Bamboo mechanics rapid lossless grading system based on depth image perception
Through the combination of deep image perception and machine vision technology, high-precision non-destructive detection of bamboo mechanical properties is achieved, the accuracy and efficiency of bamboo detection in the existing technology is solved, and the quality and performance detection capabilities of bamboo products are improved.
Patent Information
- Application Number
- CN202510658661.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-21
- Publication Date
- 2025-09-02
AI Technical Summary
In the prior art, the mechanical properties detection of bamboo mainly relies on destructive mechanical testing machines, making it difficult to achieve high-precision non-destructive testing and grading.
Using a method based on depth image perception, combined with machine vision technology and neural network, the physical characteristics and black box features of bamboo cross-section were extracted, and non-destructive detection was performed through multimodal fusion modules, and the BP neural network model was used to predict the mechanical properties of bamboo.
It realizes high-precision and high-efficiency non-destructive testing of the mechanical properties of bamboo, especially the prediction accuracy of elastic modulus and tensile strength reaches 92.8%, and expands the mechanical properties of bamboo integrated materials, improving the pass rate and durability of bamboo products.
Smart Images
Figure CN120580477A_ABST
Abstract
Description
Technical Field The present application belongs to the technical field of computer vision and bamboo mechanical property detection, and specifically relates to a system for extracting physical features and black box features of bamboo cross sections, and fusing the physical features and black box features to realize bamboo mechanical property detection. Background Art Machine learning is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Simply put, machine vision technology mimics the recognition function of the human eye, parsing visual information through algorithms and machine learning models, enabling computers to identify, process, and interpret physical objects from images or videos. Currently, the performance testing of bamboo is primarily performed through destructive testing using mechanical testing machines. This technology utilizes machine vision to combine the cross-sectional characteristics of bamboo with its mechanical properties, enabling high-precision non-destructive testing and grading of its mechanical properties. Summary of the Invention The present invention aims to solve, at least to some extent, the technical problems existing in the non-destructive testing and grading of bamboo mechanical properties. To this end, the present invention aims to provide a machine vision-based bamboo cross-section feature extraction and mechanical property testing system. The technical solution adopted in the present invention is: A rapid and non-destructive bamboo grading system based on deep image perception, including: The bamboo cross-section acquisition module is used to collect and store the features of the bamboo cross-section to facilitate the subsequent extraction of the bamboo cross-section features. The bamboo cross-section physical feature extraction module is used to extract physical features from the collected bamboo cross-section image data and transfer them to the bamboo cross-section feature fusion module. Preferably, the bamboo cross-section physical feature extraction module further includes: The bamboo cross-section feature recognition module is used to identify the features of the bamboo cross-section, and after identification, the data is transferred to the bamboo cross-section feature fusion module. The vascular bundle volume fraction extraction module extracts features from the collected bamboo cross-sections, binarizes the collected image data using the Oust algorithm, and calculates the ratio of pixel 0 to pixel 1 to obtain the vascular bundle volume fraction. The calculation formula for the vascular bundle volume fraction VF is: Where pix0 is the number of pixel 0s in the cross-sectional feature map, and pix1 is the number of pixel 1s in the cross-sectional feature map. Tlocal = μ·block + 0.5σ·block; where μ is the mean of the sub-blocks and σ is the standard deviation, σ effectively overcomes the problem of uneven illumination in bright-field imaging.
[0001] The vascular bundle number extraction module extracts features from the collected bamboo cross-sections, expands the use of computer morphological operations to merge adjacent pixel blocks, and finally uses aspect ratio filtering rules to remove noise from image features to achieve higher accuracy, ultimately realizing vascular bundle counting.
[0002] The vascular bundle average perimeter extraction module extracts features from the collected bamboo cross-sections and processes the collected image data using edge detection technology. It identifies the edge perimeters of the collected vascular bundles and calculates the pixel size they occupy. The data obtained by the vascular bundle number extraction module can be used to obtain the average perimeter of the vascular bundles. The calculation formula for the average perimeter C of the vascular bundles is: Where BC is the total perimeter of the vascular bundle, and BN is the number of vascular bundles. The Gaussian function of the LoG edge detection algorithm can be expressed as: Where σ controls the receptive field of view. A larger σ results in slower attenuation and a stronger smoothing effect. x and y are the coordinates of the image pixels.
