: Use the k-mer algorithm to matrix-represent the base sequences of the screened effective DNA barcode sequences to obtain the key feature matrix of the wood DNA sequence sample.

The key feature matrix of wood DNA sequence samples is the wood genetic characteristics.

S330: Based on the wood chemical fingerprint sample, obtain the structure and content information of the wood characteristic compounds, express the molecular weight of each characteristic compound in a matrix, and obtain the key feature matrix of the wood chemical fingerprint sample.

The key feature matrix of the wood chemical fingerprint sample is the wood chemical characteristics, which is a matrix representation of the wood characteristic compounds. When obtaining the characteristic compounds of wood, you can perform peak alignment processing on the wood chemical fingerprint sample by traversing the charge-to-mass ratio data, and retrieve the characteristic compounds by comparing the charge-to-mass ratio.

As another improvement of the embodiment of the present invention, the aforementioned S400 includes:

S410: Normalize the key feature matrices of wood section image samples, wood DNA sequence samples, and wood chemical fingerprint samples.

The normalization process makes the matrices of the three data types have the same dimensional format and retains their respective important characteristics.

S420: Fusion of key feature matrices of normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix.

The fusion method can be linear weighted fusion or matrix splicing fusion, and finally a wood classification feature fusion matrix is obtained, where:

Linear weighted fusion includes: superimposing the eigenvalues with the same number of rows and columns in the key feature matrices of normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix.

Matrix splicing and fusion includes: performing a matrix splicing operation on the key feature matrices of normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix.

For example, the tf.concat function can be used to concatenate the matrices of the three data types with unified dimensions to obtain a fusion matrix.

S430: Output the identification results in the fully connected layer of the deep learning model based on the fusion matrix, use the test set to test and adjust parameters of the deep learning model, and adjust and optimize the parameters of the deep learning model based on the identification results to improve the classification accuracy of the deep learning model. Reaching more than 99%.

After the above training process is completed, the aforementioned S500 can include:

S510: According to the actual application scenario and the data availability of the wood sample to be identified, obtain one or more of the wood cross-section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified.

S520: Input one or more wood section images, wood DNA sequences and wood chemical fingerprints of the wood sample to be identified into the deep learning model, and output the tree species name of the wood sample to be identified.

An embodiment of the present invention also provides a wood species identification device based on multi-source feature fusion. As shown in Figure 2, the device includes:

Data set establishment module 1 is used to establish a wood classification feature reference data set and divide it into a training set and a test set; among them, the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples.

Model building module 2 is used to build a deep learning model and use the training set to train the deep learning model.

Feature extraction module 3 is used to extract key features of wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples using a deep learning model using an independent learning method.

Model optimization module 4 is used to fuse the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, use the test set to verify the fusion results, and adjust and optimize the deep learning model.

The tree species identification module 5 is used to obtain the wood section image, wood DNA sequence and/or wood chemical fingerprint of the wood sample to be identified, and input the wood section image, wood DNA sequence and/or wood chemical fingerprint to the deep learning model to obtain the wood sample to be identified. Identify the species name of the wood sample.

The invention simultaneously utilizes the structural characteristics, genetic characteristics and chemical characteristics of wood for tree species identification, covers various types of tree species information of wood, and solves the problem of existing wood identification that lacks reliable feature reference data sets and relies on a single feature for tree species identification results. Unreliable and unsuitable for complex application scenarios, it achieves accurate identification of wood at the "species" level. On the one hand, it improves the identification accuracy, and on the other hand, it improves the stability and applicability of the identification results.

As an improvement to the embodiment of the present invention, the data set creation module includes:

The wood section image acquisition unit is used to obtain the wood cross section, radial section and chord section structural images collected from the wood specimen, and perform data enhancement processing to obtain the wood section image sample.

Among them, the structural images of wood cross sections, radial sections and chord sections all include macro structural images and micro structural images, and data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping;

The wood DNA sequence acquisition unit is used to obtain DNA extracted from wood specimens, conduct amplification, sequencing and DNA barcode evaluation, screen out effective DNA barcode sequences, and obtain wood DNA sequence samples.

The wood chemical fingerprint acquisition unit is used to obtain the mass spectrum data collected by scanning the surface of the wood specimen with a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample.

The data cleaning unit is used to clean the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a wood classification feature reference data set.

The data set division unit is used to divide the wood classification feature reference data set into a training set and a test set in a ratio of 8:2 using the k-fold cross-validation method.

Further, the model building modules include:

The model building unit is used to build a deep learning model based on a convolutional neural network. The deep learning model includes three parallel convolutional neural networks. Each convolutional neural network includes an input layer, a convolutional layer, an attention mechanism layer, Pooling layer and fully connected layer.

