A particleboard quality detection method and system based on image recognition

Abstract

The invention discloses a particleboard quality detection method and system based on image recognition, and specifically relates to the technical field of image recognition. By selecting particleboard samples from different batches and manually marking edge defects, combining local grayscale change analysis and texture contrast difference calculation, key features such as edge smoothness change rate and texture contrast difference are extracted, and the key features are converted into comprehensive feature vectors and input into a machine learning model for training and prediction. The method can effectively evaluate the quality recognition accuracy of the detection system in the edge area, and provide feedback based on the accuracy classification result. When incompletely accurate recognition occurs, by comparing the deviation of historical batches, abnormal particleboards are automatically marked and the operator is reminded to review, thereby improving production quality and safety, avoiding the flow of defective products into the market, and reducing potential product accident risks.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a particleboard quality detection method and system based on image recognition. Background Art

[0002] Image recognition-based particleboard quality inspection is a method that uses modern computer vision technology to automatically monitor and evaluate quality issues during particleboard production and use. By capturing particleboard surface images with a high-resolution camera and combining them with artificial intelligence algorithms (such as deep learning or traditional image processing methods), this technology can quickly identify whether there are defects on the particleboard surface, such as cracks, bubbles, discoloration, dents, or other surface problems. This method is more efficient and accurate than traditional manual inspection and can achieve real-time monitoring in mass production.

[0003] In addition, the detection system based on image recognition can also analyze the size, texture uniformity and smoothness of particleboard through algorithms to evaluate the overall quality level. By combining with industrial automation equipment, this technology can not only reduce the intensity of manual labor, but also improve the objectivity and consistency of detection results, help manufacturers optimize process flow and reduce defective rate, and provide important support for intelligent manufacturing in the particleboard industry.

[0004] The prior art has the following deficiencies:

[0005] In the edge areas of particleboard, there may be more surface quality defects, but these areas are usually difficult to detect accurately due to the camera's viewing angle or optical distortion. If the detection system cannot accurately identify quality problems at the edge, the overall quality assessment may be inaccurate. At the same time, the edge areas are usually where stress is concentrated in the particleboard, and defects such as cracks, peeling or poor pressing may occur. If quality problems in these areas are missed, defective products may be released during production. These defective products may break or delaminate during transportation, cutting or actual use, which directly affects the mechanical properties and safety of the particleboard, especially in furniture manufacturing or construction, and may lead to major product accidents. Summary of the invention

[0006] The purpose of the present invention is to provide a particleboard quality detection method and system based on image recognition to solve the shortcomings of the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a particleboard quality detection method based on image recognition, comprising the following steps:

[0008] S1: A series of particleboards were selected from different production batches, including multiple samples with edge defects, and the real defects in the edge area of ​​each sample were manually annotated, including cracks, bubbles, roughness, and uneven edges;

[0009] S2: Perform local analysis on the edge area, select multiple segmented windows to calculate the grayscale change, calculate the grayscale standard deviation of each window, mark the area where the smoothness changes suddenly, and calculate the ratio of the smoothness change by comparing the smoothness unchanged and sudden change areas to obtain the edge smoothness change rate;

[0010] S3: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing different texture blocks in the edge area of ​​the particle board, and the texture contrast difference of the edge area of ​​the particle board is calculated according to the change of the brightness difference of the texture of each area;

[0011] S4: converting the acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board into a comprehensive feature vector as the input of the machine learning model, and determining the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board according to the output result of the model;

[0012] S5: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board, the quality identification result of the detection system is divided into an accurate identification result, an incompletely accurate identification result and an inaccurate identification result;

[0013] S6: For inaccurate identification results, analyze the degree of abnormality in the quality accuracy of the edge area of ​​the particleboard identified by the detection system within a fixed time period. If the degree of abnormality is high, calculate the deviation between the identification accuracy of the current batch and the historical batches. If the deviation of the quality inspection result is large, automatically mark the corresponding particleboard and remind the operator to recheck.

[0014] Preferably, in S2, Gaussian blur is performed on the image, and Gaussian blur uses a Gaussian kernel to convolve the image, and the formula is: ; Where: G(x,y) is the value of the Gaussian kernel, x and y are the distances between the current pixel and the center of the kernel, σ is the standard deviation, which controls the degree of blur. The image is convolved with the Gaussian kernel to obtain the blurred image ;

[0015] The Sobel operator is used for edge detection, and its gradient calculation formula is: ; ;

[0016] in, and are the gradients along the x-axis and y-axis respectively, I(x,y) is the grayscale value of the image at position (x,y), and the gradient magnitude G and gradient direction θ are calculated: , ;

[0017] At each pixel location, calculate the grayscale difference between the blurred image and the original image: ; Where I(x,y) is the gray value of the original image, is the gray value of the blurred image, ΔI(x,y) is the gray difference of the pixel;

[0018] The pixels whose grayscale difference ΔI(x,y) is greater than the set threshold are regarded as the area where the edge smoothness changes suddenly. The ratio of the smoothness mutation area in the edge area is calculated, and the total number of pixels in the edge area is set to , the number of pixels in the smoothness mutation area is , then the smoothness change rate The calculation expression is: ;in, is the number of pixels whose grayscale difference is greater than the threshold, is the total number of pixels in the edge area.