[0003] The bamboo cross-section black box feature extraction module is used to perform black box feature extraction on the collected bamboo cross-section image data and pass it to the bamboo cross-section feature fusion module.
[0004] The system visualization module is used for the demonstration and operation of bamboo cross-section feature extraction and mechanical property prediction system, which can better show how to test the mechanical properties of bamboo.
[0005] Preferably, the bamboo cross-section data acquisition module includes multiple acquisition instruments, such as 3D scanners, cameras, sensors, digital meters, and industrial microscopes, which can provide accurate and reliable data to collect bamboo cross-section data.
[0006] Preferably, the bamboo cross-section data feature extraction module includes multiple modules, a bamboo cross-section image preprocessing module and a bamboo cross-section feature processing module.
[0007] The bamboo cross-section image preprocessing module refers to using the collected bamboo cross-section feature map to perform binarization processing on it, so as to make the characteristics of the bamboo cross-section more obvious.
[0008] The bamboo cross-section feature processing module transmits the preprocessed image to the bamboo cross-section feature extraction module. The feature processing module will process the features of the image and convert it from image data into digital data, which is convenient for subsequent prediction model processing.
[0009] The bamboo cross-section black-box feature extraction module uses the Grad-CAM method to extract black-box features from the collected bamboo cross-section image data and passes the algorithm results to the bamboo cross-section feature fusion module. The Grad-CAM method primarily performs global average pooling on the gradient to obtain a weight for each feature map channel. A larger weight indicates a greater contribution of the feature to the mechanical property grading.
[0010] The bamboo cross-section feature input and self-attention fusion module inputs at least three of the collected physical fingerprint features into the bamboo mechanical properties grading module, or the black box features and physical fingerprint features are fused through the self-attention mechanism and then passed into the bamboo mechanical properties grading module.
[0011] The beneficial effects of the present invention are:
[0012] 1. This paper studies the mechanical properties of bamboo, utilizing advanced machine vision technology and a neural network structure to analyze the properties. The extracted physical characteristics of bamboo are integrated with black-box features via a multimodal fusion module, enabling high-precision and high-efficiency nondestructive testing of bamboo's mechanical properties. This enables nondestructive testing of the elastic modulus and tensile strength of bamboo. The data detected by the system is then verified and analyzed based on the results of a mechanical testing machine. Ultimately, it is confirmed that the system can achieve a prediction accuracy of 92.8% for the elastic modulus and tensile strength.
[0013] 2. Provides a new testing method for bamboo mechanical properties, with expanded application to bamboo laminated timber. This method utilizes machine vision technology and a neural network model to integrate the physical and black-box characteristics of bamboo cross-sections to enable non-destructive testing of bamboo mechanical properties. This method can identify and mitigate performance deficiencies associated with bamboo, improving the pass rate and durability of bamboo products. This provides practical guidance for companies in the early stages of bamboo raw material selection and performance testing of bamboo products. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is the image feature extraction module of the bamboo material mechanical rapid non-destructive grading system based on deep image perception of the present invention
[0015] Figure 2 This is a flowchart of the bamboo material mechanical rapid non-destructive grading system based on depth image perception of the present invention.
[0016] Figure 3This is a flow chart of the bamboo mechanical grading module in the bamboo mechanical rapid non-destructive grading system based on depth image perception of the present invention.
[0017] Figure 4 The predicted results of MOE and MOR of bamboo mechanical properties of the present invention are shown in FIG.
[0018] Figure 5 This is the data acquisition module of the bamboo material mechanical rapid non-destructive grading system based on depth image perception of the present invention.
[0019] In the figure: 1. Lens fixture; 2. Adjustment nut; 3. Rotary adjustment nut; 4. Lift adjustment lever; 5. Base; 6. Industrial microscope DETAILED DESCRIPTION
[0020] The following is a clear and complete description of the technical solutions of the present invention in conjunction with the accompanying drawings. Obviously, the examples described are some examples of the present invention and not all embodiments. Based on the embodiments of the present invention, all other examples obtained by ordinary technicians in the field of the present invention without making any creative efforts are within the scope of protection of the present invention.
[0021] It should be understood that in the embodiments, the functional actions may appear in a different order than in the drawings. The order of operations performed may change, but in essence, there may be concurrent execution or other order of execution of instructions.
