A method for grading particle board based on improved decision tree algorithm

Abstract

The application discloses a method for particle board grading based on an improved decision tree algorithm, which comprises the following steps: collecting particle board images; grading the particle board images manually; selecting image data in which only one kind of defect exists on the surface of the particle board, and taking the divided particle board grade as the severity label of the defect in the image; extracting the evaluation index of the defect in the image to form a judgment attribute; generating a decision tree for the evaluation index of each kind of defect; selecting image data in which multiple kinds of defects exist on the surface of the particle board, and taking the result output by all the decision trees for the evaluation index of the defect and the particle board grade divided for each image as a training set, and taking all the defect categories as an attribute set; generating a decision tree for the division of the defect categories; constructing a grading method; and applying the grading method to the particle board production detection link. Through the method, the particle board quality can be accurately graded, the detection precision and work efficiency are improved, and the calculation cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of particle board processing quality detection, and specifically relates to a particle board grading method based on an improved decision tree algorithm, aiming to realize accurate grading of the quality of particle boards through processing and analysis of image data of the particle boards. BACKGROUND

[0002] As a widely used wood-based panel, particle board has important applications in the fields of furniture manufacturing, building decoration, etc. In the production process of particle boards, the surface quality is directly related to the overall quality of the products and market competitiveness. However, at the present stage, particle board production enterprises still mainly rely on arranging special detection personnel to conduct manual visual sorting in terms of surface defect detection.

[0003] With the continuous acceleration of the production line speed, the disadvantages of manual visual sorting method are increasingly prominent. The surface defects of particle boards are usually small relative to the entire board surface, and it is difficult for the human eye to accurately and efficiently detect these surface defects on a high-speed particle board production line. At the same time, it is also difficult to distinguish the types of surface defects of particle boards in a short period of time, which leads to low recognition rate and high false detection and missed detection rate. In addition, due to the strong subjectivity of manual visual sorting method, and there is no uniform appearance quality grading standard for the surface defects of particle boards at present, the judgments made by different detection personnel on the same batch of particle boards with different surface defects may have great differences, which leads to low accuracy and uniformity of the judgment on the appearance surface defects of particle boards.

[0004] In recent years, machine vision technology has been more and more widely applied in the field of industrial detection. A number of research teams have carried out research on particle board surface defect machine vision detection systems and have made some new progress. The particle board surface defect detection system of machine vision usually collects images through an industrial camera, and in the corresponding software on the computer, a specific detection algorithm is used to extract the features of the surface defects of the particle board in the image, and the detection and recognition of the surface defects of the particle board are realized according to these feature information. Compared with the manual visual sorting method, the particle board surface defect detection system using machine vision technology has higher detection accuracy and higher work efficiency, and can realize digital statistical management.

[0005] Nevertheless, this solution is still in the application development stage and has some issues that require further improvement. For example, current detection algorithms for particleboard surface defects mainly focus on the target detection of a few types of defects. However, in actual particleboard production, surface defects vary greatly in size, style, quantity, and type, and there is a lack of a scheme to further classify and grade particleboard based on these defects. After successfully extracting the surface defect features of particleboard, a corresponding grading algorithm needs to be designed to achieve the grading of particleboard quality. This algorithm should be able to comprehensively consider multiple dimensions such as the type, quantity, area, and severity of defects to set different quality levels for particleboard. However, since the occurrence of particleboard surface defects is random, the subsequent processing methods for different defects are not entirely the same. Therefore, the grading of particleboard is not a simple mechanical process but requires comprehensive consideration of multiple dimensions. Currently, those skilled in the art are attempting to use decision tree algorithms to evaluate the quality level of particleboard based on multiple features of surface defects. However, particleboard has many types of defects, such as sand leakage, dust spots, glue spots, and indentations, and the evaluation indicators for each defect are different, such as area, quantity, range, and dispersion. If traditional decision tree algorithms are used directly, there will be an attribute set containing dozens of elements, and the generated decision tree will be very complex and redundant, with high computational costs, which is not conducive to practical production applications. Summary of the Invention

[0006] The technical problem this invention aims to solve is to provide a particleboard grading method based on an improved decision tree algorithm, addressing the shortcomings of existing technologies. This method aims to overcome the inefficiencies, poor accuracy, and strong subjectivity of traditional manual visual sorting methods, while also addressing the deficiencies of existing machine vision inspection systems in particleboard grading, such as the lack of a grading algorithm that comprehensively considers multiple dimensions and the complexity and redundancy of decision tree algorithms. Through this invention, precise grading of particleboard quality can be achieved, improving inspection accuracy and work efficiency, reducing computational costs, and providing particleboard manufacturers with an efficient, accurate, and unified grading solution, thereby enhancing product market competitiveness and enterprise economic benefits.

