Computer vision-based raw material grading method for stone plastic floor

Abstract

The application discloses a stone-plastic floor raw material grading method based on computer vision and relates to the technical field of stone-plastic floor grading. In order to solve the problem that stone-plastic floor raw material defects cannot be accurately distinguished, the application extracts core features such as defects, colors, textures and shapes, filters key features through correlation analysis, redundancy identification and stability verification, combines artificial grading results, enhances the relevance of features and grades, covers different production conditions and states from sample preparation, sets up image capture environment and equipment standards, and optimizes multiple steps such as preprocessing noise reduction and geometric correction, and combines color-gray feature fusion and double verification of target separation to improve data reliability and consistency.

Description

Technical Field

[0001] This invention relates to the field of stone-plastic flooring grading technology, specifically a method for grading stone-plastic flooring raw materials based on computer vision. Background Technology

[0002] The current grading of raw materials for stone plastic flooring relies heavily on manual labor, which suffers from low efficiency, high subjectivity, and inconsistent standards, easily leading to grading deviations. Traditional machine vision methods often suffer from insufficient handling of image noise and background interference, inaccurate feature extraction, and difficulty in effectively identifying key indicators such as subtle defects and color differences. As a result, the grading accuracy is low and the stability is poor, making it difficult to adapt to the diverse testing needs under different production stages and raw material ratios. Summary of the Invention

[0003] The purpose of this invention is to provide a computer vision-based method for grading raw materials of stone-plastic flooring. By extracting core features such as defects, color, texture, and shape, and screening key features through correlation analysis, redundancy identification, and stability verification, the method enhances the correlation between features and grades by combining the results of manual grading. The method involves multiple optimization steps, from sample preparation covering different production conditions and states, to standardized settings of image capture environment and equipment, to preprocessing noise reduction and geometric correction, and combining color-grayscale feature fusion of target separation and dual verification to improve data reliability and consistency. This method can solve the problems in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A computer vision-based method for grading raw materials for stone-plastic flooring includes:

[0006] First, prepare the raw materials for the stone plastic flooring to be tested; capture visual images of the prepared raw materials; preprocess the captured images; separate the target object from the background or other areas in the preprocessed images; extract feature information reflecting the grade of the raw materials from the separated images; filter the grade-influencing features from the feature information; construct a sample dataset based on the selected grade-influencing features and the raw materials; train a model on the constructed sample dataset and make grade judgments based on the trained model; analyze and output the grade judgment results and provide a feedback mechanism.

[0007] Preferably, the raw materials for the stone-plastic flooring to be tested are prepared, including:

[0008] First, samples of the raw materials of the stone plastic flooring to be tested are extracted. The raw materials of the stone plastic flooring to be tested include samples from different production periods, different raw material ratios, and different appearance states.

[0009] The extracted samples are cleaned and then air-dried or dried with cold air.

[0010] The samples that are air-dried or dried with cold air are subjected to morphological pretreatment. Morphological pretreatment involves cutting or trimming the selected samples into uniform size specifications according to actual testing requirements.

[0011] After the morphological pretreatment is completed, the final raw material of the stone plastic flooring to be tested is obtained.

[0012] Preferably, the prepared stone plastic flooring raw materials are subjected to visual image capture, including:

[0013] Before visual image capture of the raw materials for stone plastic flooring, the image capture environment must first be set up;

[0014] The shooting area is set up in a closed or semi-closed space and an adjustable brightness shadowless light source system is used. At the same time, a solid color non-reflective background board is laid on the shooting table. The background board is made of matte material.

[0015] After the image capture environment is set up, the shooting equipment is debugged and its parameters are set.

[0016] First, the industrial camera is fixed on a tripod, and the vertical distance between the camera lens and the inspection surface of the stone plastic flooring raw material is adjusted. Then, the industrial camera is connected to the computer, and the parameters are set through image acquisition software.

[0017] After the shooting equipment is debugged and the parameters are set, the stone plastic flooring raw material is placed in the center of the background board for image acquisition;

[0018] The image acquisition process involves: starting the image acquisition software, taking pictures of the raw materials of the stone plastic flooring according to the preset parameters, obtaining a frontal image of the raw materials of the stone plastic flooring after taking the pictures, adjusting the angle of the industrial camera, taking pictures from the 45-degree angle and the side of the raw materials of the stone plastic flooring respectively, and acquiring at least 2 images from each angle.

[0019] Finally, the original visual images of the raw materials for stone plastic flooring were captured.

[0020] Preferably, the captured images of the stone plastic flooring undergo image preprocessing, including:

[0021] Image preprocessing is performed on the captured images of the raw materials for stone plastic flooring;

[0022] The image preprocessing workflow is as follows:

[0023] S1: Noise Reduction Processing: A multi-level filtering method is used to process the sensor noise and ambient light spots in the captured image of the raw material of the stone plastic flooring.

[0024] S2: Contrast Correction: Based on the image grayscale histogram analysis, the dynamic range of overly dark or overly bright images in the denoised raw material image of the stone plastic flooring is adjusted.

[0025] S3: Geometric normalization processing: The actual outline of the stone plastic flooring raw material image after contrast correction is located by edge detection, the redundant area of ​​the background board in the image is cropped, and only the complete area of ​​the stone plastic flooring sample is retained. The angle of the tilted part in the stone plastic flooring raw material image is corrected according to the horizontal reference line of the edge of the stone plastic flooring raw material sample.

