A wood density detection method, system, computer device and storage medium

Abstract

The application provides a kind of wood density detection method, system, computer equipment and storage medium, belong to wood detection and analysis field, this method includes: with fast-growing wood and Dalbergia wood as sample, prepare radial section wood chip and constant temperature and humidity treatment;Through terahertz time-domain spectroscopy system acquisition sample time-domain spectrum, obtains frequency domain spectrum by fast fourier transform conversion, extracts the refractive index of 0.2THz-1THz band from frequency domain spectrum as initial feature;73 terahertz refractive indexes are screened out by UVE, after comparing ENR, RF, XGBoost model, select the XGBoost optimized by PSO for training, the contribution of refractive index corresponding to terahertz refractive index to density prediction value is analyzed by combining SHAP value;Test shows that the density prediction R of cross-species reaches 0.9862, nondestructive, efficient detection can be realized, suitable for forestry processing and quality evaluation, and provides an innovative technical path for wood property analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of wood detection and analysis, and particularly relates to a wood density detection method and system, a computer device and a storage medium. BACKGROUND

[0002] Wood density is not only a key parameter for characterizing the physical quality and mechanical properties of wood, but also reflects the growth characteristics of trees. Therefore, accurate determination of wood density is crucial for wood physical property evaluation, material grading, drying process optimization, and product quality control.

[0003] Existing wood density detection methods mainly include drainage method, mechanical stress method, X-ray density detection method, and near-infrared spectroscopy method, etc. However, the existing methods still have limitations such as complex process, sample destructiveness, high environmental sensitivity, and insufficient spatial resolution. The common drainage method measures the weight of the wood sample, puts the wood into water, and calculates the volume of the wood by the mass of the discharged water, and then calculates the density of the wood sample. This method is accurate in measurement, but the operation is tedious and time-consuming, and real-time measurement cannot be achieved. The detection method based on mechanical force, such as Pilodyn, detects the basic density of living trees. Due to the different growth rates of different tree species and the same tree species in different directions, the anisotropy of density makes the detection factors complex, and the detection value of one tree species cannot be applied to other tree species. Although the micro-drill resistance instrument can quickly detect the internal situation of wood, it will leave a through-hole on the surface of the wood during measurement, which belongs to micro-damage detection, and the detection result is affected by factors such as wood diameter. The X-ray density detection method can measure the density of wood, but it has strict requirements for the detection environment. In addition, the X-ray can harm the human body, and safety protection measures need to be taken. In addition, non-destructive detection technologies such as near-infrared spectroscopy can detect wood density, but the detection accuracy needs to be improved, and it is difficult to accurately reflect the subtle structure and density changes inside the wood.

[0004] The existing technology of near-infrared spectroscopy (NIR) combined with machine learning algorithm, which collects the near-infrared spectrum of wood, uses PCA and PLS-R algorithm to construct the model for predicting the density of wood. For example, Alves et al. used NIR and x-ray data to establish a partial least squares regression (PLS-R) model to predict the density of pine wood core. ARRIEL et al. used NIR data to establish a partial least squares regression (PLS-R) model to predict the density of eucalyptus wood. The internal structure of wood is complex, and its internal structure varies greatly in different waveband spectrum absorption and reflection. When relying solely on spectral information to predict wood density, the stability of the model has a significant shortcoming, and the fluctuation range of the prediction result is large, which makes it difficult to ensure continuous and accurate prediction effect, and cannot meet the demand of accurate detection of different types of wood density in actual production. SUMMARY

[0005] In order to solve the problems of low accuracy of wood density prediction result and large loss of prediction process, the present application provides a wood density detection method, system, computer device and storage medium.

[0006] In order to achieve the above-mentioned purpose, the present application provides a wood density detection method, comprising:

[0007] Preparation of the radial section wood sample of the wood to be detected.

[0008] Acquisition of the time-domain spectrum signal of the radial section wood sample by using a terahertz time-domain spectroscopy system; conversion of the time-domain spectrum signal into a frequency domain spectrum and a phase spectrum by using fast Fourier transform (FFT); calculation of the frequency domain spectrum feature of the 0.2THz-1THz frequency band of the radial section wood sample frequency domain spectrum based on the frequency domain spectrum and the phase spectrum; and extraction of the terahertz refractive index of the frequency domain spectrum feature by using uninformative variable elimination (UVE) technology.

[0009] Inputting of the terahertz refractive index into a preset model to obtain a predicted density value of the wood to be detected.

