A method for identifying wood tree species based on quantitative anatomy
A quantitative anatomical method using machine learning models for wood species identification addresses the limitations of traditional and advanced methods by providing accurate species-level identification, enhancing reliability and reducing equipment dependency.
Patent Information
- Application Number
- CN202111497274.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-09
- Publication Date
- 2025-07-15
- Estimated Expiration
- 2041-12-09
AI Technical Summary
Traditional wood identification methods have strong subjectivity and low reliability of identification results, making it difficult to achieve accurate identification of wood at the "seed" level. The existing methods are cumbersome and costly, which is not conducive to multi-scenario promotion.
Based on quantitative anatomy, by collecting three-section microscope images of wood specimens, measuring quantitative anatomical characteristic indicators, establishing machine learning models, and analyzing samples to be inspected using the preferred machine learning model to achieve accurate identification of wood tree species.
It has achieved accurate identification of wood tree species to the "species" level, overcomes the subjectivity of traditional methods, has reliable identification results, and is not restricted by specific instruments, and can be widely used in customs law enforcement, quality supervision and laboratory testing.
Smart Images
Figure CN114187591B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wood identification, and particularly relates to a method for identifying wood tree species based on quantitative anatomy. Background Art
[0002] Traditional wood identification methods identify wood tree species by observing the macroscopic and microscopic anatomical characteristics of wood and then comparing them with wood specimens. However, due to human subjectivity, the observed wood anatomical characteristics are generally qualitative rather than quantitative in traditional wood identification methods, resulting in often deviated identification results. At the same time, traditional wood identification methods can generally only identify wood to the "genus" or "category" level and cannot achieve accurate identification of wood at the "species" level.
[0003] In the prior art, chromatography, spectroscopy or DNA barcoding can also be used to identify wood tree species. However, most of these methods require making test samples, setting conditions, detecting through specific laboratory instruments, and screening to obtain the identification results. The process is cumbersome, and subjective judgment is still required in one or more links during the implementation process, which is prone to identification deviation, and the identification cost is high, making it not conducive to popularization and application in the field of multi-scenario wood identification. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for identifying wood tree species based on quantitative anatomy to solve the technical problems of strong subjectivity and low reliability of existing wood identification methods.
[0005] To solve the above technical problems, the present invention specifically provides the following technical solutions:
[0006] A method for identifying wood tree species based on quantitative anatomy, characterized by comprising the following steps:
[0007] Step 100: Obtain a reference data set of wood quantitative anatomy data based on wood specimens;
[0008] Step 200: Train a machine learning model using the reference data set of wood quantitative anatomy data and preferably select the machine learning model with the highest classification accuracy;
[0009] Step 300: Measure and obtain the quantitative anatomy data of the sample to be tested;
[0010] Step 400: Analyze the quantitative anatomy data of the sample to be tested using the selected machine learning model, identify and output the wood tree species of the sample to be tested.
[0011] As a preferred solution of the present invention, in the step 100, the wood specimens are standard samples from a wood specimen museum and correctly named.
[0012] As a preferred embodiment of the present invention, the implementation steps for obtaining the reference dataset of wood quantitative anatomical data include:
[0013] Step 101: Prepare transverse, radial, and tangential section slices of wood tissues based on wood specimens;
[0014] Step 102: Collect microscopic images of the three sections of the slices;
[0015] Step 103: Measure the wood quantitative anatomical characteristic indexes based on the microscopic images of the three sections.
[0016] As a preferred embodiment of the present invention, the quantitative anatomical data in the step 300 is consistent with the wood quantitative anatomical characteristic indexes in the step 103;
[0017] Among them, the wood quantitative anatomical characteristic indexes include: tangential diameter of wood pores, pore frequency, height of wood rays, width of wood rays, linear frequency of wood rays, ratio of axial parenchyma, size of intervessel pits.
[0018] As a preferred embodiment of the present invention, in the step 102, the microscopic images of the three sections are collected by an optical microscope;
[0019] Among them, multiple images of different fields of view are collected for each slice, the magnification of the transverse section is 40 times, and the magnification of the radial and tangential sections is 100 times.