[0026] S4: Color Space Unification: Convert the colors in the image of the stone plastic flooring raw material after geometric normalization to the HSV color space. The HSV color space includes a luminance channel and a chrominance channel. Then, perform grayscale value processing on the luminance channel and the chrominance channel respectively.

[0027] S5: Quality Inspection: The image of the raw material of the stone plastic flooring with unified color space is evaluated for clarity. The unqualified parts in the evaluation results are pre-processed again until the clarity evaluation is qualified.

[0028] The final step is to complete the preprocessing of the captured stone plastic flooring images.

[0029] Preferably, separating the target object from the background or other areas in the preprocessed stone plastic flooring image includes:

[0030] The process for separating images of stone-plastic flooring is as follows:

[0031] First, initial segmentation is performed based on color features. The standard color range of the background board is extracted and a color threshold is set. Then, pixels in the stone plastic flooring image that meet the background features are marked as background areas, and the remaining pixels are classified as target candidate areas.

[0032] Then, the grayscale features are combined for segmentation. First, a grayscale threshold is set. After the grayscale value is set, pixels in the target candidate region that are lower than the target grayscale lower limit are reclassified as background.

[0033] Edge contour optimization is performed on the image after grayscale value segmentation, wherein the target candidate region is corrected based on the geometric normalization process.

[0034] After the edge contour optimization is completed, interference regions are removed. This involves confirming the area and shape factor of each connected region, setting a reasonable size range for the target object, and identifying regions with too small an area or shapes that do not conform to the geometric characteristics of the sample as interference, marking them as background and removing them from the target region.

[0035] After the interference area is removed, the separation results are verified. The separation results verification includes visual inspection and pixel ratio analysis. The visual inspection is to confirm that the target object covers the stone plastic floor sample and that there is no background residue or missing target area. The deviation between the edge and the actual boundary of the stone plastic floor sample is controlled within 2 pixels. The pixel ratio analysis is to count the ratio of the number of pixels in the target area to the total number of pixels in the image. The deviation between the sample size and the image size is less than 3%. If it is greater than 3%, the grayscale feature segmentation is re-performed.

[0036] After the separation results are verified, an image of the separated stone-plastic flooring is obtained.

[0037] Preferably, feature information reflecting the raw material grade is extracted from the separated stone-plastic flooring image, including:

[0038] The core indicators reflecting the characteristic information of raw material grades are defect characteristics, color characteristics, texture characteristics, and morphological detail characteristics;

[0039] Defect features, color features, texture features, and morphological detail features are extracted respectively;

[0040] Among them, the extraction of defect features is as follows: by analyzing the gray value, color difference or texture change of pixels in the image, the region that is significantly different from the normal region is identified, the potential defects are initially located, and the attribute features of each located defect are extracted, including size features, morphological features and distribution features.

[0041] The color feature extraction is as follows: perform overall color analysis on the separated stone plastic floor image, extract the main color tone, perform consistency analysis on the main color tone, determine whether there are isolated color areas that differ from the main color tone based on the analysis results, and record the area and position of the isolated color areas.

[0042] The extraction of texture features involves dividing the separated stone plastic floor image into several local texture analysis regions, and analyzing the texture direction, texture density, and texture consistency of each texture analysis region.

[0043] The extraction of morphological details involves: performing edge regularity analysis on the image contour of the separated stone plastic flooring image. Edge regularity analysis detects whether there are irregular shapes such as burrs, depressions or protrusions on the edge of the image contour, counts the number of abnormal points on the edge that deviate from the smooth curve and the degree of deviation. At the same time, for the image taken from the side, the contour line of the side edge is extracted, the thickness value at different positions is measured, and the maximum deviation value of the thickness and the overall mean deviation are confirmed.

[0044] The extracted defect features, color features, texture features, and morphological detail features are summarized to form an initial feature dataset, which ultimately yields feature information reflecting the grade of the raw materials.

[0045] Preferably, the hierarchical influence features are filtered from the feature information, including:

[0046] Standardize the description of feature information and unify the measurement method of features;

[0047] After the measurement method is standardized, each feature information is correlated with the grade label, where the grade label is the manual grading result of the introduced samples;

[0048] Determine the distribution differences of each feature based on the correlation analysis results;

[0049] Redundant features are identified based on the distribution differences of each feature information. Redundant feature identification involves analyzing the intrinsic relationship between features and identifying highly redundant features. Specifically, if two features describe the same quality attribute in physical meaning or their numerical change trends are completely consistent, they are determined to be redundant features.

[0050] The redundancy features are evaluated for stability. If the value fluctuation of a redundancy feature is less than 5% in repeated tests, it is considered stable; if the fluctuation exceeds 15%, it is considered unstable and is removed.

[0051] The results of distribution difference, redundancy feature identification, and stability verification are prioritized and ranked. The prioritization is based on a comprehensive score of the results of distribution difference, redundancy feature identification, and stability verification, and the results are ranked from highest to lowest score.

[0052] The top 80% of the features in the ranking results are selected as candidate key features, and these candidate key features are used as hierarchical influence features in the feature information.

[0053] Preferably, a sample dataset is constructed based on the selected hierarchical influence features and the raw materials of SPC flooring, including:

[0054] The basic information of the raw material samples of stone plastic flooring is summarized, including sample number, production batch, original grade label and the path of the collected multi-angle images, and a sample information summary table is established.

[0055] For each stone plastic flooring raw material sample, match its corresponding graded influence characteristic data, and obtain the sample characteristic data of the stone plastic flooring raw material sample after the matching is completed;

[0056] Based on the results of manual grading, the sample feature data are labeled with grade labels, and labeling rules are established to clarify the feature thresholds corresponding to each grade.

