Wood internal defect nondestructive testing method and system based on forestry CT image

Abstract

The invention relates to the technical field of forestry detection, and discloses a wood internal defect nondestructive testing method and system based on a forestry CT image, and the method comprises the steps: obtaining the projection data of wood through a forestry CT scanning device, reconstructing the projection data through employing a maximum likelihood expectation iteration method, and generating a wood internal density distribution image; histogram equalization and Gamma correction are carried out on the wood internal density distribution image, the contrast of a defect area is enhanced, and a preprocessed image is obtained; inputting the preprocessed image into an improved MobileNetV3 model, generating a pixel-level mask graph of the defect, and marking the position and the range of the defect to obtain a defect area; classifying defect types on the basis of the segmented defect regions, calculating the area, the volume and the density attenuation rate of the defect regions, and outputting a defect severity degree evaluation result through an improved MobileNetV3 model; the detection efficiency is improved, and the wood does not need to be physically damaged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forestry detection, and in particular to a wood internal defect nondestructive detection method and system based on forestry CT images. BACKGROUND

[0002] Wood, as an important natural resource, is widely used in many fields such as building and furniture manufacturing. However, wood often has defects such as insect damage, decay, and cracks inside, which can seriously affect the quality and performance of wood. Traditional wood internal defect detection methods, such as knocking and probe methods, not only have low detection efficiency, but also cause damage to the wood to some extent, which cannot meet the demand for wood nondestructive detection in modern forestry production. With the application of CT technology in the field of forestry, wood internal defect detection based on CT images has become a research hotspot. However, existing methods still have deficiencies in image reconstruction accuracy, defect feature extraction, and multi-scale information fusion, making it difficult to meet the demand for accurate identification of complex defects. SUMMARY

[0003] The present application relates to the technical field of forestry detection, and in particular to a wood internal defect nondestructive detection method and system based on forestry CT images.

[0004] The first aspect of the present application provides a wood internal defect nondestructive detection method based on forestry CT images, which comprises the following steps: Obtain the projection data of the wood using a forestry CT scanning device, reconstruct the projection data using the maximum likelihood expectation iteration method, and generate the internal density distribution image of the wood. Perform histogram equalization and Gamma correction on the internal density distribution image of the wood to enhance the contrast of the defect area and obtain the preprocessed image. Input the preprocessed image into an improved MobileNetV3 model to generate a pixel-level mask image of the defect, label the position and range of the defect, and obtain the defect area. Based on the segmented defect area, classify the defect type, calculate the area, volume, and density decay rate of the defect area, and output the defect severity evaluation result through the improved MobileNetV3 model.

[0005] Optionally, in the first implementation manner of the first aspect of the present application, the step of obtaining the projection data of the wood using the forestry CT scanning device and reconstructing the projection data using the maximum likelihood expectation iteration method to generate the internal density distribution image of the wood comprises: Set the initial iteration number to 100 times, introduce an ordered subset acceleration strategy, and divide the projection data into 8-16 subsets. When the maximum likelihood expectation iteration method is used to reconstruct the projection data, only one subset is processed in each iteration, and each subset is iterated alternately; The convergence threshold is set to 0.001. When the difference between the reconstructed images of two adjacent iterations is less than the convergence threshold, the iteration is stopped and the internal density distribution image of the wood is generated.

[0006] Optionally, in a second implementation of the first aspect of the present invention, performing histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area to obtain a preprocessed image includes: Obtain the number of pixels at each gray level in the wood internal density distribution image and calculate its probability of occurrence; Redistribute the grayscale according to the probability distribution, expand the narrow grayscale range in the original wood internal density distribution image to the grayscale range of 0-255, and enhance the image contrast; The Gamma value is set to 0.5-0.8, and Gamma correction is performed on the image after histogram equalization processing to compress the high grayscale area and expand the low grayscale area to obtain the preprocessed image.