[0010] Preferably, the preset model is an XGBoost model, and before the terahertz refractive index is inputted into the XGBoost model, the method further comprises:

[0011] Preparation of radial section wood samples of low-density fast-growing tree species and high-density Pterocarpus tree species, respectively, and placing the radial section wood samples in a constant temperature and humidity box for determining the real density labels of the samples by using a drainage method;

[0012] Screening of the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species, inputting of the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species and the corresponding real density label of the sample into an elastic net regression (ENR) model, a random forest (RF) model and an extreme gradient boosting (XGBoost) model for training; optimization of the learning rate and the weight of the regularization term of the ENR model by using a particle swarm optimization (PSO) algorithm during the training process; optimization of the maximum depth and the number of trees of the RF model and the XGBoost model by using the PSO algorithm;

[0013] Inputting of the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species into the trained model for prediction, and preliminary screening of the optimal pre-training model as the XGBoost according to the prediction result; and optimization of the optimal pre-training model XGBoost to obtain a final test XGBoost model.

[0014] Preferably, the optimization of the optimal pre-training model XGBoost to obtain the final test XGBoost model comprises:

[0015] Extracting key features of the radial section wood sample of low-density fast-growing tree species and high-density Dalbergia tree species respectively;

[0016] Optimizing the maximum depth and the number of trees of XGBoost by using particle swarm optimization (PSO) algorithm; introducing learning rate to correct the optimal model XGBoost; inputting the key features of the radial section wood sample of low-density fast-growing tree species and high-density Dalbergia tree species into the optimized XGBoost model for training to obtain the training prediction result of wood density;

[0017] Using the training prediction result of wood density and the real density label to calculate the cross-species determination coefficient R² and the root mean square error (RMSE) index respectively, when both R² and RMSE meet the preset threshold, determining that the XGBoost model at this time is the final test XGBoost model.

[0018] Preferably, before extracting the terahertz refractive index of the frequency domain spectrum feature by the UVE, the method further comprises: preprocessing the frequency domain spectrum feature of each radial section wood sample, specifically including: calculating the feature similarity of the frequency domain spectrum features of any two radial section wood samples; and uniformly distributing all samples according to the feature similarity calculation result by using a hierarchical KS algorithm.

[0019] The application also provides a wood density detection system, comprising:

[0020] A sample collection module is configured to prepare a radial section wood sample of wood to be detected.

[0021] A feature extraction module is configured to collect a time domain spectrum signal of the radial section wood sample by using a terahertz time domain spectrum system; convert the time domain spectrum signal into a frequency domain spectrum and a phase spectrum by using a fast Fourier transform (FFT); calculate a frequency domain spectrum feature of a 0.2THz-1THz frequency band of the radial section wood sample based on the frequency domain spectrum and the phase spectrum; and extract a terahertz refractive index of the frequency domain spectrum feature by using an uninformative variable elimination (UVE) technique.

[0022] A model application module is configured to input the terahertz refractive index into a preset model to obtain a predicted density value of the wood to be detected.

[0023] The application also provides a computer device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the wood density detection methods.

[0024] The application also provides a computer readable storage medium, wherein the storage medium stores a computer program, and the computer program can implement the steps of any one of the wood density detection methods when loaded by a processor.

[0025] The wood density detection method provided by the application has the following beneficial effects:

[0026] The application first prepares a radial section wood sample of the wood to be detected; acquires a time-domain spectrum signal of the wood radial section sample, the time-domain spectrum signal can obtain microstructure information of the wood, has a certain penetration to the wood and does not cause physical damage to the wood, avoids damage to the wood sample in the traditional detection method and ensures that the subsequent use value of the wood is not affected; extracts a frequency spectrum feature from the time-domain spectrum signal; inputs the terahertz refractive index of the frequency spectrum feature into a preset model to obtain a predicted density value of the wood to be detected; the frequency domain key feature shows good performance in the detection of wood density of different tree species and effectively improves the accuracy in the cross-tree detection. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the application and the design of the application, the drawings required by the embodiments will be briefly introduced below. The drawings in the following description are only part of the embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the drawings.

[0028] Figure 1 The flowchart of the wood density detection method of the embodiment of the application;

[0029] Figure 2 The flowchart of the wood density detection method model selection and application of the embodiment of the application. DETAILED DESCRIPTION

[0030] In order to make those skilled in the art better understand the technical solutions of the application and can be implemented, the application will be described in detail below in combination with the drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical application of the application, and cannot limit the protection scope of the application.