[0022] Example 1
[0023] like Figure 1 As shown, the rapid, non-destructive bamboo mechanical grading system based on depth image perception in this embodiment includes a bamboo cross-section acquisition module, a bamboo cross-section physical feature extraction module, a bamboo cross-section black box feature extraction module, a bamboo cross-section feature input and self-attention fusion module, a bamboo mechanical properties grading module, a bamboo cross-section feature recognition module, a system visualization module, a bamboo cross-section image preprocessing module, a bamboo cross-section feature processing module, and a bamboo mechanical properties grading module. The bamboo cross-section acquisition module is primarily used to collect and store bamboo cross-section features to facilitate subsequent extraction of bamboo cross-section features. The bamboo cross-section physical feature extraction module is used to extract physical features from the collected bamboo cross-section image data and pass them to the bamboo cross-section feature input and self-attention fusion module. The bamboo cross-section feature input and self-attention fusion module inputs three of the collected physical fingerprint features into the bamboo mechanical properties grading module. The bamboo mechanical properties grading module processes the data obtained by the feature fusion module and inputs the processed data into an optimized BP neural network model. Through continuous training and iteration of the model, the bamboo mechanical properties of the collected data are ultimately graded.
[0024] Both machine vision technology and machine learning technology use the advanced level of computers to simulate a series of human behaviors. By combining machine learning and machine vision with bamboo performance testing, it is possible to better achieve non-destructive grading of bamboo mechanical properties and truly realize high-precision and high-efficiency real-time detection of bamboo mechanical properties.
[0025] The system visualization module is used for the demonstration and operation of bamboo cross-sectional feature extraction and mechanical property prediction system, which can better show how to use bamboo cross-sectional features to detect bamboo mechanical properties.
[0026] In this embodiment, the bamboo cross-section data feature extraction module mainly includes a bamboo cross-section image preprocessing module and a bamboo cross-section feature processing module. The above two modules are used to implement preprocessing and feature classification analysis of bamboo cross-section features.
[0027] The bamboo cross-section physical feature processing module also includes a vascular bundle volume fraction extraction module, a vascular bundle number extraction module, and a vascular bundle average perimeter extraction module. The vascular bundle volume fraction extraction module extracts features from the collected bamboo cross-section, binarizes the collected image data, and calculates the vascular bundle volume fraction by calculating the ratio of pixel 0 to pixel 1.
[0028] The vascular bundle number extraction module extracts features from the collected bamboo cross-sections, uses computer morphological operations to merge adjacent pixel blocks, and finally uses aspect ratio filtering rules to remove noise from image features to achieve higher accuracy, ultimately realizing vascular bundle counting.
[0029] The vascular bundle average perimeter extraction module extracts features from the collected bamboo cross-sections, processes the collected image data through edge detection technology, identifies the edge perimeter of the collected vascular bundles and calculates the pixel size they occupy. The data obtained by the vascular bundle number extraction module can be used to obtain the average perimeter of the vascular bundles.
[0030] The bamboo cross-section data acquisition module collects data in the following way: using multiple acquisition instruments, such as 3D scanners, cameras, sensors, digital instruments, and industrial microscopes, the collected data is transmitted to the bamboo cross-section image preprocessing module to preprocess and extract and classify the bamboo cross-section features.
[0031] The specific implementation method of the bamboo mechanical properties testing module is as follows: S1, firstly, the bamboo cross-section data collected by the data acquisition module is subjected to image preprocessing and classification processing by the preprocessing module and the cross-section feature recognition module; S2. The vascular bundle volume fraction, number of vascular bundles per unit area, and average circumference of vascular bundles extracted by the bamboo cross-section physical fingerprint feature extraction module are input into the bamboo cross-section feature input and self-attention fusion module; S3. The data received by the bamboo cross-section feature input and self-attention module are input into the bamboo performance grading module, and the bamboo mechanical properties are detected and graded using the trained BP neural network model in the bamboo mechanical properties grading module. After iterative training of the BP neural network, the fitting degree between the target output and the predicted output reaches more than 80% ( Figure 4 ), where the fitting equations are: MOE=7745.935+586.157*VF-67.793*VBA,MOE=7748.2+573.1*VF+21.9*C,MOR=102.4+5.6*VBA+0.7*C,MOR=102.4+5.3*VBA-0.4*VB; Where MOE is the elastic modulus of bamboo, MOR is the static bending strength of bamboo, VBA is the number of vascular bundles per unit area, C is the average circumference of vascular bundles, and VF is the volume fraction of vascular bundles.