[0019] Preferably, in S3, for each texture block, the average brightness value of all pixels thereof is calculated, and the formula is: ;in, is the average brightness of the texture block, N is the total number of pixels in the texture block, is the gray value of the i-th pixel, and the standard deviation of the brightness of each texture block is calculated to indicate the variation of the texture. The formula is: ;in, is the brightness standard deviation of the texture block, indicating the contrast of the texture block;

[0020] For each pair of adjacent texture blocks, the brightness difference is calculated as the contrast difference, and the average brightness of the two texture blocks is set to and , then the brightness contrast difference The calculation formula is: ; Calculate the average of the contrast differences between all texture patches: ;in, is the average value of the texture contrast difference in the edge area, M is the logarithm of the texture blocks for calculating the contrast difference, and the texture contrast difference in the edge area is obtained by combining the contrast differences of all texture blocks. The expression is: ;in, is the total number of texture blocks in the edge region, is the contrast difference of the i-th texture block, and TC is the texture contrast difference.

[0021] Preferably, in S4, the obtained edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board are converted into a comprehensive feature vector as an input item of the machine learning model, and the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board is determined according to the output result of the model, specifically:

[0022] The edge smoothness change rate and texture contrast difference are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the edge area quality of the particleboard predicted by the detection system as the prediction target, and takes minimizing the sum of prediction errors of the accuracy value labels of the edge area quality of the particleboard identified by all detection systems as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The accuracy value of the edge area quality of the particleboard identified by the detection system is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0023] Preferably, in S5, according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard, the quality recognition result of the detection system is divided into an accurate recognition result, an incompletely accurate recognition result and an inaccurate recognition result;

[0024] The acquired accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the first standard threshold and the second standard threshold respectively;

[0025] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than the second standard threshold, it means that the detection system has high accuracy in identifying the quality of the edge area of ​​the particle board, and a high-accuracy recognition signal is generated at this time, and the recognition result is classified as an accurate recognition result;

[0026] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the detection system has a general accuracy in identifying the quality of the edge area of ​​the particle board, and a medium accuracy recognition signal is generated at this time, and the recognition result is classified as an incompletely accurate recognition result;

[0027] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is less than the first standard threshold, it means that the detection system has low accuracy in identifying the quality of the edge area of ​​the particle board. At this time, a low-accuracy recognition signal is generated and the recognition result is classified as an inaccurate recognition result.

[0028] Preferably, in S6, for the incompletely accurate recognition result, the abnormal degree of the quality accuracy of the edge area of ​​the particleboard recognized by the detection system within a fixed time period is analyzed, specifically:

[0029] For incompletely accurate identification results, that is, the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy values ​​of the detection system identifying the quality of the edge area of ​​the particle board generated within a subsequent fixed time period that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a data set is established, and the standard deviation of the data set is calculated as the abnormality degree value of the accuracy of the detection system identifying the quality of the edge area of ​​the particle board within the fixed time period;

[0030] The acquired abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is compared with the abnormality degree value reference threshold under the normal state set according to historical data. If the abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is greater than or equal to the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particle board identified by the detection system within a fixed time period is high, and an early warning signal is generated at this time; if the abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is less than the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particle board identified by the detection system within a fixed time period is low, and no early warning signal is generated at this time.

[0031] Preferably, if the degree of abnormality is high, the deviation between the recognition accuracy of the current batch and the historical batch is calculated, and if the deviation of the quality inspection result is large, the corresponding particle board is automatically marked and the operator is reminded to review, specifically: if the degree of abnormality is high, the deviation between the recognition accuracy of the current batch and the historical batch is calculated, and if the deviation of the quality inspection result is large, the corresponding particle board is automatically marked and the operator is reminded to review, specifically:

[0032] Set the recognition accuracy value of the current batch to , the average recognition accuracy of historical batches is , deviation calculation formula: ; Where: ΔA is the deviation between the current batch and the historical batch;

[0033] A tolerance deviation threshold δ is set. If the deviation ΔA is greater than the tolerance deviation threshold δ, it is considered that the recognition accuracy of the current batch is greatly different from that of the historical batches, and there is a potential quality problem. The current batch of particleboard is automatically marked as an abnormal product, and the operator is reminded to recheck.

[0034] The present invention also provides a particleboard quality detection system based on image recognition, including a data acquisition module, a smoothness analysis module, a texture analysis module, an accuracy evaluation module, a classification module and a detection optimization module:

[0035] Data collection module: select a series of particleboards from different production batches, ensure that multiple samples with edge defects are included, and manually annotate the real defects in the edge area of ​​each sample, including cracks, bubbles, roughness, and uneven edges;

[0036] Smoothness analysis module: perform local analysis on edge areas, select multiple segmented windows to calculate grayscale changes, calculate the grayscale standard deviation of each window, mark the areas where smoothness changes suddenly, and calculate the ratio of smoothness changes by comparing the areas where smoothness remains unchanged and changes suddenly, and obtain the edge smoothness change rate;

[0037] Texture analysis module: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing the different texture blocks in the edge area of ​​the particle board. According to the change of the brightness difference of the texture in each area, the texture contrast difference of the edge area of ​​the particle board is calculated;

[0038] Accuracy evaluation module: The acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particleboard are converted into a comprehensive feature vector as the input of the machine learning model. The accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard is determined based on the output of the model.

[0039] Classification module: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard, the quality identification results of the detection system are divided into accurate identification results, incompletely accurate identification results and inaccurate identification results;

[0040] Detection optimization module: For inaccurate identification results, the abnormal degree of the detection system's identification of the quality accuracy of the edge area of ​​the particleboard within a fixed time period is analyzed. If the abnormal degree is high, the deviation between the current batch's identification accuracy and the historical batch is calculated. If the quality inspection result has a large deviation, the corresponding particleboard is automatically marked and the operator is reminded to review it.

[0041] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0042] 1. The present invention improves the detection accuracy of edge defects by performing fine grayscale change analysis, smoothness change rate and texture contrast difference calculation on the edge area of ​​particleboard, and combining machine learning model for prediction. This method can accurately identify edge defects such as cracks, bubbles, roughness and unevenness, avoiding defective products from entering the production process due to missed edge quality problems, thereby improving product quality and production efficiency.