[0031] As Figure 2As shown, the present application discloses a kind of wood density nondestructive testing technology based on terahertz spectrum and machine learning fusion, to solve the problem of low efficiency, strong destructive of traditional detection method.The research selects fast-growing tree species and high-density tree species of Dalbergia as sample, prepares radial section wood chip, is placed in constant temperature and humidity box and balanced to constant mass, and uses drainage method to determine real density as label value.Through terahertz time-domain spectroscopy system (QT-TRS1000) acquisition sample time-domain spectrum, is converted into frequency spectrum and phase spectrum by fast fourier transform (FFT), and 560 initial characteristics of refractive index in 0.2THz-1THz frequency band are calculated.73 key features are selected from initial features by using uninformative variable elimination (UVE) method, then the performance of three models is compared, and finally XGBoost model optimized by particle swarm optimization (PSO) algorithm is selected for training, and SHAP value quantification analysis is combined to analyze the contribution of key features to prediction results.Test results show that the model performs excellently in cross-species wood density prediction, with a determination coefficient (R²) of 0.9862 and a very low root mean square error, achieving high-precision, nondestructive and efficient wood density detection.This technology can be widely used in forestry processing, wood quality evaluation and other scenes, providing an innovative technical path for wood property analysis, and has important practical value.

[0032] The present application provides a wood density detection method, specifically as Figure 1 shown, comprising:

[0033] Three fast-growing tree species (Chinese fir, poplar and eucalyptus) were prepared for sample preparation, and the wood radial section was prepared according to the specification of 20mm (transverse) x 20mm (longitudinal) x 20mm (chord direction). All samples were placed in a constant temperature and humidity chamber (20±1℃, 50±2% RH) to balance the mass. The basic density of wood was determined according to GB / T1927.5-2021 standard, and the drainage method was used for determination.

[0034] Terahertz spectrum of radial section of wood samples with different densities was collected by terahertz time-domain spectroscopy system (QT-TRS1000). Each sample was scanned five times and the average value was taken to reduce the test error. The time-domain signal and reference signal of the sample were converted into frequency spectrum and phase spectrum by fast fourier transform (FFT), and the reference signal was the sampling signal with air as the medium, which was used as a reference condition to assist the sampling of the sample, so as to calculate the terahertz refractive index of the wood sample , the calculation formula is as follows.

[0035] ;

[0036] Among them, d is the thickness of the sample, c is the speed of light, is the angular frequency, The phase difference is.

[0037] By obtaining the internal microstructure information of wood through terahertz time-domain spectroscopy technology, combining advanced machine learning algorithms for data mining and analysis, an accurate correlation model between wood density and terahertz refractive index can be established. In the cross-species wood density prediction, the R² of the test set can reach 0.9846, and after UVE algorithm feature selection, it is further improved to 0.9862, which is much higher than the accuracy of existing similar technology inventions. Terahertz waves have a certain penetration for wood and do not cause physical damage to wood, avoiding the destruction of wood samples by traditional detection methods and ensuring that the subsequent use value of wood is not affected.

[0038] The spectral data is preprocessed, and the specific steps include:

[0039] In view of the low signal-to-noise ratio characteristics of the light guide antenna in the 0-0.2THz frequency band, the processed sample data set is selected for wave band selection, and the sample refractive index is selected in the range of 0.2THz-1THz as the feature.

[0040] The hierarchical KS method is used for data set sample division, and the training set division ratio is 80%, and the remaining is used as the test. The data set after division of each species sample is combined. The KS algorithm is based on the euclidean metric of sample spectral features, and the sample with the maximum euclidean distance from the existing training set sample is selected by iteration to ensure the uniformity of the sample distribution in the feature space. The unselected samples naturally constitute the test set. The formula for calculating the euclidean distance is as follows:

[0041] ;

[0042] Among them, , M is the number of samples, and are two different samples, N is the number of sample spectral features, i is the index of the number of spectral features in the sample.

[0043] ​According to the extracted characteristic of the wave band range of 0.2THz-1THz, the model is trained by using the training set, and the prediction result of the test set is obtained by using the training set and the trained model. The specific steps include: selecting a plurality of regression models for training the wood density prediction model, the models including elastic network regression (ENR), random forest (RF) and extreme gradient boosting model (XGBoost). The particle swarm optimization algorithm (Particle Swarm Optimization, PSO) is used for hyperparameter optimization of the three models. The PSO algorithm parameters are set as: inertia weight ω=0.8, cognitive factor c1=1.5, and social factor c2=2.3. The test set results are used for model evaluation and performance comparison.