[0007] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0008] A particleboard grading method based on an improved decision tree algorithm includes the following steps:

[0009] Step 1: Collect particleboard image data and construct a dataset for generating a decision tree;

[0010] Step 2: Manually classify the collected particleboard image data into four levels;

[0011] Step 3, select the image data of the particle board surface with only one kind of defect, and take the particle board grade divided in step 2 as the severity label of the defect in the image; the evaluation index of the defect in the image is extracted by using the image processing method, the evaluation index includes area, quantity, range and dispersion degree, and constitutes the judgment attribute; the results of the judgment attribute in each image and the severity of the defect on the image are constructed as a training set, the evaluation index of the defect is constructed as an attribute set, and a decision tree for the evaluation index of the defect is generated for each kind of defect; in the generation of the decision tree, the evaluation index of the defect is divided according to the characteristics of different defects, and the attribute priority weighting is embodied in the decision tree generation algorithm;

[0012] Step 4, select the image data of the particle board surface with multiple defects, and bring each image into all the decision trees for the evaluation index of the defect generated in step 3, and save the output results of each kind of defect; the results of each kind of defect obtained from each image and the particle board grade divided in step 2 are constructed as a training set, and all kinds of defects are constructed as an attribute set; a decision tree for the classification of the defect types is generated;

[0013] Step 5, constructing a grading method: judging whether the surface of the particle board contains defects, if not, it is a first-class board; if there are defects, further judging the defect type of the surface of the particle board, and extracting the specific value of the evaluation index of each defect, if only containing a single defect, it is brought into the decision tree for the evaluation index of the defect, and the severity of the defect determined by the decision tree is taken as the grading grade of the particle board; if containing multiple defects, it is brought into the decision tree for the classification of the defect types, and in each judgment node, the evaluation index of the defect is brought into the decision tree for the evaluation index of the defect generated in step 3, and the output result is taken as the judgment result of the node, until the decision tree for the classification of the defect types is executed and the output result is taken as the grading grade of the particle board;

[0014] Step 6, applying the grading method obtained in step 5 to the particle board production detection link: collecting the particle board image and processing the image, then inputting the processed data into the grading method obtained in step 5, and grading the particle board according to the output result, thus the grading of the particle board is completed.

[0015] As a further improved technical solution of the application, the step 1 is specifically: using an industrial camera to collect particle board image data, and constructing a data set for generating a decision tree.

[0016] As a further improved technical solution of the application, the four grades in step 2 are specifically first-class board, second-class board, third-class board and fourth-class board.

[0017] As a further improved technical solution of the present application, the step 3 is specifically:

[0018] Step 3.1, selecting image data of the particle board in which only one kind of defect exists on the surface of the particle board in step 2;

[0019] Step 3.2, taking the grade of the particle board divided in step 2 for the image data as the severity of the defect in the corresponding image;

[0020] Step 3.3, for each kind of defect, using an image processing method to extract evaluation indexes of the defect in the image, the evaluation indexes including area, quantity, range and dispersion degree;

[0021] Step 3.4, constructing the results of the evaluation indexes of the defect extracted from each image and the severity of the defect on the corresponding image into a training set; each item in the training set includes evaluation index data of one kind of defect and a corresponding severity label;

[0022] Step 3.5, constructing the evaluation indexes of the defect into an attribute set;

[0023] Step 3.6, for each kind of defect, using the training set constructed in step 3.4 and the attribute set constructed in step 3.5, and using an improved ID3 decision tree learning algorithm to generate a decision tree for the evaluation indexes of the defect; the improved ID3 decision tree learning algorithm introduces weights Ka set according to different kinds of defects and different evaluation indexes, which is used to correct the influence degree of different kinds of defects on one evaluation index, so that the optimal partition attribute formula is optimized as a*=arg(Ka·max(Gain(D,a))) by multiplying the information gain by the weight of the corresponding feature; wherein, Gain(D,a) is the information gain, and Ka is the weight of one evaluation index corresponding to one kind of defect.

