A method for color classification of veneer wood based on feature extraction

Abstract

This invention belongs to the field of wood processing technology and relates to a method for color classification of veneer wood based on feature extraction. First, an image of the veneer wood is acquired and preprocessed to obtain a standardized image. Texture regions are extracted from the preprocessed image, identifying fine vein-like and coarse band-like wood texture regions respectively. These two types of regions are merged to generate a texture masking layer, and the texture masking layer is used to separate the unaffected natural color analysis region from the preprocessed image. Next, the color matrix of each channel in the four color spaces (RGB, Lab, HSV, LCH) of the natural color analysis region is extracted, and a high-dimensional color feature vector is constructed. Principal component analysis is used to perform feature dimensionality reduction. Finally, the K-means clustering algorithm is applied to the dimensionality-reduced feature vector, and the number of clusters is adaptively determined based on the color difference of the cluster centers, achieving accurate color classification of the veneer wood. This invention can effectively suppress the influence of texture interference on color classification and improve the accuracy and stability of veneer color classification.

Description

Technical Field

[0001] This invention belongs to the field of wood processing identification technology, and relates to computer identification of wood veneer color classification, and more particularly to a method for color classification of decorative wood veneer based on feature extraction. Background Technology

[0002] Veneered wood veneers are widely used in the production of furniture, flooring, and other products, and their color directly determines the visual effect. They need to be categorized by color before use, as different color families of wood veneers are suitable for different products.

[0003] Currently, the classification of veneer wood veneer colors largely relies on manual inspection. This method is inefficient and susceptible to subjective interference, resulting in poor consistency and stability of classification results. In large-scale production, it also consumes a significant amount of time and labor costs, making it difficult to balance consistency and efficiency in classification. Therefore, achieving efficient, accurate, and stable classification of veneer wood veneer colors is crucial.

[0004] The natural texture of wood veneer surfaces varies significantly between samples, and this texture intertwines with color information, making traditional color classification methods susceptible to texture interference and reducing classification accuracy. Furthermore, overly complex feature extraction leads to a surge in computation, failing to meet the real-time and efficiency requirements of large-scale industrial production. For example, Chinese patent CN113610185A has the following shortcomings: firstly, the grading intervals of the H and S channels are artificially set, lacking adaptability to the color distribution of different wood species; secondly, the V (lightness) channel is not introduced, making it difficult to distinguish between dark and light-colored woods; and thirdly, the grading intervals of the H channel have coverage gaps, resulting in some mid-tone woods not being properly classified. In addition, none of the above technical solutions consider the interference of wood surface texture on color features, making accurate classification difficult.

[0005] Therefore, there is an urgent need for a color classification method for wood veneer that can effectively shield the influence of texture, optimize the feature extraction process, and combine high efficiency and stability in industrial scenarios. Summary of the Invention

[0006] To address the problems of low color classification accuracy and complex feature extraction efficiency caused by texture interference in existing technologies, this invention provides a color classification method for veneer wood that can effectively shield texture interference, achieve high-efficiency feature extraction, and provide adaptive classification.

[0007] Technical solution

[0008] A method for classifying the color of wood veneer based on feature extraction includes the following steps:

[0009] S1. Obtain the image of the wood veneer and perform preprocessing operations on the original image to obtain a standardized image;

[0010] S2. Extract texture regions from the pre-processed wood veneer image, extracting fine vein-like wood texture and coarse band-like wood texture regions respectively, and merging the two types of texture regions to obtain a texture masking layer.

[0011] S3. Apply the texture masking layer to the preprocessed veneer image to shield texture interference and generate the natural color analysis region for veneer color analysis.

[0012] S4. Using the natural color analysis area of ​​the veneer wood veneer as the analysis unit, calculate the first moment, second moment and third moment of each color channel in the four color spaces of RGB, Lab, HSV and LCH respectively to form a high-dimensional color feature matrix G. Perform principal component analysis on G to obtain the dimension-reduced color feature matrix D.

[0013] S5. Perform K-means clustering on the dimensionality-reduced color feature matrix D, and adaptively determine the number of clusters K by the color difference between the cluster centers to obtain the color classification result of the veneer wood.