[0044] In the embodiment, particle swarm optimization PSO is used for hyperparameter tuning of models such as ENR, RF and XGBoost. For ENR, the learning rate and the weight ratio of the regularization term are adjusted to control the complexity of the model and balance the bias and variance. For integrated decision tree models such as RF and XGBoost, the main optimized hyperparameters include the number of trees (n_estimators) and the maximum depth (max_depth) for controlling the fitting ability and complexity of the model.

[0045] After the optimal model XGBoost is evaluated, the XGBoost prediction model is optimized by fusing the spectral data of multiple tree species and UVE feature selection. The specific steps are as follows:

[0046] Three high-density Dalbergia wood species, namely, Dalbergia angolensis, Dalbergia odorifera and Dalbergia cochinchinensis, are selected to fuse with the three fast-growing tree wood samples. The data set division method and spectral band selection of the three Dalbergia wood species refer to the fast-growing trees.

[0047] In order to further improve the performance of the XGBoost model, the UVE (Uninformative Variable Elimination) method is used for feature selection to screen out the features that contribute most to the wood density prediction. The UVE method is used to remove redundant and irrelevant features, reduce the complexity of the model, and improve the training efficiency and generalization ability of the model. For the 560 original refractive index feature frequencies in the 0.2THz-1THz frequency band, 87% is determined as the best feature selection threshold after multiple tests, and finally 73 key feature frequencies are selected.

[0048] The XGBoost wood density prediction model is established using the selection feature, in addition to the optimization of the XGBoost hyperparameters using PSO, the learning rate (learning_rate) and subsample ratio (subsample) parameters are introduced to suppress overfitting and improve the generalization performance of the model. The training set is input into the optimized model to obtain the corresponding prediction results, which are compared with the actual measured values to evaluate the prediction accuracy of the model. According to the evaluation indicators, the coefficient of determination (R 2 ), root mean square error (RMSE) and other evaluation indicators, the performance of the model is evaluated. According to the optimized XGBoost model, a fast and high-precision wood density prediction system is established based on terahertz time-domain spectroscopy data, which is suitable for wood density prediction of different tree species, and provides scientific basis for accurate evaluation of forestry resources and wood processing. The research covers fast-growing tree species wood and high-density dalbergia wood and other types, and the model constructed in the wood density detection of different tree species shows good performance, effectively solving the problem of significant decrease in accuracy in cross-species detection of the prior art.

[0049] The refractive index corresponding to the key feature frequency is selected as the key feature by the uninformative variable elimination technology UVE from the frequency domain spectrum feature; further, the SHAP value is used to quantify the contribution value of each frequency point, analyze the influence intensity and direction of the feature on the model output, and identify the key feature frequency which has significant contribution to the prediction result. With the help of SHAP method, the model prediction process can be analyzed in depth, and the contribution of different frequency refractive index to wood density prediction is clear, which provides a strong basis for model optimization and further understanding of the relationship between wood internal structure and density.

[0050] The present application can realize high-precision, non-destructive, fast and universal wood density detection, effectively solve the problems of destructive, low efficiency, poor real-time performance of traditional detection methods, and high cost, insufficient stability and universality of existing similar technical inventions, meet the actual needs of forestry management, wood processing, ecological monitoring and other fields for accurate wood density detection.

[0051] Based on the same inventive concept, the present application also provides a wood density detection system, comprising:

[0052] The sample collection module is used for preparing a radial section wood sample of the wood to be detected.

[0053] The feature extraction module is configured to collect a time-domain spectrum signal of the radial section wood sample by using a terahertz time-domain spectrum system; convert the time-domain spectrum signal into a frequency domain spectrum and a phase spectrum by using fast Fourier transform (FFT); calculate a frequency domain spectrum feature of a 0.2 THz-1 THz frequency band of the wood radial section sample frequency domain spectrum based on the frequency domain spectrum and the phase spectrum; and extract a terahertz refractive index of the frequency domain spectrum feature by using an uninformative variable elimination (UVE) technique.

[0054] The model application module is configured to input the terahertz refractive index into a preset model to obtain a predicted density value of the wood to be detected.

[0055] The application further provides a computer device, which comprises a processor, an internal bus, a network interface, a memory and a nonvolatile memory at a hardware level, and can further comprise other hardware required by a business. The processor reads a corresponding computer program from the nonvolatile memory into the memory and then runs to implement the wood density detection method provided above.