[0043] 2. The present invention automatically marks and prompts the operator to review the inaccurate recognition results by performing abnormal detection and deviation analysis. This intelligent feedback mechanism can timely discover potential problems in system recognition, reduce human intervention, and improve the reliability and stability of detection. By effectively reducing quality accidents caused by detection errors, especially the potential risks in furniture manufacturing or construction applications, the present invention greatly improves the safety of particleboard use and ensures the mechanical properties and safety of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0045] Figure 1 The present invention is a flow chart of the method.

[0046] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Example 1, please refer to Figure 1 and Figure 2 As shown, the particleboard quality detection method based on image recognition described in this embodiment includes the following steps:

[0049] S1: A series of particleboards were selected from different production batches, including multiple samples with edge defects, and the real defects in the edge area of ​​each sample were manually annotated, including cracks, bubbles, roughness, and uneven edges;

[0050] S2: Perform local analysis on the edge area, select multiple segmented windows to calculate the grayscale change, calculate the grayscale standard deviation of each window, mark the area where the smoothness changes suddenly, and calculate the ratio of the smoothness change by comparing the smoothness unchanged and sudden change areas to obtain the edge smoothness change rate;

[0051] S3: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing different texture blocks in the edge area of ​​the particle board, and the texture contrast difference of the edge area of ​​the particle board is calculated according to the change of the brightness difference of the texture of each area;

[0052] S4: converting the acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board into a comprehensive feature vector as the input of the machine learning model, and determining the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board according to the output result of the model;

[0053] S5: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board, the quality identification result of the detection system is divided into an accurate identification result, an incompletely accurate identification result and an inaccurate identification result;

[0054] S6: For inaccurate identification results, analyze the degree of abnormality in the quality accuracy of the edge area of ​​the particleboard identified by the detection system within a fixed time period. If the degree of abnormality is high, calculate the deviation between the identification accuracy of the current batch and the historical batches. If the deviation of the quality inspection result is large, automatically mark the corresponding particleboard and remind the operator to recheck.

[0055] In S1, a series of particleboards were selected from different production batches, including multiple samples with edge defects. The real defects in the edge area of ​​each sample were manually annotated, including cracks, bubbles, roughness, and uneven edges, specifically:

[0056] In order to ensure that the detection system can accurately identify different production batches and different types of edge defects, the selected particleboard samples should have the following characteristics:

[0057] Batch diversity: Samples are selected from different production batches. This is because the production process may vary from batch to batch, so the type and number of defects may also vary. Selecting samples from multiple batches ensures that the samples cover the various defects that may occur during the production process.

[0058] Defect Variety: Make sure the selected particleboard samples contain different types of edge defects. These defects may include, but are not limited to:

[0059] Cracks: such as fine cracks and through cracks, often occur in the stress-bearing area of ​​particleboard. Bubbles: Due to air ingress or uneven materials during the production process, bubbles usually appear at the edges of particleboard. Roughness: Due to the uneven surface of particleboard or insufficient leveling, the surface is rough and uneven, especially at the edges. Uneven edges: For example, the edges of particleboard have irregular shapes, burrs, undulations, etc.

[0060] Select a sufficient number of particleboard samples to ensure that the samples are representative in terms of production batches, defect types and defect levels. Usually, at least 10 to 20 samples are selected to cover all major defect types. After selecting the samples, the next step is to manually annotate each sample, especially the defects in the edge areas. Manual annotation is a key step to ensure that the image recognition system can correctly learn the defect characteristics. The annotation process usually includes the following steps:

[0061] In image processing, the edge area usually refers to the surrounding area of ​​the particleboard, which is a certain distance away from the outer edge of the board. When marking, the edge area should be accurately defined, which will help with subsequent image processing and feature extraction. When marking, it is important to ensure accuracy, especially the specific locations of defects such as cracks, bubbles, roughness, and uneven edges, to avoid mislabeling or missing labels.

[0062] For each defect, the annotator needs to select the corresponding defect type and mark it according to the actual situation: Crack: Use a polygon or rectangular frame to mark the crack area and indicate the crack type (such as fine cracks, through cracks, etc.). Bubble: Mark a circular or elliptical area in the bubble area, indicating the size and shape of the bubble. Roughness: The rough area can be marked by a polygon or irregular area, and try to describe its surface roughness. Uneven edges: Use polygons or broken lines to mark the uneven edge parts, and mark the irregularity of the edge as detailed as possible.

[0063] When marking defects, in addition to marking the location of the defects, they can also be classified according to their severity. For example, cracks can be divided into small cracks and large cracks, bubbles can be divided into tiny bubbles and obvious bubbles, roughness can be divided into slight roughness and severe roughness, etc. This step helps the subsequent algorithm learn the characteristics of different defect types and defect severity. You can use some open source annotation tools (such as LabelImg, VIA, RectLabel, etc.) to complete the annotation work. These tools can support multiple annotation types, such as rectangular boxes, polygons, multi-point annotations, etc. For more complex defect types, such as cracks, you may need to use precise line or irregular polygon annotation tools.

[0064] Different annotators or experts should review to ensure that each defect is correctly annotated, especially in the edge area, to check whether the defects such as cracks, bubbles, roughness, and uneven edges are accurately annotated. For some edge defects that are difficult to judge, multiple annotations or cross-verification by different annotators may be required to ensure the consistency and accuracy of the annotation results. The edge area is easily affected by factors such as lighting and viewing angle, which may cause some defects to be unannotated or mislabeled. Therefore, it is recommended to verify the consistency of the annotation of defects in the edge area by comparing the annotation results under different image angles and different lighting conditions.