[0007] Optionally, in a third implementation of the first aspect of the present invention, inputting the preprocessed image into the improved MobileNetV3 model, generating a pixel-level mask map of the defect, and marking the position and range of the defect to obtain the defect area includes: The preprocessed image is input into the improved MobileNetV3 model, and a wavelet convolution module is embedded in the backbone network to extract the multi-scale features of the defect through wavelet transform. The AIFI module is introduced to fuse feature maps of different scales through a multi-head attention mechanism. After feature fusion, a residual connection is added to add the original input and the fused features to obtain the fused feature map. The fused feature map is segmented using the adaptive local threshold method, and morphological operations are combined to eliminate noise and extract defect areas. Calculate the SSIM values ​​of the defect area and the normal area. If the SSIM value is lower than 0.8, adjust the adjustment coefficient of the adaptive local threshold method according to the SSIM value and dynamically adjust the segmentation threshold. The encoder-decoder framework of the U-Net++ structure is used to generate pixel-level mask maps of defects and mark the location and range of defects.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, embedding a wavelet convolution module in the backbone network and extracting multi-scale features of defects through wavelet transform includes: Daubechies wavelet is used for wavelet transform to decompose the preprocessed image into low-frequency approximate components and high-frequency detail components. The multi-scale features of defects are extracted through wavelet filters of different scales. Low-pass filtering is used to obtain low-frequency information to capture the overall characteristics of the internal structure of the wood, and high-pass filtering is used to obtain high-frequency information to highlight the detailed features including the edges of the defects.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the encoder-decoder framework using the U-Net++ structure generates a pixel-level mask map of the defect and marks the location and range of the defect, including: The decoder part of the U-Net++ structure upsamples the feature map of the defective area through deconvolution operations, and performs jump connections with the feature map of the corresponding scale in the encoder to fuse the feature information at different levels again to generate a pixel-level mask map of the defect.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, classifying the defect type based on the segmented defect area includes: Based on the segmented defect area, the geometric features, texture features and density features of the defect are extracted as classification features; A support vector machine classifier is adopted, and radial basis function is used as kernel function. The parameters of the support vector machine classifier are optimized through cross validation to classify defects into different types, which at least include voids, decay and insect damage.

[0011] A second aspect of the present invention provides a nondestructive detection system for internal defects of wood based on forestry CT images, the system comprising: An image acquisition module is used to acquire projection data of wood using forestry CT scanning equipment, reconstruct the projection data using the maximum likelihood expectation iteration method, and generate an image of the internal density distribution of the wood; The image preprocessing module is used to perform histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area and obtain the preprocessed image; The defect annotation module is used to input the preprocessed image into the improved MobileNetV3 model to generate a pixel-level mask of the defect, annotate the location and range of the defect, and obtain the defect area; The output module is used to classify the defect type based on the segmented defect area, calculate the area, volume and density attenuation rate of the defect area, and output the defect severity assessment results through the improved MobileNetV3 model.

[0012] The third aspect of the present invention provides a non-destructive detection device for internal defects of wood based on forestry CT images, the non-destructive detection device for internal defects of wood based on forestry CT images comprising a memory and at least one processor, the memory storing instructions; the at least one processor calls the instructions in the memory so that the non-destructive detection device for internal defects of wood based on forestry CT images performs each step of the non-destructive detection method for internal defects of wood based on forestry CT images as described in any one of the above items.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, implement the various steps of the method for non-destructive detection of internal defects of wood based on forestry CT images as described in any of the above items.

[0014] In the technical solution provided by the present invention, forestry CT scanning equipment is used to obtain projection data of wood, and the projection data is reconstructed by the maximum likelihood expectation iteration method to generate an image of the internal density distribution of the wood; histogram equalization and Gamma correction are performed on the image of the internal density distribution of the wood to enhance the contrast of the defect area and obtain a preprocessed image; the preprocessed image is input into the improved MobileNetV3 model to generate a pixel-level mask map of the defect, and the position and range of the defect are marked to obtain the defect area; based on the segmented defect area, the defect type is classified, the area, volume and density attenuation rate of the defect area are calculated, and the defect severity assessment result is output through the improved MobileNetV3 model; the present invention improves the CT image quality, introduces the wavelet convolution module and the AIFI module, enhances the model's recognition ability for complex defects, and realizes the precise positioning and boundary characterization of defects based on the dynamic threshold segmentation of SSIM and the U-Net++ structure, can comprehensively and accurately extract the internal defect characteristics of the wood, significantly improves the accuracy of defect detection, and can quickly process a large number of wood CT Image data greatly improves detection efficiency and meets the demand for rapid wood detection in modern forestry production. It does not require physical damage to the wood and truly realizes non-destructive detection of internal defects in the wood, which helps to improve the utilization rate and economic benefits of the wood. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 A flowchart of a method for nondestructive detection of internal defects in wood based on forestry CT images provided by an embodiment of the present invention; Figure 2A schematic diagram of the structure of a nondestructive detection system for internal defects of wood based on forestry CT images provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a nondestructive detection device for internal defects of wood based on forestry CT images provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar items and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or device.