[0024] As a further improved technical solution of the present application, the step 4 is specifically:

[0025] Step 4.1, selecting particle board image data in which multiple defects exist on the surface from the particle board image data set constructed in step 2;

[0026] Step 4.2, taking each image in which multiple defects exist into the decision tree for the evaluation indexes of the defect generated in step 3 for each kind of defect respectively, to obtain the output results of each kind of defect, the results including the severity of each kind of defect;

[0027] Step 4.3, constructing the multiple defect results of each image after integration and the grade of the particle board corresponding to the image divided in step 2 into a new training set;

[0028] Step 4.4, all defect categories in each image are taken as the attribute set;

[0029] Step 4.5, using the traditional ID3 decision tree learning algorithm, the training set constructed in step 4.3 and the attribute set constructed in step 4.4, a decision tree for defect category division is generated.

[0030] As a further improved technical solution of the application, the defect categories include scratches, glue spots, indentations, large shavings, dust spots, and sand leakage.

[0031] The beneficial effects of the application are:

[0032] (1) Compared with the traditional manual visual sorting method, the application uses machine vision technology and an improved decision tree algorithm to more accurately detect and grade the surface defects of the particle board. Machine vision technology can efficiently collect and process image data, and the improved decision tree algorithm can consider multiple defect characteristics to achieve more accurate grading.

[0033] (2) The manual visual sorting method is not only inefficient, but also subjective, which can lead to inconsistent results. The application reduces the dependence on manual detection through an automated method, thereby reducing labor costs and errors caused by subjective factors.

[0034] (3) The application provides a clear grading standard and method, which divides the particle board into four grades, each with specific defect characteristics and severity definitions. This helps to unify the industry's evaluation criteria for particle board quality, improving market standardization and product comparability.

[0035] (4) There are many types of surface defects on particle boards, and each defect has different effects on quality. The application generates decision trees for single defects and multiple defects, respectively, to consider the characteristics and severity of different defects, achieving more comprehensive and reasonable grading.

[0036] (5) The traditional decision tree algorithm has high computational cost when dealing with particle board defects due to complex and redundant attribute sets. The application introduces attribute priority weighting and an improved information gain calculation method to optimize the generation process of the decision tree, reducing the computational complexity and making it more suitable for practical production applications. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a schematic diagram of various types of particle board surface defects.

[0038] Figure 2 is a schematic diagram of a decision tree for sand leakage defects.

[0039] Figure 3A decision tree diagram for defect category classification. DETAILED DESCRIPTION

[0040] The specific embodiments of the present application are further described below with reference to the accompanying drawings:

[0041] A particle board grading method based on an improved decision tree algorithm, comprising:

[0042] Step 1: Use an industrial camera to collect particle board image data and build a data set for generating a decision tree.

[0043] Step 2: Manually grade the collected particle board data. Based on the specific conditions of the board surface defects and actual production experience, the particle board is divided into four grades.

[0044] Step 3: Select particle board data with only one type of defect on the surface, and use the particle board grade divided in step 2 above as the severity label of the defect in the image. Use general image processing methods to extract evaluation indicators such as the area, number, range, and dispersion of the defect in the image to form the decision attributes. The results of the decision attributes in each image and the severity of the defect on the image are built into a training set, the evaluation indicators of the defect are built into an attribute set, and a decision tree for defect evaluation indicators (i.e., defect attributes) is generated for each type of defect. When generating the decision tree, the defect evaluation indicators are prioritized according to the characteristics of different defects, which is specifically reflected in the attribute priority weighting in the decision tree generation algorithm. The defect categories include scratches, glue spots, indentations, large particles, dust spots, and sand leakage. The defect images are shown in Figure 1

[0045] Step 4: Select data with multiple defects on the surface of the particle board, input each image into all the decision trees for defect evaluation indicators generated in step 3 above, and save the output results of each defect. The results of each defect obtained from each image and the particle board grade divided in step 2 above are built into a training set, and all defect categories are built into an attribute set. A decision tree for defect category classification is generated.

[0046] ​Step 5, constructing the grading method. Determine whether the surface of the particle board contains defects, if there is no defect, it is a first-class board; if there is a defect, further determine the type of surface defect of the particle board, and extract the specific value of each defect evaluation index, if it contains only a single defect, it is brought into the decision tree for the defect evaluation index, and the severity of the defect determined by the decision tree is taken as the grading level of the particle board; if it contains multiple defects, it is brought into the decision tree for defect type classification, and the evaluation index of the defect is brought into the decision tree for the evaluation index of the defect generated in step 3 at each decision node, and the output result is taken as the decision result of the node, until the decision tree for defect type classification is executed and the output result is taken as the grading level of the particle board.