[0014] Preferably, the preprocessing step S1 specifically includes:

[0015] S11. Position, rotate, and cut the veneer to obtain a standardized image of the veneer;

[0016] S12. Perform bilateral filtering on the three RGB color channels of the standardized image respectively;

[0017] S13. Convert the processed image to a color space that includes the luminance channel, and perform histogram equalization on the luminance channel.

[0018] Preferably, the extraction process of the fine vein-like wood grain region in step S2 is as follows:

[0019] S21. Convert the preprocessed wood veneer image into a grayscale image of size M×N. gray (x,y), for I gray Perform two-dimensional discrete Fourier transform and spectral centering on (x,y);

[0020] S22. Calculate the filter radius according to the formula R = max(M,N) / 8, construct a circular low-pass filter H(u,v) with the center of the spectrum as the center to complete the frequency domain filtering, and obtain the low-frequency energy feature map E(u,v) by inverse Fourier transform;

[0021] S23. Perform binary segmentation on E(u,v) using an adaptive threshold algorithm, and then use morphological closure operation of 3×3 kernel to fill the small holes, thereby obtaining the fine vein-like wood texture region.

[0022] Preferably, the process of obtaining the texture masking layer in step S2 further includes:

[0023] For the coarse striped wood texture region in the preprocessed veneer image, the relatively dark coarse striped wood texture region is extracted using global OTU thresholding. Then, through connected component analysis, connected regions with an area greater than 200 pixels are retained. The boundaries of the retained connected regions are expanded using a 3×3 kernel dilation operation to generate the coarse striped wood texture region. The fine vein-like wood texture region is merged with the coarse striped wood texture region to obtain a texture masking layer used to shield texture interference.

[0024] Preferably, step S3 specifically includes:

[0025] Using the texture masking layer as a reference, a pixel-level masking operation is performed on the preprocessed veneer image, setting the pixel values ​​corresponding to the fine vein-like wood texture area and the coarse band-like wood texture area to zero. The remaining image area that is not set to zero is the natural color analysis area used for veneer color analysis.

[0026] Preferably, step S4 specifically includes:

[0027] Color features of the natural color analysis area of ​​the wood veneer were extracted. The R, G, and B channels of the RGB color space, the L, a, and b channels of the Lab color space, the H, S, and V channels of the HSV color space, and the L, C, and H channels of the LCH color space were extracted, for a total of 12 color channels. The first, second, and third moments of each color channel were calculated, resulting in 36 statistical measures. These 36 statistical measures were then concatenated in sequence to form a high-dimensional color feature matrix G.

[0028] The expression for the high-dimensional color feature matrix G is:

[0029] ;

[0030] The color feature vector of the natural color analysis region for each veneer is:

[0031] .

[0032] Preferably, step S4 further includes:

[0033] Principal component analysis is performed on the high-dimensional color feature matrix G to obtain the dimensionality-reduced color feature matrix D;

[0034] The color feature matrix D is expressed as follows:

[0035] .

[0036] Preferably, in step S5, the K-Means algorithm is used to perform unsupervised clustering based on the color feature matrix D; wherein the number of clusters K is adaptively determined by the following method:

[0037] Within the range of candidate cluster numbers, clustering is attempted one by one, and the color distance between each pair of cluster centers is calculated for each K value. If the color difference ∆E between all cluster centers under the current K value is greater than 5.0, the clustering of that K value is considered valid; otherwise, it is considered invalid. Starting from the smallest candidate value, K is gradually increased, and the largest valid K is selected as the number of clusters for the first stage of clustering. If all candidate K values ​​meet the valid condition, the largest candidate value is taken as the number of clusters. If none of the candidate K values ​​meet the condition, the largest candidate value is used for clustering by default.

[0038] Beneficial effects

[0039] This invention effectively avoids the interference of texture on color classification in traditional methods by masking the texture of veneer images and using only the color information of the natural color analysis region for classification, thereby improving the accuracy and stability of veneer color classification. This invention uses principal component analysis to reduce the dimensionality of high-dimensional color features, obtaining low-dimensional color feature vectors, effectively improving the efficiency of subsequent clustering operations. Simultaneously, it adaptively determines the number of clusters K based on the color difference distance of the cluster centers, ensuring the clustering effect and making it suitable for large-scale industrial veneer color sorting scenarios. This invention avoids the limitation of traditional K-means algorithms requiring manual preset of the K value by adaptively determining the number of clusters K, improving the objectivity of the classification results and adaptability to different batches of veneer. Attached Figure Description

[0040] Figure 1 A flowchart of the feature extraction-based color classification method for wood veneer of this invention;

[0041] Figure 2 The preprocessing flow and effect diagram of this invention are shown, wherein the left side represents the processing steps and the right side represents the corresponding processed image;

[0042] Figure 3 Flowchart of texture masking in this invention;

[0043] Figure 4 The input image used for texture region extraction in this invention;

[0044] Figure 5 An image illustrating the effectiveness of this invention in identifying fine vein-like wood grain regions;

[0045] Figure 6 An illustration illustrating the effectiveness of this invention in recognizing coarse, striped wood grain patterns.