[0018] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 The embodiment of the present invention provides a flowchart of a method for nondestructive detection of internal defects in wood based on forestry CT images. The method specifically includes the following steps: Step 101: Using forestry CT scanning equipment to obtain projection data of wood, reconstructing the projection data using a maximum likelihood prediction iteration method to generate an image of the internal density distribution of the wood; In this example, forestry CT scanning equipment suitable for wood inspection was selected, with a scanning voltage of 120-140 kV and a scanning current of 100-150 mA to ensure effective wood penetration and acquisition of clear projection data. The scanning layer thickness was set to 1-2 mm, and the pitch was controlled between 0.8 and 1.2 to ensure detailed scanning of the wood's internal structure and avoid missing minor defects. During scanning, the wood sample was placed on a rotating stage and scanned 360°, with projection data collected every 1°-2° to ensure comprehensive wood projection information.

[0019] In this example, when reconstructing the projection data using the maximum likelihood prediction iterative method, the initial number of iterations was set to 100. To improve reconstruction efficiency and image quality, an ordered subset acceleration strategy was introduced. The projection data was divided into 8-16 subsets, and only one subset was processed in each iteration. By alternating between subsets, reconstruction time was significantly shortened while ensuring reconstruction accuracy. Furthermore, a convergence threshold of 0.001 was set. When the difference between the reconstructed images of two consecutive iterations was less than this threshold, the iteration was terminated, and an image of the internal density distribution of the wood was generated.

[0020] Step 102: performing histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area and obtain a preprocessed image; In this embodiment, the number of pixels at each grayscale level in the wood internal density distribution image is obtained and the probability of its occurrence is calculated; the grayscale levels are redistributed according to the probability distribution, and the narrow grayscale range in the original wood internal density distribution image is expanded to the grayscale range of 0-255 to enhance the image contrast; the gamma value is set to 0.5-0.8, and gamma correction is performed on the image after histogram equalization processing to compress high grayscale areas and expand low grayscale areas to obtain a preprocessed image.

[0021] In this embodiment, after acquiring an image of the internal density distribution of the wood, the first task is to count the number of pixels at each grayscale level in the image. This step is like classifying and counting all the pixels in the image, recording the number of pixels corresponding to each grayscale level one by one from darkest to lightest. This method can clearly understand the distribution of different grayscale values ​​in the image and then calculate the probability of each grayscale level. The probability calculation is helpful for subsequent analysis of the central tendency and dispersion of grayscale values ​​in the image. Based on the calculated grayscale probability distribution, the next step is to redistribute the grayscale of the image. Since the grayscale range of the original wood internal density distribution image is often narrow, many details may be compressed into a smaller grayscale interval, resulting in unclear distinction between defective areas and normal areas. To improve this situation, the grayscale range of the image needs to be expanded to the entire range of 0-255. This process is similar to guiding vehicles originally crowded on a narrow road to a wide highway, so that pixels of different grayscale values ​​in the image can be more evenly distributed. Through this redistribution, the contrast of the image is significantly enhanced, and the edges and details of the defective area become more prominent; After completing histogram equalization, the image is further gamma-corrected. The key to gamma correction is setting an appropriate gamma value, which is set between 0.5 and 0.8 here. This range is determined based on the characteristics of the wood image and processing requirements, and can effectively adjust the image's brightness and contrast. The principle of gamma correction is to perform a nonlinear transformation on the image's pixel values. When the gamma value is less than 1, the high-grayscale areas in the image are compressed, reducing the brightness of overly bright areas; at the same time, the low-grayscale areas are expanded, increasing the brightness of darker areas. For images of the internal density distribution of wood, after gamma correction, the originally darker defective areas become more clearly visible, while the details of normal areas are not lost due to over-enhancement, resulting in the final pre-processed image.