[0047] Step 6, apply the specific grading method obtained in step 5 to the particle board production detection link. The camera collects the particle board image and after a series of necessary general image processing, the data is transmitted into the grading method, and the particle board is classified according to the output result. At this point, the particle board grading is completed.

[0048] The step 2 is specifically:

[0049] Step 2.1, the four levels are: first-class board, no defect or only a small amount of slight defect, which can be directly produced.

[0050] Step 2.2, second-class board, a certain number of slight defects or a small amount of moderate defects, which can be produced after manual turning.

[0051] Step 2.3, third-class board, multiple moderate defects or a small amount of serious defects, which can be produced after manual turning.

[0052] Step 2.4, fourth-class board, multiple serious defects, which affects use and is directly scrapped.

[0053] The step 3 is specifically:

[0054] Step 3.1, select the data of the particle board surface in step 2 which contains only one kind of defect. For example, select the image data containing only sand leakage defects, or data containing only dust spots, glue spots, indentation and other single defects.

[0055] Step 3.2, the grade of the particle board divided in step 2 for the image is taken as the severity of the defect in the image. For example, if a particle board containing sand leakage defects is manually classified as a second-class board, the severity of the sand leakage defect of the board is second-class.

[0056] Step 3.3, For each defect, use general image processing methods, including but not limited to region segmentation, edge detection, morphological operation, etc., to extract the evaluation indexes of the defect in the image, such as area, number, range, dispersion degree, etc. These evaluation indexes are key parameters for describing the characteristics of defects, and represent the severity of defects from different angles. For example, for the sand leakage defect, the image processing method can calculate the area of the sand leakage region, count the number of sand leakage points, measure the maximum extension range of the sand leakage region, and analyze the dispersion degree between the sand leakage points, etc. Quantitative data is provided for subsequent decision tree generation.

[0057] Step 3.4, The results of the extracted judgment attributes (i.e. defect evaluation indexes) in each image and the severity of the defect on the image are constructed as a training set. Each item in the training set includes the evaluation index data of the defect and the corresponding severity label. For example, for the training set of the sand leakage defect, D_lousha={(x_1:[“area”:100, “number”:5, “range”:120, “dispersion degree”:“dispersion”], y_1:“level three”), (x_2:[“area”:20, “number”:1, “range”:15, “dispersion degree”:“concentration”], y_2:“level two”), …(x_n:[…], y_n:…)}, where n is the total number of sand leakage data sets.

[0058] Step 3.5, The evaluation indexes of the defect (such as area, number, range, dispersion degree, etc.) are constructed as an attribute set. The attribute set is the input feature set in the decision tree algorithm, and in the generation of the decision tree, the subsequent algorithm will classify and decide the samples according to these attributes. For example, for the sand leakage defect, the attribute set is A_lousha={a_1:“area”, a_2:“number”, a_3:“range”, a_4:“dispersion degree”}. These attributes will be used as nodes of the decision tree, and will be split and selected in the algorithm to generate the optimal decision tree structure.

[0059] Step 3.6: For each type of defect, using the training set and attribute set constructed above, generate a decision tree for the defect evaluation index. This method improves upon the traditional ID3 decision tree learning algorithm to better adapt to particleboard classification and grading tasks. For particleboard surface defects, the same index but different types of defects have significantly different impacts on grading. For example, a particleboard with a sand leakage area of ​​20% of the board surface can be rated as a level 2 board, while a large particleboard of the same area is rated as a level 4 board. This shows that even with the same level of area as an index, different types of defects can lead to significant differences in rating. This method optimizes the optimal partitioning attribute a* selected from ID3. Traditional ID3 uses information gain as the criterion for selecting partitioning attributes, a* = arg(max(Gain(D,a))), where Gain(D,a) is the information gain. This method introduces a weight Ka set according to different defect types and various evaluation indexes to correct the degree of influence of different types of defects on a certain evaluation index. The specific weight value is set based on the degree of influence of a certain index on particleboard grading in actual production experience. Multiplying the information gain by the weight of the corresponding feature, the optimal splitting attribute formula is optimized to a*=arg(Ka·max(Gain(D,a))). For example, for sand leakage defects, the training set D_lousha and attribute set A_lousha are input into the improved decision tree learning algorithm TreeGenerate_lousha for sand leakage defects, generating a result such as... Figure 2 The decision tree shown is Tree_lousha for sand leakage defects (i.e., decision tree for sand leakage defect evaluation indicators), which is Tree_lousha = TreeGenerate_lousha(D_lousha, A_lousha).