[0057] The sample feature data with completed grade labels are divided into training set, validation set and test set. The quality of the divided dataset is verified. The quality verification is to verify whether the feature value of each sample feature data is within a reasonable range and whether the grade label is consistent with the actual situation of the sample feature data.

[0058] The validated datasets are stored as structured files for training, validation, and test sets respectively. At the same time, a dataset description document is generated, which records the data source, partition ratio, feature meaning, and processing method.

[0059] Finally, the sample dataset was constructed.

[0060] Preferably, the constructed sample dataset is used to train a model, and a grading judgment is made based on the trained model, including:

[0061] The model type is selected based on the size of the sample dataset, the type of features, and the requirements of the hierarchical task, including machine learning models or deep learning models.

[0062] The training parameters are set according to the model type. The training parameters include basic parameters and regularization parameters. Then, the training parameters are tuned. The parameter tuning is carried out by testing different combinations within the preset parameter range through grid search or random search methods. The optimal parameter combination is selected with the hierarchical accuracy of the validation set as the indicator.

[0063] After setting the training parameters, the model is trained using the training set data, and the training process is monitored in real time.

[0064] The trained model is then evaluated using a separate test set, including hierarchical accuracy evaluation, category precision and recall evaluation, and confusion matrix analysis.

[0065] The model after performance evaluation is classified and judged, in which the feature data of the sample to be tested is extracted first;

[0066] The extracted feature data of the sample to be detected is input into the model after the performance evaluation is completed for model prediction. The model outputs the predicted level label based on the learned features and hierarchical association rules, and records the confidence of the model in the prediction result.

[0067] The model prediction results are then validated, including confidence and consistency checks.

[0068] The model prediction data that passed the verification in the verification results are associated and stored with the verification prediction level and sample information, and a classification judgment result is generated.

[0069] Finally, the model training and classification judgment of the sample dataset are completed.

[0070] Preferably, the grading judgment results are analyzed and output, and a feedback mechanism is provided, including:

[0071] First, the results of the classification judgment are analyzed, including basic statistical analysis, misclassification analysis, and feature impact analysis.

[0072] The analysis results are then visualized, including a grade distribution bar chart, a misclassification heatmap, and a feature correlation scatter plot, while generating production monitoring reports, model performance reports, and raw data reports.

[0073] The generated report is sent to the corresponding display terminal for data display.

[0074] Staff members view the data on the display terminal, perform manual verification, record the verification results, and generate a structured feedback form based on the verification results.

[0075] The structured feedback table is used for classification and priority sorting, including feature recognition class, grade standard class and model stability class;

[0076] Based on the classification and priority sorting results, the tasks are assigned to the corresponding staff for processing and execution.

[0077] And record the processing results.

[0078] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0079] 1. The computer vision-based method for grading raw materials of stone plastic flooring provided by this invention covers different production conditions and states in sample preparation, standardizes the environment and equipment settings for image capture, and optimizes multiple steps such as noise reduction and geometric correction in preprocessing. Combined with color-grayscale feature fusion of target separation and dual verification, it improves data reliability and consistency.

[0080] 2. The computer vision-based grading method for stone plastic flooring raw materials provided by this invention extracts core features such as defects, color, texture, and shape, and filters key features through correlation analysis, redundancy identification, and stability verification. Combined with the results of manual grading, it enhances the correlation between features and grades.

[0081] 3. The computer vision-based grading method for stone plastic flooring raw materials provided by this invention selects models based on dataset characteristics, optimizes parameters using grid and random search, and evaluates performance using multiple indicators. By combining visual analysis and manual verification, a feedback loop is formed for issues related to features, grading standards, and model stability, thereby achieving grading accuracy and continuous optimization. Attached Figure Description

[0082] Figure 1 This is a schematic diagram of the raw material grading process for stone-plastic flooring according to the present invention;

[0083] Figure 2 This is a schematic diagram of the raw material grading process for stone-plastic flooring according to the present invention. Detailed Implementation

[0084] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] To address the issues of poor image quality caused by improper sample preparation, inadequate shooting environment and equipment, and noise, distortion, and color inconsistency resulting from insufficient preprocessing in existing technologies for testing stone plastic flooring, please refer to [link to relevant documentation]. Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0086] A computer vision-based method for grading raw materials for stone-plastic flooring includes:

[0087] First, prepare the raw materials for the stone plastic flooring to be tested; capture visual images of the prepared raw materials; preprocess the captured images; separate the target object from the background or other areas in the preprocessed images; extract feature information reflecting the grade of the raw materials from the separated images; filter the grade-influencing features from the feature information; construct a sample dataset based on the selected grade-influencing features and the raw materials; train a model on the constructed sample dataset and make grade judgments based on the trained model; analyze and output the grade judgment results and provide a feedback mechanism.

[0088] Prepare the raw materials for the stone-plastic flooring to be tested, including:

[0089] First, samples of the raw materials of the stone plastic flooring to be tested are extracted. The raw materials of the stone plastic flooring to be tested include samples from different production periods, different raw material ratios, and different appearance states.

[0090] The extracted samples are cleaned and then air-dried or dried with cold air.

[0091] The samples that are air-dried or dried with cold air are subjected to morphological pretreatment. Morphological pretreatment involves cutting or trimming the selected samples into uniform size specifications according to actual testing requirements.

[0092] After the morphological pretreatment is completed, the final raw material of the stone plastic flooring to be tested is obtained.