The model training unit is used to input the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set into the three convolutional neural networks of the deep learning model for training. In the convolution layer, attention mechanism layer and The pooling layer performs feature extraction, and the fully connected layer performs feature classification.

As a preference, the feature extraction module includes:

The first extraction unit is used to respectively extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images, and perform matrix representation to obtain the key feature matrix of the wood section image sample.

Among them, the wood anatomical structure feature information includes cross-section feature information extracted from the wood cross-section structure image, radial section feature information extracted from the wood radial section structure image, and chord section feature information extracted from the wood chord section structure image.

The cross-section feature information includes pore frequency features, pore diameter features and/or axial parenchyma frequency features, the radial section feature information includes wood ray cell type features and/or duct-ray pit type features, and chord section features. The information includes a wood ray width characteristic, a wood ray height characteristic, and/or a wood ray frequency characteristic.

The second extraction unit is used to matrix-represent the base sequences of the screened effective DNA barcode sequences using the k-mer algorithm to obtain the key feature matrix of the wood DNA sequence sample.

The third extraction unit is used to obtain the structure and content information of wood characteristic compounds based on the wood chemical fingerprint sample, and express the molecular weight of each characteristic compound in a matrix to obtain the key feature matrix of the wood chemical fingerprint sample.

Among them, peak alignment processing can be performed on wood chemical fingerprint samples by traversing charge-to-mass ratio data, and characteristic compounds can be retrieved by comparing charge-to-mass ratios.

As another improvement of the embodiment of the present invention, the model optimization module includes:

The normalization unit is used to normalize the key feature matrices of wood section image samples, wood DNA sequence samples, and wood chemical fingerprint samples.

The feature fusion unit is used to fuse the key feature matrices of normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix.

Among them, the operation of key feature matrix fusion can be:

The fusion matrix is obtained by superimposing the eigenvalues with the same number of rows and columns in the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples;

or;

The key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples are subjected to a matrix splicing operation to obtain a fusion matrix.

The model optimization unit is used to output the identification results based on the fusion matrix, use the test set to test and adjust the parameters of the deep learning model, and adjust and optimize the parameters of the deep learning model based on the identification results, so that the classification accuracy of the deep learning model reaches more than 99%. .

After the above training process is completed, the aforementioned tree species identification module includes:

The data acquisition unit is used to obtain one or more of the wood cross-section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified based on the actual application scenario and the data availability of the wood sample to be identified.

The tree species identification unit is used to input one or more wood section images, wood DNA sequences and wood chemical fingerprints of the wood sample to be identified into the deep learning model, and output the tree species name of the wood sample to be identified.

The implementation principles and technical effects of the device provided by the above embodiments correspond to those of the foregoing method embodiments. For the sake of brief description, for parts not mentioned in the embodiments of the device, please refer to the foregoing method embodiments. Corresponding content. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the modules and units described in the device can be referred to the corresponding processes in the embodiments of the foregoing method, and will not be described again here.

The wood species identification method based on multi-source feature fusion described in the above embodiments provided by the present invention can implement business logic through a computer program and record it on a storage medium. The storage medium can be read and executed by a computer to implement the method in this specification. The effect of the solution described in the embodiment. Therefore, embodiments of the present invention also provide a computer-readable storage medium for wood species identification, including a memory for storing processor-executable instructions. When the instructions are executed by the processor, the multi-source feature-based method including the foregoing embodiments is implemented. Steps of a fused wood species identification method.

The storage medium may include a physical device for storing information. The information is usually digitized and then stored in electronic, magnetic, or optical media. The storage medium may include: devices that use electrical energy to store information, such as various memories, such as RAM, ROM, etc.; devices that use magnetic energy to store information, such as hard disks, floppy disks, magnetic tapes, magnetic core memories, magnetic bubble memories, USB flash drive; a device that uses optical means to store information, such as a CD or DVD. Of course, there are other ways of readable storage media, such as quantum memory, graphene memory, and so on.

The storage medium described above may also include other implementation modes according to the description of the method embodiment. The implementation principle and technical effects of this embodiment are the same as those of the aforementioned method embodiment. For details, please refer to the description of the relevant method embodiment. Here I won’t go into details one by one.

Embodiments of the present invention also provide a device for identifying wood species. The device may be a separate computer, or may include a device using one or more of the methods or one or more embodiments of this specification. Practical operating devices, etc. The device for identifying wood species may include at least one processor and a memory that stores computer-executable instructions. When the processor executes the instructions, it implements any one or more of the steps of the wood species identification method based on multi-source feature fusion. .

The above-mentioned equipment may also include other implementation modes according to the description of the method embodiments. The implementation principle and technical effects produced by this embodiment are the same as those of the aforementioned method embodiments. For details, please refer to the description of the relevant method embodiments, which will not be discussed here. Let’s go over them one by one.