[0065] Save all annotation results as a dataset in a unified format (such as XML, JSON, or COCO format) for subsequent training and testing. Ensure that all types of defects (cracks, bubbles, roughness, uneven edges) are distributed relatively evenly in the dataset. You can ensure the diversity and comprehensiveness of the dataset by adding some samples of categories with fewer defects.

[0066] The labeled particleboard image dataset is used to train the image recognition model, especially the edge area detection model. The model is trained through deep learning or traditional machine learning algorithms (such as convolutional neural network CNN, support vector machine SVM, etc.) to enable it to recognize and classify different types of edge defects. The performance of the trained model is evaluated using methods such as cross-validation to ensure that the model can run stably on various production batches and defect types.

[0067] In this application, selecting particleboard samples from different production batches and manually marking defects is a basic step to ensure the efficient operation of the quality inspection system based on image recognition. By accurately marking the edge defects in each sample and recording the defect type, location and severity, high-quality data support can be provided for subsequent algorithm training, thereby improving the accuracy and reliability of the particleboard quality inspection system.

[0068] S2: Perform local analysis on the edge area, select multiple segmented windows to calculate the grayscale change, calculate the grayscale standard deviation of each window, mark the area where the smoothness changes suddenly, and calculate the ratio of smoothness change by comparing the smoothness unchanged and mutation areas to obtain the edge smoothness change rate.

[0069] First, ensure that the quality of the input image is suitable for subsequent analysis. In this step, the image's noise, brightness, and contrast are mainly addressed in order to extract clear edge information.

[0070] Denoising: Use methods such as Gaussian filtering or median filtering to remove noise from the image and reduce the impact of background noise on subsequent grayscale calculations.

[0071] Grayscale: Convert the image to grayscale. Grayscale images can better reflect the changes in the surface texture of particleboard, especially the defects in the edge area.

[0072] Normalization: Normalize the image to make the grayscale value range more uniform and reduce the impact of lighting differences between different images.

[0073] The edge area of ​​the particle board is extracted by edge detection algorithms (such as Canny edge detection, Sobel operator, Laplacian operator, etc.). The purpose of edge detection is to determine the part of the image with large grayscale changes, which is usually the area where defects appear.

[0074] Edge detection: Apply edge detection algorithms to obtain clear edge parts in the image. Edge area delineation: Based on the detected edge parts, manually or automatically define a rectangular box or irregular area to limit the edge area of ​​the image for subsequent analysis.

[0075] The edge region is divided into multiple segmentation windows, each of which represents a small local area in the image. The size of each window can be adjusted according to actual needs, and smaller windows (such as 3x3 or 5x5 pixels) are usually used for more detailed analysis.

[0076] Window size selection: Select a window of appropriate size based on the resolution of the particleboard and the inspection target. If it is a larger defect (such as a crack), a larger window may be required; for smaller defects (such as tiny bubbles or slight roughness), a smaller window should be selected.

[0077] Gaussian blurring the image makes the details and smoothness of the image more obvious. Gaussian blurring uses a Gaussian kernel to convolve the image. The formula is: ; Where: G(x,y) is the value of the Gaussian kernel, x and y are the distances between the current pixel and the center of the kernel, σ is the standard deviation, which controls the degree of blur. The larger the σ value, the stronger the blur effect. Convolve the image with the Gaussian kernel to get the blurred image .

[0078] Edge detection is a key step in the calculation of smoothness change rate, which can identify areas in the image where the smoothness changes suddenly (usually cracks, bubbles or uneven places).

[0079] Use the Sobel operator or Canny edge detection to detect the edges in the image. Use the Sobel operator for edge detection, and its gradient calculation formula is: ; ;in, and are the gradients along the x-axis and y-axis respectively, I(x,y) is the grayscale value of the image at position (x,y), and then the gradient magnitude G and gradient direction θ are calculated: , ; By setting a threshold to determine whether it is an edge pixel, the commonly used threshold methods are Otsu adaptive threshold method or fixed threshold.

[0080] The smoothness change rate of the edge area can be obtained by comparing the changes in the original image and the blurred image.

[0081] At each pixel location, calculate the grayscale difference between the blurred image and the original image: ; Where I(x,y) is the gray value of the original image, is the gray value of the blurred image, and ΔI(x,y) is the gray difference of the pixel.

[0082] The pixels whose grayscale difference ΔI(x,y) is greater than the set threshold are regarded as the area where the edge smoothness changes suddenly. The ratio of the smoothness mutation area in the edge area is calculated, and the total number of pixels in the edge area is set to , the number of pixels in the smoothness mutation area is , then the smoothness change rate The calculation expression is: ;in, is the number of pixels whose grayscale difference is greater than the threshold, is the total number of pixels in the edge area.

[0083] When the edge smoothness change rate of the particleboard edge area is larger, it usually means that the smoothness of the edge area in the image has undergone a large mutation, and there may be obvious quality defects, such as cracks, bubbles, roughness or uneven edges. This shows that the detection system can successfully identify these mutation areas and the detection accuracy is high. Therefore, a larger smoothness change rate usually indicates that the detection system can accurately capture the quality problems in the edge area, improving the reliability of the overall quality assessment.

[0084] On the contrary, when the smoothness change rate of the edge area of ​​the particleboard is smaller, it means that the edge area in the image is relatively smooth and the smoothness change is not obvious. This may indicate that the edge quality of the particleboard is good, or that the detection system fails to effectively identify subtle defects. A small smoothness change rate may indicate that the detection system is not capable of identifying quality problems in the edge area, misjudging it as a defect-free area, thereby affecting the accuracy of the overall quality assessment. Therefore, a too small smoothness change rate may indicate that the detection system has errors in identifying edge defects, resulting in inaccurate quality assessment.