[0046] Figure 7The image shows the effect of applying the texture masking layer of the present invention, wherein the left image is the texture masking layer and the right image is the original color analysis area. Detailed Implementation

[0047] The present invention will be described in detail below with reference to embodiments to enable those skilled in the art to better understand the present invention, but the present invention is not limited to the following embodiments. Unless otherwise specified, the raw materials and equipment involved in the embodiments are all conventional commercially available products in the art.

[0048] Example 1

[0049] A method for classifying the color of wood veneer based on feature extraction, such as Figure 1 As shown, it includes the following steps:

[0050] S1. Obtain the image of the wood veneer and perform preprocessing operations on the original image to obtain a standardized image;

[0051] S2. Extract texture regions from the pre-processed wood veneer image, extracting fine vein-like wood texture and coarse band-like wood texture regions respectively, and merging the two types of texture regions to obtain a texture masking layer.

[0052] S3. Apply the texture masking layer to the preprocessed veneer image to shield texture interference and generate the natural color analysis region for veneer color analysis.

[0053] S4. Using the natural color analysis area of ​​the veneer wood veneer as the analysis unit, calculate the first moment, second moment and third moment of each color channel in the four color spaces of RGB, Lab, HSV and LCH respectively to form a high-dimensional color feature matrix G. Perform principal component analysis on G to obtain the dimension-reduced color feature matrix D.

[0054] S5. Perform K-means clustering on the dimensionality-reduced color feature matrix D, and adaptively determine the number of clusters K by the color difference between the cluster centers to obtain the color classification result of the veneer wood.

[0055] like Figure 2 As shown, step S1 includes the following steps: positioning, rotating and cutting the veneer to obtain a standardized image of the veneer, that is, obtaining a closed contour through image binarization, determining contour points based on pixel jump boundaries, calculating the area of ​​the closed contour using Green's formula, filtering the outer contour with the largest area and extracting the convex hull, and determining the minimum bounding rectangle of the veneer through the rotation caliper method.

[0056] Using the original angle θ of the smallest bounding rectangle as a reference, compare the width w with the height h. If the width is less than the height, then adjust the original angle by an additional 90° to obtain the rotation angle θ. corrA rotation matrix is ​​generated using the center of the rectangle as the rotation center and the corrected angle as the rotation angle. An affine transformation is performed on the coordinate points of the original image and the outline to map them to a new pixel coordinate system. Then, the veneer area is cropped to obtain a veneer image with a standardized size of 4186×1070.

[0057] The closed contour is obtained based on the pixel transition boundary between binarized image 0 and 255. The binarized value I of the boundary pixel (x,y) must satisfy:

[0058] ,

[0059] Then, Green's formula is used to calculate the area of ​​the closed contour, the outer contour with the largest area is selected and the convex hull is extracted, and the minimum bounding rectangle of the veneer is determined by the rotating caliper method.

[0060] The original angle of the smallest bounding rectangle Using the minimum bounding rectangle as a reference, compare its width w with its height h. If the width is less than the height, adjust the original angle by an additional 90° to obtain the final rotation angle. satisfy:

[0061] ;

[0062] A rotation matrix is ​​then generated using the center of the rectangle as the rotation center and the corrected angle as the rotation angle. An affine transformation is performed on the coordinate points of the original image and the outline, mapping them to a new pixel coordinate system. The veneer area is then cropped, resulting in a veneer image with a standardized size of 4186×1070. Those skilled in the art can adjust the standardized size accordingly based on the resolution of the actual acquired image.

[0063] The above steps can accurately locate the veneer in the acquired image, complete rigid rotation correction, accurately remove the background, and standardize the size, ultimately obtaining a veneer image without background interference, with uniform size and unchanged original color information.