[0022] Step 103: Input the preprocessed image into the improved MobileNetV3 model to generate a pixel-level mask of the defect, mark the location and range of the defect to obtain the defect area; In this embodiment, the preprocessed image is input into the improved MobileNetV3 model, a wavelet convolution module is embedded in the backbone network, and the multi-scale features of the defect are extracted through wavelet transform; the AIFI module is introduced, and the feature maps of different scales are fused through the multi-head attention mechanism. After the feature fusion, the residual connection is added, and the original input and the fused features are added to obtain the fused feature map; the fused feature map is segmented using the adaptive local threshold method, and the noise is eliminated by combining morphological operations to extract the defect area; the SSIM value of the defect area and the normal area is calculated. If the SSIM value is lower than 0.8, the adjustment coefficient of the adaptive local threshold method is adjusted according to the SSIM value, and the segmentation threshold is dynamically adjusted; the encoder-decoder framework of the U-Net++ structure is used to generate a pixel-level mask map of the defect, and the location and range of the defect are marked.

[0023] In this embodiment, Daubechies wavelet is used for wavelet transform to decompose the preprocessed image into low-frequency approximate components and high-frequency detail components. Multi-scale features of defects are extracted through wavelet filters of different scales. Low-frequency information is obtained through low-pass filtering to capture the overall characteristics of the internal structure of the wood, and high-pass filtering is used to obtain high-frequency information to highlight the detailed features including the edges of the defects.

[0024] In this embodiment, the decoder part of the U-Net++ structure upsamples the feature map with the defective area through a deconvolution operation, and performs a jump connection with the feature map of the corresponding scale in the encoder to fuse the feature information of different levels again to generate a pixel-level mask map of the defect.

[0025] In this embodiment, the AIFI module fuses feature maps of different scales through a multi-head attention mechanism. The multi-head attention mechanism divides the feature map into multiple subspaces, each of which is processed by an independent attention head. This design enables the model to simultaneously focus on feature information at different positions and scales. By calculating the attention weight of each position, the model can automatically focus on key areas related to defects. After feature fusion, residual connections are introduced to add the original input features to the fused features. This structural design effectively solves the gradient vanishing problem in deep neural networks, ensuring that the model can learn richer feature representations while retaining important information in the original input.

[0026] In this embodiment, after obtaining the fused feature map, an adaptive local threshold method is used for image segmentation. This method dynamically calculates the threshold based on the local area characteristics around each pixel, rather than using a global unified threshold. For the internal density distribution image of wood, this method can better adapt to the grayscale variation characteristics of different regions, especially when the boundary between the defect area and normal tissue is blurred. After segmentation, morphological operations (such as opening and closing operations) are applied to eliminate noise and fill holes. The opening operation can remove small noise points, while the closing operation helps to connect and fill small holes in the defect area, thereby obtaining a more complete and accurate defect area outline.

[0027] In this embodiment, in order to further optimize the segmentation results, the structural similarity index between the defective area and the normal area is calculated. The SSIM value measures the similarity between the two areas in terms of brightness, contrast and structure. The value range is from -1 to 1. The higher the value, the more similar it is. When the SSIM value is lower than 0.8, it means that there may be errors in the segmentation result and the segmentation threshold needs to be adjusted. The system will dynamically adjust the adjustment coefficient in the adaptive local threshold method according to the current SSIM value and re-segment. This feedback mechanism enables the segmentation process to be automatically optimized according to the actual image characteristics, thereby improving the accuracy of defect area extraction.

[0028] Step 104: Based on the segmented defect area, classify the defect type, calculate the area, volume and density attenuation rate of the defect area, and output the defect severity assessment result through the improved MobileNetV3 model.

[0029] In this embodiment, based on the segmented defect area, the geometric features, texture features and density features of the defects are extracted as classification features; a support vector machine classifier is adopted, and the radial basis function is used as the kernel function. The parameters of the support vector machine classifier are optimized through cross-validation to classify the defects into different types, which include at least cavities, decay and insect bites.