[0085] S3: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing different texture blocks in the edge area of ​​the particle board. According to the change in the brightness difference of the texture of each area, the texture contrast difference of the edge area of ​​the particle board is calculated.

[0086] Divide the edge region in the image into multiple smaller texture blocks. These texture blocks are usually rectangular or square sub-regions of the image. Choose an appropriate texture block size based on the resolution of the image and the complexity of the edge. Common sizes are m×m windows (e.g. 16×16 or 32×32 pixels). Divide the edge region into multiple non-overlapping texture blocks. The size of each texture block should ensure that local texture information can be captured.

[0087] In each texture block, the texture features of the area are calculated, such as brightness value, local standard deviation or gray level co-occurrence matrix, etc. For each texture block, the average brightness value of all pixels (i.e. the average gray value) is calculated, and the formula is: ;in, is the average brightness of the texture block, N is the total number of pixels in the texture block, is the gray value of the i-th pixel, and the standard deviation of the brightness of each texture block is calculated to indicate the variation of the texture. The formula is: ;in, is the brightness standard deviation of the texture block, indicating the contrast of the block.

[0088] For each pair of adjacent texture blocks, the brightness difference is calculated as the contrast difference, and the average brightness of the two texture blocks is set to and , then the brightness contrast difference The calculation formula is: ; In order to more comprehensively reflect the texture inhomogeneity in the edge area, the average value of the contrast difference between all texture blocks is calculated: ;in, is the average value of the texture contrast difference in the edge area, M is the logarithm of the texture blocks for calculating the contrast difference, and the texture contrast difference in the edge area is obtained by combining the contrast differences of all texture blocks. The expression is: ;in, is the total number of texture blocks in the edge region, is the contrast difference of the i-th texture block, and TC is the texture contrast difference.

[0089] When the texture contrast difference is larger, it usually means that there is a larger texture difference in the edge area of ​​the particleboard, which may be caused by cracks, roughness or other quality defects. A larger texture contrast difference indicates that the texture inhomogeneity in the image is higher, and the detection system can identify obvious edge problems and the detection results are more accurate. Therefore, a larger texture contrast difference usually reflects the high accuracy of the detection system in identifying edge quality defects, and can effectively locate irregular areas and improve the reliability of quality assessment.

[0090] On the contrary, when the texture contrast difference is smaller, it means that the texture of the edge area of ​​the particleboard is more uniform, smoother, with fewer quality defects or no obvious texture changes. This indicates that the detection system may fail to identify minor defects or changes when analyzing the edge area, resulting in a smaller contrast difference. A smaller texture contrast difference may mean that the detection system fails to fully capture subtle problems at the edge, and there may be misjudgments, resulting in inaccurate quality assessments. Therefore, a smaller contrast difference suggests that the detection system has certain limitations in identifying the quality of the edge area of ​​the particleboard.

[0091] S4: The acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board are converted into a comprehensive feature vector as the input of the machine learning model, and the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board is determined according to the output results of the model.

[0092] The edge smoothness change rate and texture contrast difference are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the edge area quality of the particleboard predicted by the detection system as the prediction target, and takes minimizing the sum of prediction errors of the accuracy value labels of the edge area quality of the particleboard identified by all detection systems as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The accuracy value of the edge area quality of the particleboard identified by the detection system is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0093] The method for obtaining the accuracy value of the detection system for identifying the quality of the edge area of ​​the particleboard is as follows: from the comprehensive feature vector training data of the trained machine learning model, the corresponding function expression is obtained: ; In the formula, is the output function of the model, is the smoothness change rate, TC is the texture contrast difference, Accuracy value for the detection system in identifying the quality of the edge area of ​​particleboard.

[0094] S5: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board, the quality identification result of the detection system is divided into an accurate identification result, an incompletely accurate identification result and an inaccurate identification result;

[0095] The acquired accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the first standard threshold and the second standard threshold respectively;

[0096] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than the second standard threshold, it means that the detection system has high accuracy in identifying the quality of the edge area of ​​the particle board, and a high-accuracy recognition signal is generated at this time, and the recognition result is classified as an accurate recognition result;

[0097] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the detection system has a general accuracy in identifying the quality of the edge area of ​​the particle board, and a medium accuracy recognition signal is generated at this time, and the recognition result is classified as an incompletely accurate recognition result;

[0098] If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is less than the first standard threshold, it means that the detection system has low accuracy in identifying the quality of the edge area of ​​the particle board. At this time, a low-accuracy recognition signal is generated and the recognition result is classified as an inaccurate recognition result.

[0099] For particleboards that are identified as accurate, the system confirms that the quality of their edge areas meets the standard requirements. Secondary verification can be performed to ensure the reliability of the test. For example, other test methods (such as manual inspection or other automated equipment) are used for verification to ensure that there are no missed tests or misjudgments. Accurately identified products are recorded and relevant test data is archived. The records include: particleboard batch number, production date, test time, test results, etc. This data will be helpful for future quality traceability and production optimization.

[0100] We analyze the production process data to accurately identify the result product, find the best operating parameters in the current process, and confirm which conditions (such as temperature, pressure, machine speed, etc.) can ensure high-quality particleboard production. We use this data to feed back into the production process to further optimize the production process and reduce the occurrence of unqualified products.

[0101] For particleboards with inaccurate identification results, the production line is first paused or slowed down to avoid more defective products. The detected particleboards are manually inspected or other inspection equipment is used to further confirm whether there are edge defects to avoid erroneous results affecting a large number of products.