[0020] As a preferred embodiment of the present invention, in the step 103, the Image J software is used to measure the wood quantitative anatomical indexes, and the measurement process includes:
[0021] Step 131: Import the microscopic images of each field of view of the three sections into the Image J software;
[0022] Step 132: Observe and measure the quantitative anatomical index data visually represented in each field of view image;
[0023] Step 133: Calculate the other quantitative anatomical index data indirectly represented based on the visually represented quantitative anatomical data;
[0024] Step 134: Establish the reference dataset of wood quantitative anatomical data based on the visually represented quantitative anatomical index data and the other quantitative anatomical index data indirectly represented;
[0025] Among them, the quantitative anatomical index data represented by each section is the average value of the quantitative anatomical data represented by each field of view image of each section slice, and the reference dataset of wood quantitative anatomical data includes: average value, maximum value, minimum value of each quantitative anatomical index of each wood species, and the index data interval with the maximum value and the minimum value as the interval extreme values.
[0026] As a preferred embodiment of the present invention, the field-of-view image of each slice includes 1 main field of view and 4 sub-fields of view. The main field of view is set at the center of the slice, and the sub-fields of view are one-quarter field-of-view images evenly divided from each complete slice. Moreover, the priority level of the anatomical feature indexes measured by the main field-of-view image is higher than that of the sub-fields of view.
[0027] As a preferred embodiment of the present invention, the quantitative anatomical indexes obtained from the cross-sectional microscopic image include the tangential diameter of the wood pores, the pore frequency, and the proportion of axial parenchyma. The quantitative anatomical indexes extracted from the tangential-sectional microscopic image include the height of the wood ray, the width of the wood ray, and the linear frequency of the wood ray. The quantitative anatomical index obtained from the radial-sectional microscopic image includes the size of the intervessel pits. Moreover, the judgment priority level of the quantitative anatomical feature data characterized by the cross-sectional microscopic image is the highest.
[0028] As a preferred embodiment of the present invention, the machine learning model in step 200 includes models such as artificial neural network, support vector machine, naive Bayes, decision tree, and random forest.
[0029] As a preferred embodiment of the present invention, the screening criteria for the machine learning model are as follows:
[0030] The average value of the recognition accuracies of the quantitative anatomical indexes obtained from the multiple field-of-view images of each slice is used as the recognition accuracy of the quantitative anatomical indexes characterized by each section surface. The weighted average value of the recognition accuracies of the three section surface indexes of the machine learning model is used as the standard for screening the accuracy of the machine learning model.
[0031] The present invention has the following beneficial effects compared with the prior art:
[0032] Based on the wood specimen with correct identification, the present invention obtains the reference data set of wood quantitative anatomy data to ensure the accuracy and reliability of the identification results. Then, according to the obtained reference data set, a machine learning model is trained, and the machine learning model with the highest classification accuracy is preferably used to analyze the anatomical data of the sample to be detected, so as to obtain the identification result of the wood tree species of the sample to be detected, which can realize the accurate identification of wood at the "species" level, overcome the subjectivity of the traditional manual identification method, and the identification result is reliable and well-founded. Compared with some existing detection methods, the technology of the present invention is not limited by specific instruments and can be widely applied in fields such as customs law enforcement, quality supervision, and laboratory testing. Description of the Drawings
[0033] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary. For those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained based on the provided drawings.
[0034] Figure 1 Schematic flow diagram of the wood species identification method provided by the embodiment of the present invention;
[0035] Figure 2 Provided by the embodiment of the present invention Figure 1 Schematic diagram of the implementation steps of step 100 in
[0036] Figure 3 Schematic measurement flow diagram of step 103 provided by the embodiment of the present invention. Detailed implementation manners
[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.
[0038] As Figure 1 shown, the present invention provides a wood species identification method based on quantitative anatomy, including the following steps:
[0039] Step 100: Obtain a reference data set of wood quantitative anatomy data based on wood specimens;
[0040] Step 200: Train a machine learning model using the reference data set of wood quantitative anatomy data, and preferably select the machine learning model with the highest classification accuracy;
[0041] Step 300: Measure and obtain the quantitative anatomy data of the sample to be inspected;
[0042] Step 400: Analyze the quantitative anatomy data of the sample to be inspected using the selected machine learning model, identify and output the wood species of the sample to be inspected.