[0093] Specifically, samples from different production periods, with varying raw material ratios, and in different appearance states ensure broad representativeness. This comprehensively reflects the quality of raw materials under various production conditions and their inherent state, avoiding biased results due to a single sample and more accurately reflecting the overall quality level of the raw materials. The cleaning process removes surface impurities from the samples through natural air drying or cold air drying, effectively preventing these impurities from interfering with the test results. Whether it's surface dust or potential oil stains, these can affect the accurate assessment of the raw material's performance. Cleaning ensures the purity of the tested object, thereby improving the accuracy of the test results. Morphological pretreatment involves cutting or trimming the samples to a uniform size, ensuring all samples are under the same morphological conditions. This operation allows subsequent testing to be conducted on a consistent basis, reducing testing errors caused by differences in sample morphology and making the test data from different samples more comparable, providing favorable conditions for accurate analysis and evaluation of raw material quality.

[0094] Visual image capture of the prepared stone plastic flooring raw materials was performed, including:

[0095] Before visual image capture of the raw materials for stone plastic flooring, the image capture environment must first be set up;

[0096] The shooting area is set up in a closed or semi-closed space and an adjustable brightness shadowless light source system is used. At the same time, a solid color non-reflective background board is laid on the shooting table. The background board is made of matte material.

[0097] After the image capture environment is set up, the shooting equipment is debugged and its parameters are set.

[0098] First, the industrial camera is fixed on a tripod, and the vertical distance between the camera lens and the inspection surface of the stone plastic flooring raw material is adjusted. Then, the industrial camera is connected to the computer, and the parameters are set through image acquisition software.

[0099] After the shooting equipment is debugged and the parameters are set, the stone plastic flooring raw material is placed in the center of the background board for image acquisition;

[0100] The image acquisition process involves: starting the image acquisition software, taking pictures of the raw materials of the stone plastic flooring according to the preset parameters, obtaining a frontal image of the raw materials of the stone plastic flooring after taking the pictures, adjusting the angle of the industrial camera, taking pictures from the 45-degree angle and the side of the raw materials of the stone plastic flooring respectively, and acquiring at least 2 images from each angle.

[0101] Finally, the original visual images of the raw materials for stone plastic flooring were captured.

[0102] Specifically, in the environment setup phase, the enclosed space and shadowless light source system can isolate external light interference and avoid image differences caused by natural light fluctuations; the matte solid color background can eliminate reflections and noise interference, ensuring image background uniformity, highlighting sample details, and eliminating environmental interference factors for image recognition; the fixed industrial camera and vertical distance calibration ensure shooting stability and reduce image blurring caused by equipment shaking; standardized acquisition is achieved through software preset parameters to ensure that images of different samples are consistent in terms of resolution, exposure, and other indicators, improving image quality stability; image acquisition adopts a multi-dimensional shooting strategy, with a combination of front, 45-degree angle, and side views covering all angle features of the sample, which can comprehensively capture details such as surface texture and edge state; the setting of at least 2 images for each angle can reduce the random error of a single shot and improve the integrity and reliability of sample images.

[0103] The captured images of the stone plastic flooring underwent image preprocessing, including:

[0104] Image preprocessing is performed on the captured images of the raw materials for stone plastic flooring;

[0105] The image preprocessing workflow is as follows:

[0106] S1: Noise Reduction Processing: A multi-level filtering method is used to process the sensor noise and ambient light spots in the captured image of the raw material of the stone plastic flooring.

[0107] S2: Contrast Correction: Based on the image grayscale histogram analysis, the dynamic range of overly dark or overly bright images in the denoised raw material image of the stone plastic flooring is adjusted.

[0108] S3: Geometric normalization processing: The actual outline of the stone plastic flooring raw material image after contrast correction is located by edge detection, the redundant area of ​​the background board in the image is cropped, and only the complete area of ​​the stone plastic flooring sample is retained. The angle of the tilted part in the stone plastic flooring raw material image is corrected according to the horizontal reference line of the edge of the stone plastic flooring raw material sample.

[0109] S4: Color Space Unification: Convert the colors in the image of the stone plastic flooring raw material after geometric normalization to the HSV color space. The HSV color space includes a luminance channel and a chrominance channel. Then, perform grayscale value processing on the luminance channel and the chrominance channel respectively.

[0110] S5: Quality Inspection: The image of the raw material of the stone plastic flooring with unified color space is evaluated for clarity. The unqualified parts in the evaluation results are pre-processed again until the clarity evaluation is qualified.

[0111] The final step is to complete the preprocessing of the captured stone plastic flooring images.

[0112] Specifically, multi-level filtering noise reduction effectively eliminates sensor noise and ambient light spots, preventing interference from masking image features and ensuring the purity of the original image. This preserves true details for subsequent processing. Contrast correction based on grayscale histograms dynamically adjusts overly dark or bright areas, enhancing the layering of image details and making previously blurred textures and colors easier to identify, thus improving the image's information expression. Geometric standardization, through edge detection to crop redundant backgrounds and correct tilt angles, unifies the geometric shape of sample images, eliminating interference from positional deviations during shooting and providing a spatial benchmark for horizontal comparison between different sample images. Conversion to the HSV color space and separate processing of brightness and color channels enable structured analysis of color information, unifying color analysis standards and facilitating targeted extraction of color features, reducing errors caused by differences in color modes. A quality verification mechanism ensures image clarity through secondary preprocessing, forming a closed-loop control that prevents unqualified images from entering subsequent stages, ensuring the reliability of preprocessing results from the source.