[0102] Conduct in-depth analysis of inaccurate recognition results to identify the causes of recognition errors. Possible causes include: Lighting changes: uneven lighting conditions during detection, resulting in blurred or distorted edge texture information in the image. Image noise: noise or interference in the image affects the detection accuracy of texture contrast or smoothness. Insufficient model training: The machine learning model may not be fully trained, or certain types of defects (such as tiny cracks and slight surface roughness changes) are not fully reflected in the training data. For these problems, the detection algorithm can be adjusted or enhanced (for example, by adding training data, adjusting the optical system, or adopting more advanced algorithms).

[0103] If the recognition error is found to be related to the training model, the machine learning model can be retrained to enhance the model's ability to recognize complex texture changes or sudden changes in smoothness. Incremental learning methods can be used to continue training on the basis of the existing model to gradually adapt the model to more samples and complex defect patterns. It may be necessary to add more training data containing edge defect samples, especially those types that are misidentified by the system. By manually annotating more samples, the generalization ability of the system can be improved.

[0104] According to the accuracy value of the model output, adjust the standard threshold to optimize the recognition accuracy of the system under different production conditions. For example, if the system often misjudges in certain specific situations, you can adjust the parameters or set different thresholds according to these situations to reduce errors. For certain edge areas that are particularly difficult to judge, you can set additional "tolerance" or "secondary confirmation" processes to ensure that these uncertain results are checked twice before deciding whether to be counted as unqualified.

[0105] Particleboards that are marked as unqualified by manual confirmation or other verification methods need to be sorted and processed. These unqualified products can be reprocessed (such as trimming the edges or replacing some boards) or completely removed from the production line to avoid entering the market.

[0106] S6: For inaccurate identification results, analyze the degree of abnormality in the quality accuracy of the edge area of ​​the particleboard identified by the detection system within a fixed time period. If the degree of abnormality is high, calculate the deviation between the identification accuracy of the current batch and the historical batches. If the deviation of the quality inspection result is large, automatically mark the corresponding particleboard and remind the operator to recheck.

[0107] For incompletely accurate identification results, that is, the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy values ​​of the detection system identifying the quality of the edge area of ​​the particle board generated within a subsequent fixed time period that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a data set is established, and the standard deviation of the data set is calculated as the abnormality degree value of the accuracy of the detection system identifying the quality of the edge area of ​​the particle board within the fixed time period;

[0108] The acquired abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is compared with the abnormality degree value reference threshold under the normal state set according to historical data. If the abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is greater than or equal to the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particle board identified by the detection system within a fixed time period is high, and an early warning signal is generated at this time; if the abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is less than the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particle board identified by the detection system within a fixed time period is low, and no early warning signal is generated at this time.

[0109] If the degree of abnormality is high, the deviation between the recognition accuracy of the current batch and the historical batch is calculated. If the deviation of the quality inspection result is large, the corresponding particle board is automatically marked and the operator is reminded to recheck. Specifically:

[0110] Set the recognition accuracy value of the current batch to , the average recognition accuracy of historical batches is , deviation calculation formula: ; Where: ΔA is the deviation between the current batch and the historical batch.

[0111] A tolerance deviation threshold δ is set. If the deviation ΔA is greater than the tolerance deviation threshold δ, it is considered that the recognition accuracy of the current batch is greatly different from that of the historical batch, and there is a potential quality problem. The threshold δ can be set according to the specific situation, and is usually adjusted according to factors such as the fluctuation range of historical data and the fluctuation of production process.

[0112] If the deviation ΔA is greater than the preset tolerance deviation threshold δ, the current batch of particleboard is automatically marked as an abnormal product. At this time, the system can mark the abnormal products in the following ways: automatically mark the current batch of particleboard, mark it in the system, or distinguish these products through labels, warning prompts, etc. The detection system will send a review reminder to the operator, requiring the operator to manually check the products marked as abnormal. This can remind the operator through interface pop-up windows, mobile phone push notifications, warning lights, etc.

[0113] After the operator conducts a review, if it is confirmed that the current batch has quality problems, and if the review confirms that the problem is serious, the operator can temporarily stop the production line to prevent more defective products from flowing out. Products identified as unqualified are eliminated or repaired to ensure that the quality of products on the market meets the standards. Through the feedback information from the review, the error source of the detection system is analyzed, and the algorithm is further optimized or the detection parameters are adjusted to reduce the occurrence of similar deviations.

[0114] When the inspection system finds that the quality inspection results of the current batch have large deviations, in addition to manual review, the system should also have certain self-learning and optimization capabilities. When the system identifies abnormal deviations, it automatically adds new training data to help the model self-learn and improve detection accuracy. According to historical deviation data and production processes, the tolerance deviation threshold δ is dynamically adjusted to make the system more adaptable to the current production environment and product characteristics.

[0115] In this embodiment, first, samples are selected from different production batches, and edge defects (such as cracks, bubbles, roughness, uneven edges, etc.) are marked by manual annotation. Then, the grayscale change of the edge area is calculated by local analysis, the smoothness mutation area is marked, and the smoothness change ratio is calculated to obtain the edge smoothness change rate. Subsequently, the texture contrast difference of the edge area is further analyzed by calculating the contrast difference between different texture blocks in the edge area of ​​the particleboard. These feature information are converted into comprehensive feature vectors as input items of the machine learning model to evaluate the accuracy of the detection system for the quality of the edge area. According to the accuracy value output by the model, the detection results are divided into accurate, incompletely accurate and inaccurate recognition results. When there is a deviation in recognition accuracy, the system analyzes the deviation between the current batch and the historical batch. If the deviation is too large, the abnormal particleboard is automatically marked and the operator is reminded to review to ensure the continuous optimization of product quality and production process.