[0032] Example 2
[0033] The difference from Example 1 is that in the physical fingerprint feature processing module, Example 1 extracts three physical fingerprint features and inputs them into the bamboo cross-section feature input and self-attention fusion module, while Example 2 inputs all 12 extracted physical fingerprint features into the bamboo cross-section feature input and self-attention fusion module. The subsequent process is the same as that of Example 1.
[0034] Example 3
[0035] The difference from Examples 1 and 2 is that in the physical fingerprint feature processing module and the bamboo cross-section feature input and self-attention fusion module, Example 1 extracts three physical fingerprint features and inputs them into the bamboo cross-section feature input and self-attention fusion module, while Example 2 inputs all 12 extracted physical fingerprint features into the bamboo cross-section feature input and self-attention fusion module. Figure 2 As shown, in Example 3, the extracted physical fingerprint features and black-box fingerprint features are fed into the bamboo cross-section feature input and self-attention fusion module. During the black-box feature extraction process, the Crad-CAM method primarily performs global average pooling on the gradients to obtain weights for each feature map channel. A larger weight indicates a greater contribution of the feature to the mechanical property grading. Finally, the input data is passed to the bamboo performance grading module, where the trained BP neural network model is used to detect and grade the bamboo mechanical properties.
[0036] Example 4
[0037] The difference from Examples 1, 2, and 3 is that there are three physical fingerprint feature processing modules and bamboo cross-section feature input and self-attention fusion modules. Example 1 extracts three physical fingerprint features and inputs them into the bamboo cross-section feature input and self-attention fusion module, while Example 2 inputs all 12 extracted physical fingerprint features into the bamboo cross-section feature input and self-attention fusion module. In Example 3, the extracted physical fingerprint features and black box fingerprint features are input into the bamboo cross-section feature input and self-attention fusion module. Figure 4 As shown, Example 4 refers to the extraction of physical fingerprint features and black box features through bamboo cross-section feature input and feature self-attention fusion module into a one-dimensional vector, which is then passed into the bamboo mechanical properties grading module to detect and grade the mechanical properties of bamboo.
[0038] The present invention is not limited to the above-mentioned optional embodiments. Anyone can derive other forms of products under the guidance of the present invention. However, regardless of any changes in shape or structure, all technologies that fall within the scope defined by the patent requirements of the present invention are within the scope of protection of the present invention.
Claims
1. A rapid and non-destructive bamboo mechanics grading system based on deep image perception, characterized by: include: The bamboo cross-section multispectral imaging module, which includes a visible light camera, a near-infrared camera, and a laser 3D scanner, acquires bamboo surface texture, internal density distribution, and 3D topography data, facilitating subsequent extraction of bamboo cross-section features. The bamboo cross-section feature extraction module is used to extract data from the black box characteristics and physical fingerprint features of the collected bamboo cross-section images. The physical fingerprint features include 12 core parameters: bamboo color, color differentiation index, bamboo node spacing, bamboo density, vascular bundle volume fraction, vascular bundle average perimeter, vascular bundle diameter variation coefficient, number of vascular bundles per unit area, void ratio, void fractal dimension, thin-wall cell wall thickness, and fiber orientation angle. The bamboo cross-section feature input and self-attention fusion module inputs at least three of the collected physical fingerprint features into the bamboo mechanical properties grading module, or the black box features and physical fingerprint features are fused through the self-attention mechanism and then passed into the bamboo mechanical properties grading module; The bamboo mechanical properties grading module processes the data obtained through the feature fusion module and finally realizes the detection of the bamboo mechanical properties of the collected data.
2. The rapid non-destructive bamboo mechanical grading system based on depth image perception according to claim 1, wherein the bamboo cross-section feature extraction module specifically comprises: The bamboo cross-section physical feature extraction module is used to extract physical features from the collected bamboo cross-section image data and pass the obtained features, such as vascular bundle volume fraction, average vascular bundle perimeter, number of vascular bundles per unit area, and several other features, to the bamboo cross-section feature input and self-attention fusion module; The bamboo cross-section black box feature extraction module uses the Grad-CAM method to extract black box features from the collected bamboo cross-section image data and pass it to the bamboo cross-section feature fusion module; the Grad-CAM method mainly obtains the weight of each feature map channel by performing global average pooling on the gradient. The larger the weight, the greater the contribution of the feature to the mechanical property grading.