[0043] In this embodiment, by collecting the three-section images of wood and digitizing the anatomical feature data represented by the three-section images, using the reference data set of anatomical features determined by wood specimens to train a machine learning model, and preferably selecting the optimal machine learning model to automatically identify the wood species. Compared with identifying wood based on subjectively accumulated sensory experience, this implementation manner can achieve the identification of wood at the "species" level.
[0044] In step 100, the wood specimen is a standard sample from a wood herbarium and correctly identified. In this embodiment, an anatomical reference data set for the specimen is established separately to ensure the accuracy and reliability of the data source of the reference data set.
[0045] As Figure 2 shown, the implementation steps for obtaining the reference data set of wood quantitative anatomical data include:
[0046] Step 101: Prepare transverse, radial, and tangential section slices of the wood tissue based on the wood specimen;
[0047] Step 102: Use an optical microscope to collect the microscopic images of the three-section slices;
[0048] Step 103: Measure the wood quantitative anatomical characteristic indexes based on the microscopic images of the three sections.
[0049] Among them, the quantitative anatomical data in step 300 is consistent with the wood quantitative anatomical characteristic indexes in step 103. That is to say, the data collection method for the quantitative anatomical indexes of the sample wood is the same as that for the quantitative anatomical indexes of the specimen wood, and the categories of the collected quantitative anatomical indexes are also the same.
[0050] Among them, the wood quantitative anatomical characteristic indexes include: the tangential diameter of wood pores, pore frequency, height of wood rays, width of wood rays, linear frequency of wood rays, proportion of axial parenchyma, and size of intervessel pits.
[0051] Specifically, the quantitative anatomical indexes obtained from the transverse section microscopic image include the tangential diameter of wood pores, pore frequency, and proportion of axial parenchyma. The quantitative anatomical indexes extracted from the tangential section microscopic image include the height of wood rays, width of wood rays, and linear frequency of wood rays. The quantitative anatomical index obtained from the radial section microscopic image includes the size of intervessel pits.
[0052] Since there are certain differences in the images observed in different fields of view when observing the section images with a microscope, such as differences in the clarity and image morphology of the field of view images obtained by different combinations of eyepiece magnifications and objective magnifications, and differences in the morphology and size of annual rings, wood rays, and pores in the field of view images observed at different parts of the same section. Therefore, in this embodiment, multiple fields of view of the same section are selected for observation, and the average value of the characteristic data measured in multiple fields of view is taken as the anatomical characteristic data represented by this section.
[0053] Randomly determine the number of fields of view for observation and measurement. It is recommended to follow the principle of as many fields of view as possible, non-repeating images, and significant features. For example, select 5 or 10 random fields of view, with a magnification of 40 times for the transverse section and 100 times for the radial and tangential sections.
[0054] In this embodiment, it is assumed that the five fields of view of each section slice include one main field of view and four secondary fields of view. The main field of view is focused on the central area of the slice. The secondary field of view is an image of the field of view divided into one-fourth of each complete slice, and a cross-cross equalization method is preferably used to ensure that the four secondary field of view images do not overlap. The main field of view presents a complete image of the central area of the slice. Multiple fields of view are combined for observation and measurement to obtain characteristic data, making the data source more reliable.
[0055] This embodiment uses the image processing function of Image J to extract feature data from the microscopic image.
[0056] Specifically, in step 103, the quantitative anatomical indexes of the wood are measured using Image J software, such as Figure 3 As shown in , the measurement process includes:
[0057] Step 131: Import the microscopic images of each field of view of the three sections into Image J software;
[0058] Step 132: Observe and measure the quantitative anatomical index data intuitively represented by each field of view image;
[0059] Step 133: Calculating other quantitative anatomical index data represented indirectly based on the quantitative anatomical data represented intuitively;
[0060] Step 134: Establish a wood quantitative anatomical data reference data set based on the intuitively characterized quantitative anatomical index data and the indirectly characterized other quantitative anatomical index data.