[0113] To address the issues of inaccurate target-background separation and significant edge deviation in existing technologies for stone plastic flooring images, please refer to [link to relevant documentation]. Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0114] Separating the target object from the background or other areas in the preprocessed stone plastic flooring image includes:

[0115] The process for separating images of stone-plastic flooring is as follows:

[0116] First, initial segmentation is performed based on color features. The standard color range of the background board is extracted and a color threshold is set. Then, pixels in the stone plastic flooring image that meet the background features are marked as background areas, and the remaining pixels are classified as target candidate areas.

[0117] Then, the grayscale features are combined for segmentation. First, a grayscale threshold is set. After the grayscale value is set, pixels in the target candidate region that are lower than the target grayscale lower limit are reclassified as background.

[0118] Edge contour optimization is performed on the image after grayscale value segmentation, wherein the target candidate region is corrected based on the geometric normalization process.

[0119] After the edge contour optimization is completed, interference regions are removed. This involves confirming the area and shape factor of each connected region, setting a reasonable size range for the target object, and identifying regions with too small an area or shapes that do not conform to the geometric characteristics of the sample as interference, marking them as background and removing them from the target region.

[0120] After the interference area is removed, the separation results are verified. The separation results verification includes visual inspection and pixel ratio analysis. The visual inspection is to confirm that the target object covers the stone plastic floor sample and that there is no background residue or missing target area. The deviation between the edge and the actual boundary of the stone plastic floor sample is controlled within 2 pixels. The pixel ratio analysis is to count the ratio of the number of pixels in the target area to the total number of pixels in the image. The deviation between the sample size and the image size is less than 3%. If it is greater than 3%, the grayscale feature segmentation is re-performed.

[0121] After the separation results are verified, an image of the separated stone-plastic flooring is obtained.

[0122] Specifically, by extracting standard intervals for the background based on color features and setting thresholds, the background area can be quickly identified, defining a basic range for the target candidate area and reducing interference from irrelevant information. Further segmentation using grayscale features, through setting grayscale thresholds for secondary screening of candidate areas, overcomes the limitations of single color features, effectively eliminating false targets with similar colors but inconsistent grayscale, thus improving the accuracy of area segmentation. Edge contour optimization corrects candidate areas based on geometric standardization results, making the target contour more closely match the actual shape of the sample and avoiding contour shifts caused by factors such as shooting angle. Interference area removal accurately removes interference areas that are too small or have abnormal shapes by calculating the area and shape factor of connected regions, ensuring the integrity and purity of the target area. Separation result verification employs a dual standard of visual inspection and pixel proportion analysis. This involves both visually controlling edge deviations and using quantitative indicators to ensure reasonable area proportions. A backtracking adjustment mechanism forms a closed loop when standards are not met, significantly reducing separation errors and providing high-quality separation results for subsequent detection. The table below shows the separation of the target object from the background or other areas:

[0123]

[0124]

[0125] Extract feature information reflecting the raw material grade from the separated stone-plastic flooring images, including:

[0126] The core indicators reflecting the characteristic information of raw material grades are defect characteristics, color characteristics, texture characteristics, and morphological detail characteristics;

[0127] Defect features, color features, texture features, and morphological detail features are extracted respectively;

[0128] Among them, the extraction of defect features is as follows: by analyzing the gray value, color difference or texture change of pixels in the image, the region that is significantly different from the normal region is identified, the potential defects are initially located, and the attribute features of each located defect are extracted, including size features, morphological features and distribution features.

[0129] The color feature extraction is as follows: perform overall color analysis on the separated stone plastic floor image, extract the main color tone, perform consistency analysis on the main color tone, determine whether there are isolated color areas that differ from the main color tone based on the analysis results, and record the area and position of the isolated color areas.

[0130] The extraction of texture features involves dividing the separated stone plastic floor image into several local texture analysis regions, and analyzing the texture direction, texture density, and texture consistency of each texture analysis region.

[0131] The extraction of morphological details involves: performing edge regularity analysis on the image contour of the separated stone plastic flooring image. Edge regularity analysis detects whether there are irregular shapes such as burrs, depressions or protrusions on the edge of the image contour, counts the number of abnormal points on the edge that deviate from the smooth curve and the degree of deviation. At the same time, for the image taken from the side, the contour line of the side edge is extracted, the thickness value at different positions is measured, and the maximum deviation value of the thickness and the overall mean deviation are confirmed.

[0132] The extracted defect features, color features, texture features, and morphological detail features are summarized to form an initial feature dataset, which ultimately yields feature information reflecting the grade of the raw materials.

[0133] Specifically, defect feature extraction combines grayscale, color, and texture differences to locate potential defects and refines size, shape, and distribution features, accurately identifying the impact of various flaws on grade assessment. Color feature analysis identifies the consistency of the main color tone and isolated color areas, reflecting the uniformity of raw material coloring and avoiding misjudgments caused by local color variations. Texture features analyze direction, density, and consistency by segmenting local areas, taking into account both overall and local texture states, capturing the uniformity of the material structure. Morphological detail features detect both the regularity of contour edges and measure side thickness deviations, comprehensively reflecting the material processing precision. Finally, these features are aggregated to form an initial feature dataset, systematizing scattered feature information and providing structured data support for subsequent grade assessment. Each feature extraction stage is designed to address the core needs of raw material grade determination, ensuring feature relevance while enhancing information reliability through multi-dimensional cross-validation, effectively guaranteeing the accuracy of grade assessment.