3. The rapid non-destructive bamboo mechanical grading system based on depth image perception according to claim 2, wherein the bamboo cross-section physical feature extraction module specifically comprises: The vascular bundle volume fraction extraction module extracts features from the collected bamboo cross-sections, performs binarization on the collected image data using the Oust algorithm, and calculates the ratio of pixel 0 to pixel 1 in the image to obtain the vascular bundle volume fraction. The vascular bundle number extraction module extracts features from the collected bamboo cross-sections, uses computer morphological operations to merge adjacent pixel blocks, and finally uses aspect ratio filtering rules to remove noise from image features to achieve high accuracy, ultimately realizing vascular bundle counting. The vascular bundle average perimeter extraction module extracts features from the collected bamboo cross-sections and processes the collected image data using LoG edge detection technology. It identifies the edge perimeters of the collected vascular bundles and calculates the pixel size they occupy. The data obtained by the vascular bundle number extraction module can be used to obtain the average perimeter of the vascular bundles. The system visualization module is used for the demonstration and operation of bamboo cross-section feature extraction and mechanical property prediction system, which can better show how to test the mechanical properties of bamboo.
4. The rapid non-destructive mechanical grading system for bamboo materials based on depth image perception according to claim 1, wherein the bamboo cross-section image preprocessing module and the bamboo cross-section feature processing module specifically comprise: The bamboo cross-section image preprocessing module is to use the collected bamboo cross-section feature map to perform binarization processing using the Oust algorithm, so as to make the characteristics of the bamboo cross-section more obvious. The bamboo cross-section feature processing module transmits the preprocessed image to the bamboo cross-section feature extraction module. The feature processing module will process the features of the image and convert it from image data into digital data, which is convenient for subsequent prediction model processing.
5. The rapid non-destructive bamboo mechanical grading system based on deep image perception according to claim 1 is characterized in that the weight distribution formula of bamboo cross-section feature input and self-attention fusion module, physical features and black box features is: Where h i is the i-th eigenvector, is the learnable parameter matrix.
6. The rapid non-destructive bamboo mechanical grading system based on depth image perception according to claim 1 is characterized in that the bamboo mechanical properties grading module specifically comprises: The data obtained through the feature fusion module is processed and divided into a dataset. The resulting dataset is divided into a training set, a validation set, and a test set. A bamboo mechanical properties grading module is constructed. The sample set is a subset of the dataset. Each sample set includes the fused feature values of the bamboo and the mechanical properties of the bamboo obtained through a three-point bending test. The bamboo mechanical properties grading module includes an input layer, a normalization layer, a fully connected layer, a DropOut layer, and a fully connected layer; the input layer is the data after the fusion of physical features and black box features, and the output layer is the mechanical properties of bamboo; the output layer adopts a linear activation function, and the bamboo mechanical properties grading module is an optimized BP neural network model.
7. According to the rapid non-destructive bamboo mechanical grading system based on depth image perception of claim 3, the calculation formula of the vascular bundle volume fraction VF is: ; Where pix0 is the number of pixel 0 in the cross-sectional feature map, and pix1 is the number of pixel 1 in the cross-sectional feature map. The improved Otsu algorithm is used to divide the 1460×380 pixel image into 129×129 sub-blocks, and the local threshold is dynamically calculated. The Otsu binarization formula is: Tlocal = μ·block + 0.5σ·block; where μ is the mean value of the sub-block and σ is the standard deviation. The existence of σ can effectively overcome the problem of uneven illumination in imaging. The vascular bundle average perimeter extraction module is characterized in that: The calculation formula for the average perimeter C of the vascular bundle is: ; Where BC is the total circumference of the vascular bundle, and BN is the number of vascular bundles. The Gaussian function of the LoG edge detection algorithm can be expressed as: where σ controls the receptive field of view. A larger σ results in slower attenuation and a stronger smoothing effect. x and y are the coordinates of the image pixels. The vascular bundle number extraction module adds a calculation process for the number of vascular bundles per unit area, using the bamboo cross-sectional area and the number of vascular bundles measured in advance for calculation. The calculation formula for the number of vascular bundles per unit area VBA is: ; Where BN is the number of vascular bundles and S is the cross-sectional area of bamboo.