[0061] The data represented intuitively mainly refers to the characteristic index data directly observed by the software. For example, the tangential diameter of the wood pores displayed in the cross-section field of view, the height and width of the wood rays displayed in the tangential section field of view, and the size of the inter-vessel pits in the radial section field of view are measured directly with the help of the software. The quantitative anatomical index data represented by each section is the average value of the quantitative anatomical data represented by the images of each field of view of each section. That is to say, each field of view of each section is measured and recorded, and the measurement average value of 5 fields of view is taken as the main parameter.
[0062] Other quantitative anatomical index data characterized indirectly mainly refer to characteristic data obtained through secondary calculation using directly measured data, such as the number of pores in the field of view, combined with the field of view area, to calculate the pore frequency, and the width and height of the wood rays obtained by observation and measurement, combined with the field of view width, to calculate the linear frequency of the wood rays. Similarly, the average value of these indirectly obtained characteristic data is taken as the main reference data.
[0063] In this embodiment, the three-section images of wood are collected, the anatomical features in the three-section images are extracted, and the features are digitalized. However, individual differences generally exist in the same species of wood. If the digitalized feature indexes are derived from a single sample, it will not be conducive to improving the accuracy of the identification results. To improve the reliability of the reference dataset as much as possible, all the observed data are recorded and a database is established, including the mean value, variance, maximum value, minimum value, and the index data interval with the maximum and minimum values as the interval extreme values for each quantitative anatomical index of each wood species. At the same time, the data used to train the machine learning model are the quantitative anatomical data accurately identified to the "species", and the same tree species contains different samples, covering the variability of the wood species, ensuring that the trained model also has a good recognition effect on unknown samples. Moreover, the wood specimen dataset contains multiple quantitative anatomical index values of multiple tree species, with rich and reliable data volume, which helps to accurately and quickly identify a large number of wood species.
[0064] The machine learning model in step 200 includes models such as artificial neural network, support vector machine, naive Bayes, decision tree, and random forest. The reference dataset is used to train various learning models, and the machine learning model with the highest identification accuracy is selected to identify the sample to be detected.
[0065] Considering the differences presented by the three-section features when identifying wood species using the three-section features of wood. For example, the cross-section contains the main anatomical features of the wood, comprehensively reflecting the wood cell features and the mutual connections between cells, and is the most important section for wood identification. Secondly, most features are presented in the tangential section, while fewer features are presented in the radial section. Therefore, this embodiment stipulates that the judgment priority of the quantitative anatomical feature data represented by the cross-section microscopic image is the highest. First, judge the anatomical feature index data shown in the cross-section, which can screen out more tree species first, reducing the subsequent identification workload. Subsequently, according to the anatomical feature index data shown in the tangential section, more tree species are screened out, and finally, the final screening is performed by the anatomical feature index data of the radial section, so as to identify the species name of the tree species.
[0066] There are differences in the identification accuracy of different machine learning models. The screening criteria provided in this embodiment are: the average value of the recognition accuracies of the quantitative anatomical indexes obtained from multiple field images of each section is used as the recognition accuracy of the quantitative anatomical index represented by each section, and the weighted average value of the recognition accuracies of the anatomical indexes of each machine learning model is used as the standard for screening the accuracy of the machine learning model. For example, the weighted proportion of the recognition accuracies of the three sections is 50% for the cross-section, 30% for the tangential section, and 20% for the radial section. The best machine learning model is selected in this way.
[0067] Collect the quantitative anatomical feature data of the sample to be detected, and import the collected data into the selected machine learning model. The machine model compares and analyzes the quantitative anatomical feature data of the sample with the reference data set, and directly identifies and outputs the tree species name.
[0068] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.