[0134] The hierarchical influence features are filtered from the feature information, including:

[0135] Standardize the description of feature information and unify the measurement method of features;

[0136] After the measurement method is standardized, each feature information is correlated with the grade label, where the grade label is the manual grading result of the introduced samples;

[0137] Determine the distribution differences of each feature based on the correlation analysis results;

[0138] Redundant features are identified based on the distribution differences of each feature information. Redundant feature identification involves analyzing the intrinsic relationship between features and identifying highly redundant features. Specifically, if two features describe the same quality attribute in physical meaning or their numerical change trends are completely consistent, they are determined to be redundant features.

[0139] The redundancy features are evaluated for stability. If the value fluctuation of a redundancy feature is less than 5% in repeated tests, it is considered stable; if the fluctuation exceeds 15%, it is considered unstable and is removed.

[0140] The results of distribution difference, redundancy feature identification, and stability verification are prioritized and ranked. The prioritization is based on a comprehensive score of the results of distribution difference, redundancy feature identification, and stability verification, and the results are ranked from highest to lowest score.

[0141] The top 80% of the features in the ranking results are selected as candidate key features, and these candidate key features are used as hierarchical influence features in the feature information.

[0142] Specifically, the features are first standardized in description and measurement, eliminating differences in measurement between different features and laying a consistent basis for comparison in subsequent analysis, avoiding screening bias caused by inconsistent measurement methods. Feature correlation analysis is then performed with manual grading results, anchoring feature selection to actual grading standards and ensuring that selected features are directly related to grading determination, enhancing the practical value of features. Through distribution difference judgment and redundant feature identification, features with high grading discrimination are retained while redundant information with repetitive descriptions or consistent trends is eliminated, reducing interference from data redundancy and improving screening efficiency. The stability verification stage uses fluctuation thresholds to determine feature stability, eliminating unstable features with large fluctuations, ensuring the reliability of retained features, and reducing the error risk of subsequent grading models. Priority ranking and multi-dimensional scoring avoid the one-sidedness of a single indicator, making the screening results more comprehensive. The top 80% of features are selected as candidates, controlling the number of features while retaining key information, balancing feature comprehensiveness and model efficiency. The entire process forms a closed loop from standardization to verification, ensuring that the selected grading-influencing features are relevant, stable, and representative, providing a high-quality feature foundation for accurate grading.

[0143] To address the issues of insufficient data standardization and low efficiency due to reliance on manual labor in existing stone plastic flooring grading technologies, please refer to... Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0144] A sample dataset was constructed based on the selected hierarchical influence characteristics and the raw materials of SPC flooring, including:

[0145] The basic information of the raw material samples of stone plastic flooring is summarized, including sample number, production batch, original grade label and the path of the collected multi-angle images, and a sample information summary table is established.

[0146] For each stone plastic flooring raw material sample, match its corresponding graded influence characteristic data, and obtain the sample characteristic data of the stone plastic flooring raw material sample after the matching is completed;

[0147] Based on the results of manual grading, the sample feature data are labeled with grade labels, and labeling rules are established to clarify the feature thresholds corresponding to each grade.

[0148] The sample feature data with completed grade labels are divided into training set, validation set and test set. The quality of the divided dataset is verified. The quality verification is to verify whether the feature value of each sample feature data is within a reasonable range and whether the grade label is consistent with the actual situation of the sample feature data.

[0149] The validated datasets are stored as structured files for training, validation, and test sets respectively. At the same time, a dataset description document is generated, which records the data source, partition ratio, feature meaning, and processing method.

[0150] Finally, the sample dataset was constructed.

[0151] Specifically, the process begins by summarizing basic sample information and creating a master table. This integrates key information such as sample number and production batch, ensuring traceability throughout the sample's lifecycle. This facilitates subsequent tracking of data sources and sample backgrounds, supporting the reliability of the analysis results. When matching grading influence feature data, a one-to-one correspondence mechanism ensures accurate association of feature data for each sample, preventing feature mismatches and guaranteeing dataset consistency. Combining manual grading results with grade labeling and establishing threshold rules standardizes and normalizes the labels, reducing subjective labeling differences and improving label reliability. The design of dividing the training, validation, and test sets aligns with the scientific logic of model training, effectively evaluating the model's generalization ability and avoiding overfitting. The quality verification process employs double checks (feature value rationality and label consistency) to eliminate outlier data, further ensuring dataset quality. Structured storage and the generation of explanatory documentation ensure uniform data format and information transparency, facilitating subsequent access and sharing, and lowering the barrier to data use. The overall process forms a closed loop from information aggregation to storage and archiving, ensuring the dataset's completeness, accuracy, and operability, laying a solid foundation for the training and optimization of the raw material grading assessment model.

[0152] The model is trained on the constructed sample dataset, and a classification judgment is made based on the trained model, including:

[0153] The model type is selected based on the size of the sample dataset, the type of features, and the requirements of the hierarchical task, including machine learning models or deep learning models.

[0154] The training parameters are set according to the model type. The training parameters include basic parameters and regularization parameters. Then, the training parameters are tuned. The parameter tuning is carried out by testing different combinations within the preset parameter range through grid search or random search methods. The optimal parameter combination is selected with the hierarchical accuracy of the validation set as the indicator.

[0155] After setting the training parameters, the model is trained using the training set data, and the training process is monitored in real time.

[0156] The trained model is then evaluated using a separate test set, including hierarchical accuracy evaluation, category precision and recall evaluation, and confusion matrix analysis.