[0116] Embodiment 2, a particleboard quality detection system based on image recognition described in this embodiment includes a data acquisition module, a smoothness analysis module, a texture analysis module, an accuracy evaluation module, a classification module and a detection optimization module:

[0117] Data collection module: select a series of particleboards from different production batches, ensure that multiple samples with edge defects are included, and manually annotate the real defects in the edge area of ​​each sample, including cracks, bubbles, roughness, and uneven edges;

[0118] Smoothness analysis module: perform local analysis on edge areas, select multiple segmented windows to calculate grayscale changes, calculate the grayscale standard deviation of each window, mark the areas where smoothness changes suddenly, and calculate the ratio of smoothness changes by comparing the areas where smoothness remains unchanged and changes suddenly, and obtain the edge smoothness change rate;

[0119] Texture analysis module: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing the different texture blocks in the edge area of ​​the particle board. According to the change of the brightness difference of the texture in each area, the texture contrast difference of the edge area of ​​the particle board is calculated;

[0120] Accuracy evaluation module: The acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particleboard are converted into a comprehensive feature vector as the input of the machine learning model. The accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard is determined based on the output of the model.

[0121] Classification module: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard, the quality identification results of the detection system are divided into accurate identification results, incompletely accurate identification results and inaccurate identification results;

[0122] Detection optimization module: For inaccurate identification results, the abnormal degree of the detection system's identification of the quality accuracy of the edge area of ​​the particleboard within a fixed time period is analyzed. If the abnormal degree is high, the deviation between the current batch's identification accuracy and the historical batch is calculated. If the quality inspection result has a large deviation, the corresponding particleboard is automatically marked and the operator is reminded to review it.

[0123] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0124] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0125] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0126] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A particleboard quality detection method based on image recognition, characterized in that: The following steps are involved: S1: A series of particleboards were selected from different production batches, including multiple samples with edge defects, and the real defects in the edge area of ​​each sample were manually annotated, including cracks, bubbles, roughness, and uneven edges; S2: Perform local analysis on the edge area, select multiple segmented windows to calculate the grayscale change, calculate the grayscale standard deviation of each window, mark the area where the smoothness changes suddenly, and calculate the ratio of the smoothness change by comparing the smoothness unchanged and sudden change areas to obtain the edge smoothness change rate; Gaussian blur is performed on the image. Gaussian blur uses a Gaussian kernel to convolve the image. The formula is: Where: G(x,y) is the value of the Gaussian kernel, x and y are the distances between the current pixel and the center of the kernel, σ is the standard deviation, which controls the degree of blur. The image is convolved with the Gaussian kernel to obtain the blurred image I blurred (x,y); The Sobel operator is used for edge detection, and its gradient calculation formula is: Among them, G x and G y are the gradients along the x-axis and y-axis respectively, I(x,y) is the grayscale value of the image at position (x,y), and the gradient magnitude G and gradient direction θ are calculated: θ=arctan2(G y ,G x ); At each pixel position, calculate the grayscale difference between the blurred image and the original image: ΔI(x,y)=|I(x,y)-I blurred (x,y)|; where I(x,y) is the grayscale value of the original image, I blurred (x, y) is the gray value of the blurred image, ΔI(x, y) is the gray difference of the pixel; The pixels whose grayscale difference ΔI(x,y) is greater than the set threshold are regarded as the area where the edge smoothness has a sudden change. The ratio of the smoothness mutation area in the edge area is calculated, and the total number of pixels in the edge area is set to N. edges , the number of pixels in the smoothness mutation area is The smoothness change rate R smooth The calculation expression is: in, is the number of pixels whose grayscale difference is greater than the threshold, N edges is the total number of pixels in the edge area; S3: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing different texture blocks in the edge area of ​​the particle board, and the texture contrast difference of the edge area of ​​the particle board is calculated according to the change of the brightness difference of the texture of each area; For each texture block, calculate the average brightness value of all its pixels. The formula is: Among them, I avg is the average brightness of the texture block, N is the total number of pixels in the texture block, and I ( x i ,y i) is the gray value of the i-th pixel, and the standard deviation of the brightness of each texture block is calculated to indicate the variation of the texture. The formula is: Among them, σ block is the brightness standard deviation of the texture block, indicating the contrast of the texture block; For each pair of adjacent texture blocks, the brightness difference is calculated as the contrast difference, and the average brightness of the two texture blocks is set to I and avg1 and I avg2 , then the brightness contrast difference ΔI contrast The calculation formula is: ΔI contrast = | I avg1 -I avg2 | ; Calculate the average of the contrast differences between all texture patches: Among them, C contrast is the average value of the texture contrast difference in the edge area, M is the logarithm of the texture blocks for calculating the contrast difference, and the texture contrast difference in the edge area is obtained by combining the contrast differences of all texture blocks. The expression is: Among them, N blocks is the total number of texture blocks in the edge region, is the contrast difference of the i-th texture block, TC is the texture contrast difference; S4: converting the acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board into a comprehensive feature vector as the input of the machine learning model, and determining the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board according to the output result of the model; S5: According to the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board, the quality identification result of the detection system is divided into an accurate identification result, an incompletely accurate identification result and an inaccurate identification result; S6: For inaccurate identification results, analyze the degree of abnormality in the quality accuracy of the edge area of ​​the particleboard identified by the detection system within a fixed time period. If the degree of abnormality is high, calculate the deviation between the identification accuracy of the current batch and the historical batches. If the deviation of the quality inspection result is large, automatically mark the corresponding particleboard and remind the operator to recheck.