Claims
1. A method for identifying wood tree species based on quantitative anatomy, characterized in that, It includes the following steps: Step 100: Obtain a reference dataset of wood quantitative anatomical data based on wood specimens; The implementation steps for obtaining the reference dataset of wood quantitative anatomical data include: Step 101: Prepare transverse, radial, and tangential section slices of wood tissues based on wood specimens; Step 102: Collect microscopic images of the three-section slices; Step 103: Measure wood quantitative anatomical characteristic indexes based on the microscopic images of the three sections; In Step 102, the microscopic images of the three-section slices are collected through an optical microscope, and multiple images of different fields of view are collected for each slice; The quantitative anatomical index data characterized by each section is the average value of the quantitative anatomical data characterized by the images of each field of view of each slice. The reference dataset of wood quantitative anatomical data includes the average value, maximum value, and minimum value of each quantitative anatomical index of each wood species, and the index data interval with the maximum value and the minimum value as the interval extreme values; Step 200: Train a machine learning model using the reference dataset of wood quantitative anatomical data and preferably select the machine learning model with the highest classification accuracy; Step 300: Measure and obtain the quantitative anatomical data of the sample to be tested; Step 400: Analyze the quantitative anatomical data of the sample to be tested using the preferably selected machine learning model, identify and output the wood species of the sample to be tested.
2. The method for identifying wood tree species based on quantitative anatomy according to claim 1, characterized in that, In Step 100, the wood specimens are standard samples from a wood herbarium and correctly named.
3. The method for identifying wood tree species based on quantitative anatomy according to claim 1, wherein, The quantitative anatomical data in Step 300 is consistent with the wood quantitative anatomical characteristic indexes in Step 103; Among them, the wood quantitative anatomical characteristic indexes include: tangential diameter of vessel pores, vessel pore frequency, height of wood rays, width of wood rays, linear frequency of wood rays, proportion of axial parenchyma, and size of intervessel pits.
4. A method for identifying wood tree species based on quantitative anatomy according to claim 1, characterized in that, The magnification of the transverse section is 40 times, and the magnifications of the radial and tangential sections are 100 times.
5. The wood tree species identification method based on quantitative anatomy according to claim 1, characterized in that, In Step 103, Image J software is used to measure the wood quantitative anatomical indexes, and the measurement process includes: Step 131: Import the microscopic images of each field of view of the three sections into the Image J software; Step 132: Observe and measure the quantitative anatomical index data visually characterized by the images of each field of view; Step 133: Calculate other quantitative anatomical index data indirectly characterized based on the quantitatively anatomically data visually characterized; Step 134: Establish the reference dataset of wood quantitative anatomical data based on the quantitatively anatomically index data visually characterized and the other quantitatively anatomically index data indirectly characterized.
6. A method for identifying wood species based on quantitative anatomy according to claim 1, wherein The field-of-view images of each slice include 1 main field of view and 4 sub-fields of view. The main field of view is set at the center of the slice, and the sub-fields of view are quarter-field images evenly divided from each complete slice. Moreover, the priority level of the anatomical characteristic indexes measured from the main field-of-view image is higher than that of the sub-fields of view.
7. A method for identifying wood tree species based on quantitative anatomy according to claim 1, characterized in that, The quantitative anatomical indexes obtained from the transverse section microscopic images include the tangential diameter of wood pores, pore frequency, and the ratio of axial parenchyma. The quantitative anatomical indexes extracted from the tangential section microscopic images include the height of wood rays, the width of wood rays, and the linear frequency of wood rays. The quantitative anatomical indexes obtained from the radial section microscopic images include the size of intervessel pits, and the judgment priority level of the quantitative anatomical feature data characterized by the transverse section microscopic images is the highest.
8. A method for identifying wood tree species based on quantitative anatomy according to claim 7, characterized in that The machine learning model in the step 200 includes models such as artificial neural network, support vector machine, naive Bayes, decision tree, and random forest.
9. The method for identifying wood tree species based on quantitative anatomy according to claim 8, characterized in that, The screening criteria for the machine learning model are as follows: The average value of the recognition accuracy of the quantitative anatomical indexes obtained from the multiple field-of-view images of each section is used as the recognition accuracy of the quantitative anatomical indexes characterized by each section, and the weighted average value of the recognition accuracy of the three-section indexes of the machine learning model is used as the standard for screening the accuracy of the machine learning model.