[0030] In this embodiment, when calculating the area of ​​the defect region, the number of pixels contained in the defect region is automatically counted for the two-dimensional CT image. Since a fixed pixel size is determined during forestry CT scanning, for example, each pixel corresponds to an actual area of ​​0.1mm×0.1mm, a simple multiplication operation can be used to convert the number of pixels into the actual area, thereby accurately determining the coverage of the defect at a certain level. For the calculation of the three-dimensional volume, the defect areas of the wood at different scanning levels need to be accumulated in sequence. These superimposed area data reflect the extension of the defect in space. Multiplying by the scanning layer thickness, just like stacking layers of thin slices, the final result is the three-dimensional space occupied by the defect region within the wood. This value can intuitively reflect the scale of the defect. The key to calculating the density attenuation rate of a defective area lies in comparing the density difference between the defective area and the normal wood area. First, the average density of the defective area is calculated by averaging the CT values ​​of all pixels within the defective area. At the same time, a normal area without defects in the wood is selected and its average density is calculated as a reference standard. These two average density values ​​are substituted into the density attenuation rate calculation formula. By comparing the ratio of the difference between the two to the average density of the normal area, the degree of density attenuation is expressed as a percentage. This value can effectively reflect the degree of damage to the density of the wood material caused by the defect. For example, the higher the density attenuation rate, the more significant the decrease in wood density caused by the defect, and the more serious the damage to the structural integrity of the wood. When formulating defect severity assessment standards, multiple key parameters such as defect type, area, volume and density decay rate are comprehensively considered. For cavity-type defects, if the area is small, the volume is limited and the density decay rate is at a low level, it can be judged as a mild defect; when the cavity area is large and penetrates multiple layers of the wood, resulting in a significant increase in volume, and the density decay rate is high, it is classified as a severe defect. Decay-type defects, if they only appear on the local surface of the wood, with a small area and volume and a low degree of density decay, are classified as mild; and when the decayed area penetrates deep into the wood, the area and volume are greatly expanded, seriously affecting the density and structural performance of the wood, it is determined to be a severe insect-infested defect. Classification is also based on its scale and the degree of impact on the density of the wood. By setting specific area, volume and density decay rate threshold ranges for different types of defects, a set of comprehensive assessment standards is formed to achieve a scientific classification of defect severity. According to the established evaluation standards, defects are accurately divided into three levels: mild, moderate and severe. A detailed evaluation report is output, which not only clearly marks the level of the defect, but also lists in detail the type of defect, the calculated area, volume and density attenuation rate and other specific parameters. At the same time, the corresponding CT image and mask map of the defect area are attached to intuitively show the position and shape of the defect in the wood. In addition, for defects of different levels, corresponding treatment suggestions are provided. For example, mild defects can be properly repaired and then used, while severe defects need to be carefully evaluated for their usability, providing a comprehensive and reliable decision-making basis for wood quality assessment, subsequent processing and rational resource utilization. In this embodiment, based on the defect area obtained after segmentation, three types of key features are extracted for subsequent classification. In terms of geometric feature extraction, parameters such as the perimeter, area, major-minor axis ratio, and circularity of the defect area are calculated. The perimeter reflects the boundary length of the defect, the area directly reflects the size of the defect, the major-minor axis ratio can distinguish between long strips and approximately circular defects, and the circularity measures the degree of proximity of the defect shape to a standard circle. These parameters describe the geometric morphology of the defect from different dimensions. Texture feature extraction is achieved using the gray-level co-occurrence matrix. By calculating statistics such as energy, entropy, contrast, and correlation, the gray-level distribution pattern within the defect area is captured. Energy characterizes the uniformity of the image's gray-level distribution, entropy reflects the randomness of the gray-level distribution, contrast reflects the image's clarity, and correlation describes the degree of linear dependence between gray-levels. These texture features can effectively distinguish the surface characteristics of different types of defects. For example, the texture of a decayed area is usually more complex than that of a cavity. Density feature extraction focuses on the CT value distribution of the defect area and calculates indicators such as average density, density standard deviation, maximum density, and minimum density. Average density reflects the overall density level of the defect area, standard deviation reflects the degree of density dispersion, and maximum and minimum density reveal the extremes of density variation. Different types of defects exhibit unique density characteristics due to their different physical structures. For example, the density of a cavity area is extremely low, while the density of a decayed area is between that of normal wood and a cavity. To achieve accurate classification of defect types, support vector machine is used as the classifier, and radial basis function is selected as the kernel function. RBF kernel function has good nonlinear mapping ability and can handle complex classification boundaries. It is suitable for multi-feature and nonlinear classification problems such as wood defects. During the model training phase, the 5-fold cross-validation method was used to optimize the SVM parameters. The data set was randomly divided into 5 subsets, 4 subsets were used for training each time, and 1 subset was used for validation. This cycle was repeated 5 times to ensure that each subset was used for validation. The target parameters for optimization included the penalty factor C and the kernel function parameter γ. The penalty factor C controls the balance between classification error and model complexity. A larger C value makes the model fit the training data more strictly, but may lead to overfitting; a smaller C value allows more classification error and improves the generalization ability of the model. The kernel function parameter γ determines the distribution of the data after mapping it to high-dimensional space and has an important influence on the shape of the classification boundary. Through grid search or heuristic algorithms, the optimal parameter combination is found in the preset parameter space to achieve the highest cross-validation accuracy. The SVM model finally trained can accurately classify wood defects into different types such as cavities, decay, and insect bites based on the extracted geometric, texture, and density features. For each defect area to be classified, the model outputs the probability of it belonging to each category, and the category with the highest probability is selected as the final classification result.