[0060] Step 4 specifically includes:

[0061] Step 4.1: From the particleboard image dataset constructed in Step 2, select data with multiple surface defects. These data may simultaneously contain various defect types such as sand leakage, dust spots, and glue spots, ensuring that the selected data can cover multiple defect combinations that may occur in actual production, providing a foundation for subsequently generating a universally applicable multi-defect decision tree.

[0062] Step 4.2: For each selected image containing multiple defects, input them into the decision trees generated in Step 3 for each single defect evaluation metric. For example, for an image containing both sand leakage and glue spot defects, input it into the sand leakage defect decision tree (i.e., the decision tree for the sand leakage defect evaluation metric) and the glue spot defect decision tree (i.e., the decision tree for the glue spot defect evaluation metric), respectively. Through these single defect decision trees, obtain the output results for each defect, including information such as the severity determination of each defect.

[0063] Step 4.3: Construct a new training set using the multiple defect results of each integrated image and the particleboard grade corresponding to the image as defined in Step 2. For example, construct the training set D_multi={(x_1:[“Sand Leakage”: “Level 2”, “Dust Spot”: “Level 2”, “Glue Spot”: “Level 1”, “Large Particles”: “Level 1”], y_1: “Level 2”), (x_2:[“Sand Leakage”: “Level 2”, “Dust Spot”: “Level 2”, “Glue Spot”: “Level 2”, “Large Particles”: “Level 3”], y_1: “Level 4”), …(x_m:[…], y_m:…)}, where m is the total number of datasets containing multiple defects.

[0064] Step 4.4: Treat all defect types as an attribute set. For example, if there are four defects mentioned above: sand leakage, dust spots, glue spots, and large wood shavings, then the attribute set would be A_multi = {a_1: "Sand leakage", a_2: "Dust spots", a_3: "Glue spots", a_4: "Large wood shavings"}. This attribute set is crucial for generating the multi-defect decision tree. The decision tree will split and make decisions based on these attributes to determine the quality grade of the particleboard.

[0065] Step 4.5: Using the traditional ID3 decision tree learning algorithm TreeGenerate, and leveraging the training set D_multi and attribute set A_multi constructed above, generate the following... Figure 3 The decision tree shown is Tree_multi, which is defined as Tree_multi = TreeGenerate(D_multi, A_multi). During the generation process, the algorithm automatically learns and determines the importance and priority of different defect types in classifying quality grades based on the correlation between various defect feature information and particleboard grades in the training set.

[0066] Compared with the traditional manual visual sorting method, the present application utilizes machine vision technology and improved decision tree algorithm to more accurately detect and grade the surface defects of particle board. Machine vision technology can efficiently collect and process image data, while the improved decision tree algorithm can consider multiple defect features to achieve more accurate grading. Manual visual sorting method is not only inefficient, but also subjective, which can lead to inconsistent results. The present application reduces the dependence on manual detection through an automated method, thereby reducing labor costs and errors caused by subjective factors. The present application provides a clear grading standard and method, which divides the particle board into four grades, each grade has specific defect features and severity definitions. This helps to unify the industry's evaluation criteria for particle board quality, and improves the standardization of the market and the comparability of products. There are many types of surface defects on particle board, and each defect has different effects on quality. The present application generates a decision tree for single defect and multiple defects respectively, which can consider the characteristics and severity of different defects to achieve more comprehensive and reasonable grading. The traditional decision tree algorithm has high computational cost when dealing with particle board defects due to the complexity and redundancy of the attribute set. The present application introduces attribute priority weighting and improved information gain calculation method to optimize the generation process of decision tree, reduce the computational complexity, and make it more suitable for practical application in production.

[0067] The protection scope of the present application includes but is not limited to the above embodiments, and the protection scope of the present application is subject to the claims, any replacement, deformation, improvement of the present technology that can be easily thought by those skilled in the art falls within the protection scope of the present application.