2. The particleboard quality detection method based on image recognition according to claim 1 is characterized in that: In S4, the edge smoothness change rate and texture contrast difference of the edge area of ​​the particle board are converted into a comprehensive feature vector as the input of the machine learning model. The accuracy of the detection system in identifying the quality of the edge area of ​​the particle board is determined according to the output result of the model, specifically: The edge smoothness change rate and texture contrast difference are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the edge area quality of the particleboard predicted by the detection system as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of the edge area quality of the particleboard identified by all detection systems as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The accuracy value of the edge area quality of the particleboard identified by the detection system is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

3. The particleboard quality detection method based on image recognition according to claim 2 is characterized in that: In S5, according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particle board, the quality recognition result of the detection system is divided into an accurate recognition result, an incompletely accurate recognition result and an inaccurate recognition result; The acquired accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board is compared with the first standard threshold and the second standard threshold respectively; If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than the second standard threshold, it means that the detection system has high accuracy in identifying the quality of the edge area of ​​the particle board, and a high-accuracy recognition signal is generated at this time, and the recognition result is classified as an accurate recognition result; If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the detection system has a general accuracy in identifying the quality of the edge area of ​​the particle board, and a medium accuracy recognition signal is generated at this time, and the recognition result is classified as an incompletely accurate recognition result; If the accuracy value of the detection system in identifying the quality of the edge area of ​​the particle board is less than the first standard threshold, it means that the detection system has low accuracy in identifying the quality of the edge area of ​​the particle board. At this time, a low-accuracy recognition signal is generated and the recognition result is classified as an inaccurate recognition result.

4. The particleboard quality detection method based on image recognition according to claim 1 is characterized in that: In S6, for the incompletely accurate recognition results, the abnormal degree of the quality accuracy of the edge area of ​​the particleboard recognized by the detection system within a fixed time period is analyzed, specifically: For incompletely accurate identification results, that is, the accuracy value of the detection system identifying the quality of the edge area of ​​the particle board generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy values ​​of the detection system identifying the quality of the edge area of ​​the particle board generated within a subsequent fixed time period that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected, and a data set is established, and the standard deviation of the data set is calculated as the abnormality degree value of the accuracy of the detection system identifying the quality of the edge area of ​​the particle board within the fixed time period; The acquired abnormality degree value of the quality accuracy of the edge area of ​​the particleboard identified by the detection system is compared with the abnormality degree value reference threshold under the normal state set according to historical data. If the abnormality degree value of the quality accuracy of the edge area of ​​the particleboard identified by the detection system is greater than or equal to the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particleboard identified by the detection system within a fixed time period is high, and an early warning signal is generated at this time; If the abnormality degree value of the quality accuracy of the edge area of ​​the particle board identified by the detection system is less than the abnormality degree value reference threshold, it means that the abnormality degree of the quality accuracy of the edge area of ​​the particle board identified by the detection system within a fixed time period is low, and no warning signal is generated at this time.

5. The particleboard quality detection method based on image recognition according to claim 4 is characterized in that: If the degree of abnormality is high, the deviation between the recognition accuracy of the current batch and the historical batch is calculated. If the deviation of the quality inspection result is large, the corresponding particle board is automatically marked and the operator is reminded to review it. Specifically: If the degree of abnormality is high, the deviation between the recognition accuracy of the current batch and the historical batch is calculated. If the deviation of the quality inspection result is large, the corresponding particle board is automatically marked and the operator is reminded to review it. Specifically: Set the recognition accuracy value of the current batch to Acurrent, and the average recognition accuracy value of the historical batch to Ahistory. The deviation calculation formula is: ΔA=|Acurrent-Ahistory|; where: ΔA is the deviation between the current batch and the historical batch; A tolerance deviation threshold δ is set. If the deviation ΔA is greater than the tolerance deviation threshold δ, it is considered that the recognition accuracy of the current batch is greatly different from that of the historical batches, and there is a potential quality problem. The current batch of particleboard is automatically marked as an abnormal product, and the operator is reminded to recheck.

6. A particleboard quality detection system based on image recognition, used to implement a particleboard quality detection method based on image recognition as claimed in any one of claims 1 to 5, characterized in that: It includes data acquisition module, smoothness analysis module, texture analysis module, accuracy assessment module, classification module and detection optimization module: Data collection module: select a series of particleboards from different production batches, ensure that multiple samples with edge defects are included, and manually annotate the real defects in the edge area of ​​each sample, including cracks, bubbles, roughness, and uneven edges; Smoothness analysis module: perform local analysis on edge areas, select multiple segmented windows to calculate grayscale changes, calculate the grayscale standard deviation of each window, mark the areas where smoothness changes suddenly, and calculate the ratio of smoothness changes by comparing the areas where smoothness remains unchanged and changes suddenly, and obtain the edge smoothness change rate; Texture analysis module: If the edge smoothness change rate is large, the contrast difference between each area is calculated by comparing the different texture blocks in the edge area of ​​the particle board. According to the change of the brightness difference of the texture in each area, the texture contrast difference of the edge area of ​​the particle board is calculated; Accuracy evaluation module: The acquired edge smoothness change rate and texture contrast difference of the edge area of ​​the particleboard are converted into a comprehensive feature vector as the input of the machine learning model. The accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard is determined based on the output of the model. Classification module: according to the accuracy of the detection system in identifying the quality of the edge area of ​​the particleboard, the quality identification results of the detection system are divided into accurate identification results, incompletely accurate identification results and inaccurate identification results; Detection optimization module: For inaccurate identification results, the abnormal degree of the detection system's identification of the quality accuracy of the edge area of ​​the particleboard within a fixed time period is analyzed. If the abnormal degree is high, the deviation between the current batch's identification accuracy and the historical batch is calculated. If the quality inspection result has a large deviation, the corresponding particleboard is automatically marked and the operator is reminded to review it.

Images