[0031] See also Figure 2 , a schematic structural diagram of a nondestructive detection system for internal defects of wood based on forestry CT images provided by an embodiment of the present invention, the system comprising: An image acquisition module is used to acquire projection data of wood using forestry CT scanning equipment, reconstruct the projection data using the maximum likelihood expectation iteration method, and generate an image of the internal density distribution of the wood; The image preprocessing module is used to perform histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area and obtain the preprocessed image; The defect annotation module is used to input the preprocessed image into the improved MobileNetV3 model to generate a pixel-level mask of the defect, annotate the location and range of the defect, and obtain the defect area; The output module is used to classify the defect type based on the segmented defect area, calculate the area, volume and density attenuation rate of the defect area, and output the defect severity assessment results through the improved MobileNetV3 model Figure 3 This is a schematic diagram of the structure of a nondestructive detection device for internal defects in wood based on forestry CT images, provided by an embodiment of the present invention. The device 300 can vary significantly depending on configuration or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors), memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions for operating on the nondestructive detection device 300 based on forestry CT images. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, allowing the nondestructive detection device 300 based on forestry CT images to execute the instructions stored in the storage medium 330 to implement the method provided in the aforementioned embodiment.

[0032] The nondestructive detection device 300 for wood internal defects based on forestry CT images may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating devices 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The structure of the nondestructive detection equipment for internal defects of wood based on forestry CT images shown does not constitute a limitation on the computer equipment provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0033] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the various steps of the non-destructive detection method for internal defects of wood based on forestry CT images provided in the above-mentioned embodiments.

[0034] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0035] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0036] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A nondestructive detection method for internal defects of wood based on forestry CT images, characterized in that: The method comprises the following steps: The projection data of wood is obtained by using forestry CT scanning equipment, and the maximum likelihood expectation iteration method is used to reconstruct the projection data to generate the internal density distribution image of the wood. Perform histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area and obtain the preprocessed image; The preprocessed image is input into the improved MobileNetV3 model to generate a pixel-level mask of the defect, marking the location and range of the defect to obtain the defect area; Based on the segmented defect area, the defect type is classified, the area, volume and density attenuation rate of the defect area are calculated, and the defect severity assessment results are output through the improved MobileNetV3 model.

2. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 1, characterized in that: The method comprises: obtaining projection data of wood using forestry CT scanning equipment, reconstructing the projection data using the maximum likelihood expectation iteration method, and generating an image of the internal density distribution of the wood, including: The initial number of iterations is set to 100, and the ordered subset acceleration strategy is introduced to divide the projection data into 8-16 subsets; When the maximum likelihood expectation iteration method is used to reconstruct the projection data, only one subset is processed in each iteration, and each subset is iterated alternately; The convergence threshold is set to 0.