[0064] Bilateral filtering is applied to each of the three RGB color channels of the standardized wood veneer image. Specifically, a 19×19 pixel window is used to traverse each pixel of the image, and areas where the edge does not meet the window size are filled with mirrored pixels. A pixel value similarity parameter σ is set. color =75, Spatial proximity parameter σ space =75, calculate the spatial weight and pixel similarity weight separately, and the product of the two is the pixel weight value of the bilateral filter. Divide the weighted sum of all pixels in the window by the total weight to obtain the value of the center pixel after filtering, thus preserving edge details while reducing noise.

[0065] The above steps can effectively filter out image noise while preserving the original texture edges and pixel features of the wood veneer.

[0066] The standardized wood veneer image is converted to a color space that includes a luminance channel (such as Lab color space), and histogram equalization is performed on the luminance channel. Specifically, the L channel image is divided into 8×8=64 local sub-regions, each with S pixels. A local luminance histogram is calculated independently for each sub-region. A contrast upper limit threshold c=2.0 is set, and the result is calculated according to formula N. max = (S×c) / 256 to calculate the threshold, and truncate the histogram when the number of pixels exceeds N. max The peak portion is selected, and excess pixels are evenly distributed to other brightness ranges. Simultaneously, the effective brightness range of each sub-region is adaptively stretched, linearly mapping the brightness values ​​originally concentrated in dark or bright areas to the full dynamic range of 0-255. Finally, bilinear interpolation is used to smoothly stitch together the equalization results of the 64 sub-regions, eliminating block boundary effects and obtaining an L-channel with uniform brightness distribution.

[0067] The above steps can effectively improve the problem of uneven brightness in local areas of wood veneer images.

[0068] like Figure 3 As shown, step S2 includes the following steps:

[0069] like Figure 4 The image shown is an example of a pre-processed wood veneer image. For the fine vein-like wood texture regions in the pre-processed wood veneer image, two-dimensional discrete Fourier transform and frequency domain low-pass filtering are used to extract the low-frequency energy features of the wood veneer.

[0070] Convert the image to grayscale. gray The image is (x, y) with dimensions M × N. A two-dimensional discrete Fourier transform is performed to obtain the spectrum F(u, v). The origin of the spectrum is then moved to the center of the image to obtain Fu. shift (u,v). Construct a circular low-pass filter mask H(u,v) centered at the image center, with a filter radius R = max(M,N) / 8. Multiply the centered spectrum by the mask, retaining the low-frequency component F. low (u,v). An inverse Fourier transform is performed on the spectrum retaining the low-frequency components to obtain the low-frequency reconstructed energy map E(u,v), which is then linearly normalized to the grayscale range of 0–255. The normalized energy map is then binarized using a Gaussian weighted adaptive thresholding algorithm, with a window size of 31×31 pixels and a constant C=3, resulting in a binary image. Finally, a morphological closing operation is performed using a 3×3 rectangular structure to obtain the final fine-veined wood-textured region, as shown in the image. Figure 5 As shown.

[0071] Specifically, the preprocessed wood veneer image is converted into a grayscale image, and denoted as... The size is ;

[0072] Performing a two-dimensional discrete Fourier transform on the grayscale image yields the spectrum:

[0073] ;

[0074] By shifting the origin of the spectrum to the center of the image, we obtain the centered spectrum. The coordinates of the image center are... .

[0075] Construct a circular low-pass filter mask with the image center as the center:

[0076] ,

[0077] Among them, filter radius Multiply the centered spectrum by the mask to preserve the low-frequency components:

[0078] ,

[0079] Perform an inverse Fourier transform on the spectrum retaining the low-frequency components. First, perform a spectrum shift and restoration, then perform the inverse transform to obtain the reconstructed low-frequency energy map:

[0080] ;

[0081] The low-frequency energy map is linearly normalized to the grayscale range of 0–255 to obtain the normalized energy map.

[0082] ;

[0083] A Gaussian weighted adaptive thresholding algorithm is used to perform binarization segmentation on the normalized energy map, and the threshold is calculated within a local window. The window size is set to... Pixels, constants This yields a binary image:

[0084] ,in, This is a locally adaptive threshold;

[0085] Finally, to fill in the small holes in the binary image and connect the adjacent regions, the following method is used: Morphological closing operations are performed on the rectangular structure to obtain the final fine vein-like wood grain region.