[0056] The application further provides a computer readable storage medium, which stores a computer program, and the computer program can be used to execute the wood density detection method provided above.

[0057] The specific limitations of the wood density detection method calculation system can be referred to the limitations of the wood density detection method provided above, and will not be described here. The modules in the wood density detection system can be realized by software, hardware and a combination thereof in whole or in part. The modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the modules.

[0058] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application. In addition, the above embodiments only express several embodiments of the present application, and the description is more specific and detailed, however, it should not be considered as a limitation on the patent scope. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A method of detecting the density of wood, characterized by, The method comprises the following steps: Preparation of a radial section wood sample of the wood to be detected; Collecting the time-domain spectrum signal of the radial section wood sample by using a terahertz time-domain spectroscopy system; converting the time-domain spectrum signal into a frequency domain spectrum and a phase spectrum by using fast Fourier transform (FFT); calculating the frequency domain spectrum feature of the 0.2-1 THz frequency band of the wood radial section sample based on the frequency domain spectrum and the phase spectrum; Extracting the terahertz refractive index of the frequency domain spectrum feature by using the uninformative variable elimination (UVE) technique; Inputting the terahertz refractive index into a preset model to obtain the predicted density value of the wood to be detected.

2. The method of claim 1, wherein The preset model is an XGBoost model, and before the terahertz refractive index is input into the XGBoost model, the method further comprises the following steps: Preparation of radial section wood samples of low-density fast-growing tree species and high-density Dalbergia tree species, respectively, and placing the radial section wood samples in a constant temperature and humidity box to determine the real density labels of the samples by using a drainage method; Screening the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species, and inputting the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species and the corresponding real density label of the sample into an elastic net regression (ENR) model, a random forest (RF) model and an extreme gradient boosting (XGBoost) model for training; in the training process, the learning rate and the weight of the regularization term of the ENR model are optimized by using a particle swarm optimization (PSO) algorithm; the maximum depth and the number of trees of the RF model and the XGBoost model are optimized by using the PSO algorithm; Inputting the terahertz refractive index of the radial section wood sample of the low-density fast-growing tree species into the trained model for prediction, and preliminarily screening the optimal pre-training model as the XGBoost according to the prediction result; and optimizing the optimal pre-training model XGBoost to obtain a final test XGBoost model.

3. A method of wood density detection according to claim 2, wherein The optimization of the optimal pre-training model XGBoost to obtain the final test XGBoost model comprises the following steps: Extracting the key features of the radial section wood samples of the low-density fast-growing tree species and the high-density Dalbergia tree species, respectively; Optimizing the maximum depth and the number of trees of the XGBoost by using the PSO algorithm; introducing the learning rate to correct the optimal model XGBoost; inputting the key features of the radial section wood samples of the low-density fast-growing tree species and the high-density Dalbergia tree species into the optimized XGBoost model for training to obtain the training prediction result of the wood density; Using the training prediction result of the wood density and the real density label to calculate the cross-species determination coefficient R² and the root mean square error (RMSE) index, respectively; when the R² and the RMSE both satisfy a preset threshold, it is determined that the XGBoost model at this time is the final test XGBoost model.

4. The method of claim 1, wherein Before the terahertz refractive index of the frequency domain spectrum feature is extracted by using the UVE technique, the method further comprises the following steps: preprocessing the frequency domain spectrum feature of each radial section wood sample, specifically comprising: calculating the feature similarity of the frequency domain spectrum features of any two radial section wood samples; and uniformly distributing all the samples according to the feature similarity calculation results by using a hierarchical KS algorithm.

5. A wood density detection system characterized by, The method comprises the following steps: A sample collection module is configured to prepare a radial section wood sample of the wood to be detected; A feature extraction module is configured to collect a time-domain spectrum signal of the radial section wood sample by using a terahertz time-domain spectrum system, convert the time-domain spectrum signal into a frequency-domain spectrum and a phase spectrum by using a fast Fourier transform (FFT), and calculate a frequency-domain spectrum feature of a 0.2 THz-1 THz frequency band of the radial section wood sample based on the frequency-domain spectrum and the phase spectrum; A terahertz refractive index of the frequency-domain spectrum feature is extracted by using an uninformative variable elimination (UVE) technique; A model application module is configured to input the terahertz refractive index into a preset model to obtain a predicted density value of the wood to be detected.

6. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the processor to perform the method of any one of claims 1-5. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is loaded by the processor to implement the steps of the method in any one of claims 1 to 4.

Images