001. When the difference between the reconstructed images of two adjacent iterations is less than the convergence threshold, the iteration is stopped and the internal density distribution image of the wood is generated.

3. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 1, characterized in that: The method of performing histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area to obtain a preprocessed image includes: Obtain the number of pixels at each gray level in the wood internal density distribution image and calculate its probability of occurrence; Redistribute the grayscale according to the probability distribution, expand the narrow grayscale range in the original wood internal density distribution image to the grayscale range of 0-255, and enhance the contrast of the image; The Gamma value is set to 0.5-0.8, and Gamma correction is performed on the image after histogram equalization processing to compress the high grayscale area and expand the low grayscale area to obtain the preprocessed image.

4. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 1, characterized in that: The preprocessed image is input into the improved MobileNetV3 model to generate a pixel-level mask of the defect, and the position and range of the defect are marked to obtain the defect area, including: The preprocessed image is input into the improved MobileNetV3 model, and a wavelet convolution module is embedded in the backbone network to extract the multi-scale features of the defect through wavelet transform. The AIFI module is introduced to fuse feature maps of different scales through a multi-head attention mechanism. After feature fusion, a residual connection is added to add the original input and the fused features to obtain the fused feature map. The fused feature map is segmented using the adaptive local threshold method, and morphological operations are combined to eliminate noise and extract defect areas. Calculate the SSIM values ​​of the defect area and the normal area. If the SSIM value is lower than 0.8, adjust the adjustment coefficient of the adaptive local threshold method according to the SSIM value and dynamically adjust the segmentation threshold. The encoder-decoder framework of the U-Net++ structure is used to generate pixel-level mask maps of defects and mark the location and range of defects.

5. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 4, characterized in that: The wavelet convolution module is embedded in the backbone network to extract multi-scale features of defects through wavelet transform, including: Daubechies wavelet is used for wavelet transform to decompose the preprocessed image into low-frequency approximate components and high-frequency detail components. The multi-scale features of defects are extracted through wavelet filters of different scales. Low-pass filtering is used to obtain low-frequency information to capture the overall characteristics of the internal structure of the wood, and high-pass filtering is used to obtain high-frequency information to highlight the detailed features including the edges of the defects.

6. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 4, characterized in that: The encoder-decoder framework using the U-Net++ structure generates a pixel-level mask of the defect, annotating the location and extent of the defect, including: The decoder part of the U-Net++ structure upsamples the feature map of the defective area through deconvolution operations, and performs jump connections with the feature map of the corresponding scale in the encoder to fuse the feature information at different levels again to generate a pixel-level mask map of the defect.

7. The nondestructive detection method for internal defects of wood based on forestry CT images according to claim 1, characterized in that: The defect type classification based on the segmented defect area includes: Based on the segmented defect area, the geometric features, texture features and density features of the defect are extracted as classification features; A support vector machine classifier is adopted, and radial basis function is used as kernel function. The parameters of the support vector machine classifier are optimized through cross validation to classify defects into different types, which at least include voids, decay and insect damage.

8. A nondestructive detection system for wood internal defects based on forestry CT images, characterized in that: The system includes: An image acquisition module is used to acquire projection data of wood using forestry CT scanning equipment, reconstruct the projection data using the maximum likelihood expectation iteration method, and generate an image of the internal density distribution of the wood; The image preprocessing module is used to perform histogram equalization and gamma correction on the wood internal density distribution image to enhance the contrast of the defect area and obtain the preprocessed image; The defect annotation module is used to input the preprocessed image into the improved MobileNetV3 model to generate a pixel-level mask of the defect, annotate the location and range of the defect, and obtain the defect area; The output module is used to classify the defect type based on the segmented defect area, calculate the area, volume and density attenuation rate of the defect area, and output the defect severity assessment results through the improved MobileNetV3 model.

9. A non-destructive detection device for wood internal defects based on forestry CT images, characterized in that: The nondestructive detection device for internal defects of wood based on forestry CT images includes a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the nondestructive detection device for internal defects of wood based on forestry CT images performs each step of the nondestructive detection method for internal defects of wood based on forestry CT images as described in any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for non-destructive detection of internal defects of wood based on forestry CT images as described in any one of claims 1 to 7 are implemented.

Images