[0086] For the coarse, banded wood texture areas in the preprocessed veneer image, a global OTU thresholding method is used to extract relatively dark, coarse, banded wood texture areas. This is illustrated using the grayscale image of the veneer.gray Using the input as input, the globally optimal threshold T is calculated using the Otsu's method. OTSU Introducing an offset C2=5, we obtain the correction threshold T. OTSU1 = T OTSU + C2, based on this, perform binarization. Perform connected component detection on the obtained initial region, and calculate the area S of each connected component. i Set area threshold S min =200, keep only S i > S min In the area, a 3×3 structuring element is used to perform an expansion operation on the preserved area to obtain a coarse, banded wood grain area, the effect of which is as follows. Figure 6 As shown.

[0087] That is, OTSU thresholding is performed using the grayscale image of the veneer wood. Using the input as input, the global optimal threshold is calculated using the Otsu's method. Introducing offset The corrected threshold is as follows:

[0088] ;

[0089] based on Binarization:

[0090] ;

[0091] The area with a value of 255 is the initial area of ​​the coarse, banded wood grain.

[0092] By merging the fine-veined wood grain areas with the coarse-banded wood grain areas, a texture masking layer is obtained to shield against texture interference, the effect of which is as follows: Figure 7 As shown.

[0093] Step S3 specifically involves: using the texture masking layer as a reference, performing a pixel-level masking operation on the preprocessed veneer image, setting the pixel values ​​corresponding to the fine vein-like wood texture area and the coarse band-like wood texture area to zero, and the remaining image area that is not set to zero is the natural color analysis area used for veneer color analysis.

[0094] Step S4 specifically includes: extracting color features from the natural color analysis area of ​​the veneer wood, extracting three channels (R, G, B) in the RGB color space, three channels (L, a, b) in the Lab color space, three channels (H, S, V) in the HSV color space, and three channels (L, C, H) in the LCH color space, for a total of 12 color channels; calculating the first moment (mean), second moment (standard deviation), and third moment (skewness) for each color channel, obtaining a total of 36 statistical measures, and concatenating these 36 statistical measures in sequence to form a high-dimensional color feature matrix G. Then, performing principal component analysis on the high-dimensional color feature matrix G reduces the feature dimension to 5 dimensions, obtaining the dimensionality-reduced color feature matrix D.

[0095] The expression for the high-dimensional color feature matrix G is: ;

[0096] The color feature vector of the natural color analysis region for each veneer is:

[0097] ;

[0098] Principal component analysis is performed on the high-dimensional color feature matrix G to obtain the dimensionality-reduced color feature matrix D;

[0099] The color feature matrix D is expressed as follows: .

[0100] Step S5 specifically includes: using the K-Means algorithm to perform unsupervised clustering based on the color feature matrix D, where the value of the cluster center K is adaptively determined based on the color difference distance between the cluster centers. Specifically, within the candidate cluster number range (e.g., 2 to 10), clustering is attempted one by one, and the color distance ∆E between all pairs of cluster centers obtained under each K value is calculated. The clustering is deemed valid if and only if the color difference ∆E between all cluster centers under the current K value is greater than 5.0; starting from the smallest candidate value, K is gradually increased, and the largest valid K is selected as the number of clusters; if all candidate Ks satisfy the valid condition, the largest candidate value is taken; if none of the candidate Ks satisfy the condition, the largest candidate value is used for clustering by default.

[0101] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for classifying the color of decorative wood veneer based on feature extraction, characterized in that, Includes the following steps: S1. Obtain the image of the wood veneer and perform preprocessing operations on the original image to obtain a standardized image; S2. Extract texture regions from the pre-processed wood veneer image, extracting fine vein-like wood texture and coarse band-like wood texture regions respectively, and merging the two types of texture regions to obtain a texture masking layer. S3. Apply the texture masking layer to the preprocessed veneer image to shield texture interference and generate the natural color analysis region for veneer color analysis. S4. Using the natural color analysis area of ​​the veneer wood veneer as the analysis unit, calculate the first moment, second moment and third moment of each color channel in the four color spaces of RGB, Lab, HSV and LCH respectively to form a high-dimensional color feature matrix G. Perform principal component analysis on G to obtain the dimension-reduced color feature matrix D. S5. Perform K-means clustering on the dimensionality-reduced color feature matrix D, and adaptively determine the number of clusters K by the color difference between the cluster centers to obtain the color classification result of the veneer wood.