[0157] The model after performance evaluation is classified and judged, in which the feature data of the sample to be tested is extracted first;

[0158] The extracted feature data of the sample to be detected is input into the model after the performance evaluation is completed for model prediction. The model outputs the predicted level label based on the learned features and hierarchical association rules, and records the confidence of the model in the prediction result.

[0159] The model prediction results are then validated, including confidence and consistency checks.

[0160] The model prediction data that passed the verification in the verification results are associated and stored with the verification prediction level and sample information, and a classification judgment result is generated.

[0161] Finally, the model training and classification judgment of the sample dataset are completed.

[0162] Specifically, the model type is selected based on the dataset size, feature type, and task requirements to ensure the model is adapted to the real-world scenario and avoid performance deviations caused by inappropriate model selection. Parameter tuning employs grid search or random search methods, using validation set accuracy as an indicator to select the optimal parameter combination. This ensures the scientific nature of parameter settings while maximizing model performance and reducing errors from human experience. The training process combines real-time monitoring with independent test set evaluation. Multi-dimensional indicators such as classification accuracy, precision, recall, and confusion matrix are used to comprehensively test the model's generalization ability and classification stability, effectively avoiding overfitting and ensuring stable model performance in practical applications. Confidence and consistency checks are introduced in the classification judgment stage to filter out unreliable predictions with low confidence and eliminate abnormal results through consistency checks, further improving the accuracy of classification judgments. Finally, the prediction results are associated and stored with sample information to achieve traceability, facilitating subsequent analysis and traceability. The entire process forms a complete closed loop from model selection to result verification, balancing scientific rigor and practicality, providing an efficient and reliable solution for the accurate classification of SPC flooring raw materials.

[0163] The results of the grading judgment are analyzed and output, and a feedback mechanism is provided, including:

[0164] First, the results of the classification judgment are analyzed, including basic statistical analysis, misclassification analysis, and feature impact analysis.

[0165] The analysis results are then visualized, including a grade distribution bar chart, a misclassification heatmap, and a feature correlation scatter plot, while generating production monitoring reports, model performance reports, and raw data reports.

[0166] The generated report is sent to the corresponding display terminal for data display.

[0167] Staff members view the data on the display terminal, perform manual verification, record the verification results, and generate a structured feedback form based on the verification results.

[0168] The structured feedback table is used for classification and priority sorting, including feature recognition class, grade standard class and model stability class;

[0169] Based on the classification and priority sorting results, the tasks are assigned to the corresponding staff for processing and execution.

[0170] And record the processing results.

[0171] Specifically, the analysis phase encompasses basic statistics, error classification, and feature impact analysis, comprehensively covering the overall distribution of results, causes of anomalies, and key influencing factors. It grasps both macro trends and delves into micro-level issues, providing precise direction for subsequent optimization. Visual presentation utilizes diverse charts and categorized reports, transforming complex data into intuitive information for quick understanding by personnel in different roles. Production monitoring reports assist frontline management, model performance reports support technical optimization, and information transmission efficiency is improved. The feedback mechanism introduces manual review and structured feedback forms, reducing automation errors through human-machine collaborative verification. The structured design ensures standardized and consistent feedback information, laying the foundation for problem classification. Categorized processing and prioritization accurately allocate tasks according to problem type, avoiding chaotic processing flows and improving problem-solving efficiency. Clear division of responsibilities allows professionals to focus on their respective areas, accelerating rectification execution. The entire process of recording processing results forms a complete closed loop, facilitating the tracking of problem-solving trajectories and accumulating data for system iteration, driving continuous optimization of the hierarchical model and judgment criteria. The overall solution balances analytical depth and execution efficiency, enabling full-process empowerment from result interpretation to continuous improvement, enhancing the reliability and adaptability of the hierarchical system.

[0172] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0173] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A computer vision-based method for grading raw materials for stone-plastic flooring, characterized in that, include: First, prepare the raw materials for the stone-plastic flooring to be tested; Visual image capture of the prepared stone plastic flooring raw materials; Preprocess the captured images of the stone plastic flooring; Separate the target object from the background or other areas in the pre-processed stone plastic flooring image; Extract feature information reflecting the raw material grade from the separated stone-plastic flooring images; The process involves: filtering the grading impact features from the feature information; constructing a sample dataset based on the filtered grading impact features and the raw materials of SPC flooring; training a model on the constructed sample dataset and making grading judgments based on the trained model; analyzing and outputting the grading judgment results and providing a feedback mechanism. Extract feature information reflecting the raw material grade from the separated stone-plastic flooring images, including: The core indicators reflecting the characteristic information of raw material grades are defect characteristics, color characteristics, texture characteristics, and morphological detail characteristics; Defect features, color features, texture features, and morphological detail features are extracted respectively; Among them, the extraction of defect features is as follows: by analyzing the gray value, color difference or texture change of pixels in the image, the region that is significantly different from the normal region is identified, the potential defects are initially located, and the attribute features of each located defect are extracted, including size features, morphological features and distribution features. The color feature extraction is as follows: perform overall color analysis on the separated stone plastic floor image, extract the main color tone, perform consistency analysis on the main color tone, determine whether there are isolated color areas that differ from the main color tone based on the analysis results, and record the area and position of the isolated color areas. The extraction of texture features involves dividing the separated stone plastic floor image into several local texture analysis regions, and analyzing the texture direction, texture density, and texture consistency of each texture analysis region. The extraction of morphological details involves: performing edge regularity analysis on the image contour of the separated stone plastic flooring image. Edge regularity analysis detects whether there are burrs, depressions or protrusions on the edge of the image contour. It also counts the number and degree of deviation of abnormal points that deviate from the smooth curve on the edge. At the same time, for the image taken from the side, the contour line of the side edge is extracted, the thickness value at different positions is measured, and the maximum deviation value and the overall mean deviation of the thickness are confirmed. The extracted defect features, color features, texture features, and morphological detail features are summarized to form an initial feature dataset, which ultimately yields feature information reflecting the grade of the raw materials. The hierarchical influence features are filtered from the feature information, including: Standardize the description of feature information and unify the measurement method of features; After the measurement method is unified, each feature information is correlated with the level label, where the level label is the manual classification result of the introduced samples; Determine the distribution differences of each feature based on the correlation analysis results; Redundant features are identified based on the distribution differences of each feature information. Redundant feature identification involves analyzing the intrinsic relationship between features and identifying highly redundant features. Specifically, if two features describe the same quality attribute in physical meaning or their numerical change trends are completely consistent, they are determined to be redundant features. The redundancy features are evaluated for stability. If the value fluctuation of a redundancy feature is less than 5% in repeated tests, it is considered stable; if the fluctuation exceeds 15%, it is considered unstable and is removed. The results of distribution difference, redundancy feature identification, and stability verification are prioritized and ranked. The prioritization is based on a comprehensive score of the results of distribution difference, redundancy feature identification, and stability verification, and the results are ranked from highest to lowest score. The top 80% of the features in the ranking results are selected as candidate key features, and these candidate key features are used as hierarchical influence features in the feature information.