Finally, it should be noted that the above-mentioned embodiments are only specific implementations of the present invention and are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although refer to the foregoing The embodiments illustrate the present invention in detail. Those of ordinary skill in the art should understand that any person familiar with the technical field can still modify the technical solutions recorded in the foregoing embodiments within the technical scope disclosed by the present invention. It is possible to easily think of changes or equivalent substitutions of some of the technical features; however, these modifications, changes or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All are covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (2)

1. A wood species identification method based on multi-source feature fusion, characterized in that the method includes: Establish a wood classification feature reference data set and divide it into a training set and a test set; wherein the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples; Construct a deep learning model, and use the training set to train the deep learning model; The key features of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples are respectively extracted through the deep learning model using an independent learning method; Fusion of key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, using the test set to verify the fusion results, and adjusting and optimizing the deep learning model; Obtain the wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified, input the wood section image, wood DNA sequence and wood chemical fingerprint to the deep learning model to obtain the tree species name of the wood sample to be identified; The wood classification feature reference data set is established and divided into a training set and a test set, including: Obtaining structural images of wood cross sections, radial sections and chord sections collected from wood specimens, and performing data enhancement processing to obtain the wood section image samples; Obtain the DNA extracted from the wood specimen, conduct amplification, sequencing and DNA barcode evaluation, screen out effective DNA barcode sequences, and obtain the wood DNA sequence sample; Obtain the mass spectrometry data collected by scanning the surface of the wood specimen using a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample; Perform data cleaning on the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a reference data set of wood classification characteristics; Using the k-fold cross-validation method, the wood classification feature reference data set is divided into the training set and the test set at a ratio of 8:2; The key features of the wood cross-section image sample, wood DNA sequence sample and wood chemical fingerprint sample were respectively extracted through the deep learning model using an independent learning method, including: Extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images respectively, and perform matrix representation to obtain a key feature matrix of the wood section image sample; The base sequences of the screened effective DNA barcode sequences are represented in a matrix using the k-mer algorithm to obtain the key feature matrix of the wood DNA sequence sample; Based on the wood chemical fingerprint sample, obtain the structure and content information of the wood characteristic compounds, and perform a matrix representation of the molecular weight of each characteristic compound to obtain the key feature matrix of the wood chemical fingerprint sample; fusing the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, using the test set to verify the fusion results, and adjusting and optimizing the deep learning model, include: Normalize the key feature matrices of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples; Fusion matrix of key feature matrices of the normalized wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample to obtain a fusion matrix; Based on the fusion matrix, the identification result is output, the test set is used to test and adjust parameters of the deep learning model, and each parameter of the deep learning model is adjusted and optimized according to the identification result, so that the deep learning model The classification accuracy reaches more than 99%; Constructing a deep learning model and using the training set to train the deep learning model includes: Construct a deep learning model based on a convolutional neural network. The deep learning model includes three convolutional neural networks in parallel. Each convolutional neural network includes an input layer, a convolutional layer, an attention mechanism layer, a pooling layer, and Fully connected layer; The wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set are respectively input into the three convolutional neural networks of the deep learning model for training. In the convolution layer, attention mechanism layer and The pooling layer performs feature extraction, and the fully connected layer performs feature classification; The wood cross-section, radial section and chord section structure images all include macro-structure images and micro-structure images, and the data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping; When obtaining wood characteristic compounds, perform peak alignment processing on the wood chemical fingerprint sample by traversing the charge-to-mass ratio data, and retrieve the characteristic compounds by comparing the charge-to-mass ratio; The wood anatomical structure feature information includes cross-section feature information extracted from the wood cross-section structure image, radial section feature information extracted from the wood radial section structure image, and radial section feature information extracted from the wood chord section structure image. Chord section feature information; The cross-section characteristic information includes pore frequency characteristics, pore diameter characteristics and/or axial parenchyma frequency characteristics, and the radial section characteristic information includes wood ray cell type characteristics and/or duct-ray pit type characteristics. , the chord section characteristic information includes wood ray width characteristics, wood ray height characteristics and/or wood ray frequency characteristics; The fusion of key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples includes: Superpose the eigenvalues with the same number of rows and columns in the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix; or; Perform a matrix splicing operation on the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix; The wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified are obtained, and the wood section image, wood DNA sequence and wood chemical fingerprint are input into the deep learning model to obtain the tree species of the wood sample to be identified. Name, including: According to the actual application scenario and the data availability of the wood sample to be identified, the wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified are obtained; The obtained wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified are input into the deep learning model, and the tree species name of the wood sample to be identified is output. 