Claims

1. A particleboard grading method based on an improved decision tree algorithm, characterized in that, Includes the following steps: Step 1: Collect particleboard image data and construct a dataset for generating a decision tree; Step 2: Manually classify the collected particleboard image data into four levels; Step 3: Select image data where only one type of defect exists on the particleboard surface, and use the particleboard grade classified in Step 2 as the severity label of the defect in the image; use image processing methods to extract the evaluation index of the defect in the image, which includes area, quantity, range, and dispersion, to constitute the judgment attribute; construct the training set with the judgment attribute results and the severity of the defect in each image, construct the evaluation index of the defect as an attribute set, and generate a decision tree for each defect evaluation index; when generating the decision tree, prioritize the defect evaluation index according to the characteristics of different defects, which is reflected in the decision tree generation algorithm as attribute priority weighting; Step 4: Select image data with multiple defects on the particleboard surface, input each image into the decision tree generated in Step 3 above for all defect evaluation indicators, and save the output results for each defect; construct a training set by the results of each defect obtained for each image and the particleboard grade classified in Step 2 above, construct an attribute set for all defect types; generate a decision tree for defect type classification. Step 5: Construct a grading method: Determine whether the particleboard surface contains defects. If there are no defects, it is a Grade 1 board. If there are defects, further determine the type of defect on the particleboard surface and extract the specific value of the evaluation index for each type of defect. If there is only a single defect, input it into the decision tree for the evaluation index of that defect, and use the severity of the defect determined by the decision tree as the grading level of the particleboard. If there are multiple defects, input it into the decision tree for the classification of defect types. At each decision node, input the evaluation index of that type of defect into the decision tree for the evaluation index of that defect generated in Step 3, and use its output as the decision result of that node, until the decision tree for the classification of defect types is completed, and use the output result as the grading level of the particleboard. Step 6: Apply the grading method obtained in Step 5 to the particleboard production and testing process: After acquiring particleboard images and processing them, input the processed data into the grading method obtained in Step 5, and grade the particleboard according to the output results. At this point, the particleboard grading is complete.

2. The particleboard grading method based on the improved decision tree algorithm according to claim 1, characterized in that, Step 1 specifically involves: using an industrial camera to collect image data of particleboard and constructing a dataset for generating a decision tree.

3. The particleboard grading method based on the improved decision tree algorithm according to claim 1, characterized in that, The four levels in step 2 are specifically level 1 board, level 2 board, level 3 board and level 4 board.

4. The particleboard grading method based on the improved decision tree algorithm according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Select image data from Step 2 that show only one type of defect on the particleboard surface; Step 3.2: The particleboard grade determined in Step 2 for the image data is used as the severity of the defects in the corresponding image. Step 3.3: For each type of defect, use image processing methods to extract the evaluation indicators of the defects in the image. The evaluation indicators include area, number, range and dispersion. Step 3.4: Construct a training set by extracting the defect evaluation index results from each image and the severity of the corresponding defect in the image; each item in the training set includes the evaluation index data of a defect and the corresponding severity label. Step 3.5: Construct the evaluation indicators of defects into an attribute set; Step 3.6: For each defect, using the training set constructed in Step 3.4 and the attribute set constructed in Step 3.5, and employing the improved ID3 decision tree learning algorithm, a decision tree for the defect evaluation index is generated. The improved ID3 decision tree learning algorithm introduces weights Ka set according to different defect types and various evaluation indexes to correct the degree of influence of different types of defects on a certain evaluation index. The optimal attribute partitioning formula is optimized by multiplying the information gain by the weight of the corresponding feature: a*=arg(Ka·max(Gain(D,a))); where Gain(D,a) is the information gain, and Ka is the weight corresponding to a certain evaluation index in a defect.

5. The particleboard grading method based on the improved decision tree algorithm according to claim 1, characterized in that, Step 4 specifically includes: Step 4.1: Select particleboard image data with multiple surface defects from the particleboard image dataset constructed in Step 2; Step 4.2: Input each selected image with multiple defects into the decision tree generated in Step 3 for each defect evaluation index, and obtain the output result for each defect, including the severity of each defect. Step 4.3: Construct a new training set using the multiple defect results of each integrated image and the particleboard grade corresponding to the image as defined in Step 2. Step 4.4: Collect all defect types in each image as an attribute set; Step 4.5: Using the traditional ID3 decision tree learning algorithm, and utilizing the training set constructed in Step 4.3 and the attribute set constructed in Step 4.4, generate a decision tree for the classification of defect types.

6. The particleboard grading method based on the improved decision tree algorithm according to claim 1, characterized in that, The types of defects mentioned include scratches, glue spots, indentations, large shavings, dust spots, and sand leakage.

Images