2. The method for classifying the color of veneer wood based on feature extraction according to claim 1, characterized in that, The preprocessing described in step S1 specifically includes: S11. Position, rotate, and cut the veneer wood to obtain a standardized image; S12. Perform bilateral filtering on the three RGB color channels of the standardized image respectively; S13. Convert the processed image to a color space that includes the luminance channel, and perform histogram equalization on the luminance channel.

3. The method for classifying the color of veneer wood based on feature extraction according to claim 1, characterized in that, The extraction process of the fine vein-like wood grain region in step S2 is as follows: S21. Convert the preprocessed wood veneer image into a grayscale image of size M×N. gray (x,y), for I gray Perform two-dimensional discrete Fourier transform and spectral centering on (x,y); S22. Calculate the filter radius according to the formula R = max(M,N) / 8, construct a circular low-pass filter H(u,v) with the center of the spectrum as the center to complete the frequency domain filtering, and obtain the low-frequency energy feature map E(u,v) by inverse Fourier transform; S23. Apply an adaptive thresholding algorithm to perform binary segmentation on E(u,v), and then use a 3×3 kernel morphological closure operation to fill the small holes, thereby obtaining the fine vein-like wood texture region.

4. The method for classifying the color of veneer wood based on feature extraction according to claim 1, characterized in that, The process of obtaining the texture masking layer in step S2 also includes: for the coarse strip wood texture region in the preprocessed veneer image, the relatively dark coarse strip wood texture region is extracted using a global OTU threshold, and connected regions with an area greater than 200 pixels are retained through connected component analysis. The boundaries of the retained connected regions are expanded using a 3×3 kernel dilation operation to generate the coarse strip wood texture region; the fine vein wood texture region is merged with the coarse strip wood texture region to obtain the texture masking layer used to shield texture interference.

5. The method for classifying the color of veneer wood based on feature extraction according to claim 1, characterized in that, Step S3 specifically includes: using the texture masking layer as a reference, performing a pixel-level masking operation on the preprocessed veneer image, setting the pixel values ​​corresponding to the fine vein-like wood texture area and the coarse band-like wood texture area to zero, and the remaining image area that is not set to zero is the natural color analysis area used for veneer color analysis.

6. The method for classifying the color of decorative wood veneer based on feature extraction according to claim 5, characterized in that, Step S4 specifically includes: extracting the color features of the natural color analysis area of ​​the veneer wood, extracting three channels (R, G, B) in the RGB color space, three channels (L, a, b) in the Lab color space, three channels (H, S, V) in the HSV color space, and three channels (L, C, H) in the LCH color space, for a total of 12 color channels; calculating the first moment, second moment, and third moment for each color channel, obtaining a total of 36 statistical quantities, and concatenating the 36 statistical quantities in sequence to form a high-dimensional color feature matrix G.

7. The method for classifying the color of veneer wood based on feature extraction according to claim 1 or 6, characterized in that, Step S4 further includes: performing principal component analysis on the high-dimensional color feature matrix G to reduce the feature dimension to 5 dimensions, thereby obtaining the dimensionality-reduced color feature matrix D.

8. The method for classifying the color of veneer wood based on feature extraction according to claim 1, characterized in that: In step S5, the K-Means algorithm is used to perform unsupervised clustering based on the color feature matrix D.

9. The method for classifying the color of veneer wood based on feature extraction according to claim 8, characterized in that, The adaptive determination of the number of clusters K by the color difference between cluster centers specifically includes: within the candidate number of clusters range, attempting clustering one by one and calculating the color distance between each pair of all cluster centers obtained under each K value; if and only if the color difference ∆E between all cluster centers under the current K value is > 5.0, the clustering of the K value is determined to be valid, otherwise it is considered invalid; starting from the smallest candidate value, K is gradually increased, and the largest valid K is selected as the number of clusters in the first stage of clustering.

10. The method for classifying the color of veneer wood based on feature extraction according to claim 9, characterized in that: The adaptive determination of the number of clusters K further includes: if all candidate Ks meet the valid conditions, then the largest candidate value is taken as the number of clusters; if none of the candidate Ks meet the conditions, then the largest candidate value is used for clustering by default.

Images