2. A wood species identification device based on multi-source feature fusion, characterized in that the device includes: A data set creation module is used to establish a wood classification feature reference data set and divide it into a training set and a test set; wherein the wood classification feature reference data set includes wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples. ; A model building module, used to build a deep learning model and use the training set to train the deep learning model; A feature extraction module, configured to extract key features of the wood section image sample, wood DNA sequence sample and wood chemical fingerprint sample through the deep learning model using an independent learning method; The model optimization module is used to fuse the key features of the extracted wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, use the test set to verify the fusion results, and verify the deep learning model Make adjustments and optimizations; The tree species identification module is used to obtain the wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified, and input the wood section image, wood DNA sequence and wood chemical fingerprint to the deep learning model to obtain the wood sample to be identified. The species name of the wood sample; The data set creation module includes: A wood section image acquisition unit is used to obtain the wood cross section, radial section and chord section structural images collected from the wood specimen, and perform data enhancement processing to obtain the wood section image sample; A wood DNA sequence acquisition unit is used to obtain DNA extracted from the wood specimen, perform amplification, sequencing and DNA barcode evaluation, screen out effective DNA barcode sequences, and obtain the wood DNA sequence sample; A wood chemical fingerprint acquisition unit is used to obtain mass spectrometry data collected by scanning the surface of the wood specimen using a mass spectrometer, and perform normalization processing to obtain the wood chemical fingerprint sample; A data cleaning unit, used to perform data cleaning on the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples, remove outliers, and establish a reference data set for the wood classification characteristics; A data set dividing unit is used to divide the wood classification feature reference data set into the training set and the test set in a ratio of 8:2 using a k-fold cross-validation method; The feature extraction module includes: The first extraction unit is used to respectively extract the wood anatomical structure feature information contained in the wood cross-section, radial section and chord section structural images, and perform matrix representation to obtain a key feature matrix of the wood section image sample; The second extraction unit is used to matrix-represent the base sequences of the screened effective DNA barcode sequences using the k-mer algorithm to obtain the key feature matrix of the wood DNA sequence sample; The third extraction unit is used to obtain the structure and content information of wood characteristic compounds based on the wood chemical fingerprint sample, and perform matrix representation of the molecular weight of each characteristic compound to obtain the key feature matrix of the wood chemical fingerprint sample. ; The model optimization module includes: A normalization unit, used to normalize the key feature matrices of the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples; A feature fusion unit is used to fuse the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain a fusion matrix; A model optimization unit, configured to output identification results based on the fusion matrix, use the test set to test and adjust parameters of the deep learning model, and adjust and optimize each parameter of the deep learning model according to the identification results, so that The classification accuracy of the deep learning model reaches more than 99%; The model building blocks include: A model building unit used to build a deep learning model based on a convolutional neural network. The deep learning model includes three convolutional neural networks in parallel. Each convolutional neural network includes an input layer, a convolutional layer, and an attention mechanism. layer, pooling layer and fully connected layer; The model training unit is used to input the wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples of the training set into three convolutional neural networks of the deep learning model for training. In the convolution layer , the attention mechanism layer and the pooling layer perform feature extraction, and the fully connected layer performs feature classification; The wood cross-section, radial section and chord section structure images all include macro-structure images and micro-structure images, and the data enhancement processing includes image rotation, image scaling, image mirroring and/or image cropping; When obtaining wood characteristic compounds, perform peak alignment processing on the wood chemical fingerprint sample by traversing the charge-to-mass ratio data, and retrieve the characteristic compounds by comparing the charge-to-mass ratio; The wood anatomical structure feature information includes cross-section feature information extracted from the wood cross-section structure image, radial section feature information extracted from the wood radial section structure image, and radial section feature information extracted from the wood chord section structure image. Chord section feature information; The cross-section characteristic information includes pore frequency characteristics, pore diameter characteristics and/or axial parenchyma frequency characteristics, and the radial section characteristic information includes wood ray cell type characteristics and/or duct-ray pit type characteristics. , the chord section characteristic information includes wood ray width characteristics, wood ray height characteristics and/or wood ray frequency characteristics; The feature fusion unit is used for: Superpose the eigenvalues with the same number of rows and columns in the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix; or; Perform a matrix splicing operation on the key feature matrices of the normalized wood section image samples, wood DNA sequence samples and wood chemical fingerprint samples to obtain the fusion matrix; The tree species identification module includes: The data acquisition unit is used to obtain the wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified based on the actual application scenario and the data availability of the wood sample to be identified; The tree species identification unit is used to input the obtained wood section image, wood DNA sequence and wood chemical fingerprint of the wood sample to be identified into the deep learning model, and output the tree species name of the wood sample to be identified.