2. The method for grading raw materials of stone-plastic flooring based on computer vision according to claim 1, characterized in that, Prepare the raw materials for the stone-plastic flooring to be tested, including: First, samples of the raw materials of the stone plastic flooring to be tested are extracted. The raw materials of the stone plastic flooring to be tested include samples from different production periods, different raw material ratios, and different appearance states. The extracted samples are cleaned and then air-dried or dried with cold air. The samples that are air-dried or dried with cold air are subjected to morphological pretreatment. Morphological pretreatment involves cutting or trimming the selected samples into uniform size specifications according to actual testing requirements. After the morphological pretreatment is completed, the final raw material of the stone plastic flooring to be tested is obtained.

3. The method for grading raw materials of stone-plastic flooring based on computer vision according to claim 2, characterized in that, Visual image capture of the prepared stone plastic flooring raw materials was performed, including: Before visual image capture of the raw materials for stone plastic flooring, the image capture environment must first be set up; The shooting area is set up in a closed or semi-closed space and an adjustable brightness shadowless light source system is used. At the same time, a solid color non-reflective background board is laid on the shooting table. The background board is made of matte material. After the image capture environment is set up, the shooting equipment is debugged and its parameters are set. First, the industrial camera is fixed on a tripod, and the vertical distance between the camera lens and the inspection surface of the stone plastic flooring raw material is adjusted. Then, the industrial camera is connected to the computer, and the parameters are set through image acquisition software. After the shooting equipment is debugged and the parameters are set, the stone plastic flooring raw material is placed in the center of the background board for image acquisition; The image acquisition process involves: starting the image acquisition software, taking pictures of the raw materials of the stone plastic flooring according to the preset parameters, obtaining a frontal image of the raw materials of the stone plastic flooring after taking the pictures, adjusting the angle of the industrial camera, taking pictures from the 45-degree angle and the side of the raw materials of the stone plastic flooring respectively, and acquiring at least 2 images from each angle. Finally, the original visual images of the raw materials for stone plastic flooring were captured.

4. The computer vision-based method for grading raw materials for stone-plastic flooring according to claim 3, characterized in that, The captured images of the stone plastic flooring underwent image preprocessing, including: Image preprocessing is performed on the captured images of the raw materials for stone plastic flooring; The image preprocessing workflow is as follows: S1: Noise Reduction Processing: A multi-level filtering method is used to process the sensor noise and ambient light spots in the captured image of the raw material of the stone plastic flooring. S2: Contrast Correction: Based on the image grayscale histogram analysis, the dynamic range of overly dark or overly bright images in the denoised raw material image of the stone plastic flooring is adjusted. S3: Geometric normalization processing: The actual outline of the stone plastic flooring raw material image after contrast correction is located by edge detection, the redundant area of ​​the background board in the image is cropped, and only the complete area of ​​the stone plastic flooring sample is retained. The angle of the tilted part in the stone plastic flooring raw material image is corrected according to the horizontal reference line of the edge of the stone plastic flooring raw material sample. S4: Color Space Unification: Convert the colors in the image of the stone plastic flooring raw material after geometric normalization to the HSV color space. The HSV color space includes a luminance channel and a chrominance channel. Then, perform grayscale value processing on the luminance channel and the chrominance channel respectively. S5: Quality Inspection: The image of the raw material of the stone plastic flooring with unified color space is evaluated for clarity. The unqualified parts in the evaluation results are pre-processed again until the clarity evaluation is qualified. The final step is to complete the preprocessing of the captured stone plastic flooring images.

5. The computer vision-based method for grading raw materials for stone-plastic flooring according to claim 4, characterized in that, Separating the target object from the background or other areas in the preprocessed stone plastic flooring image includes: The process for separating images of stone-plastic flooring is as follows: First, initial segmentation is performed based on color features. The standard color range of the background board is extracted and a color threshold is set. Then, pixels in the stone plastic flooring image that meet the background features are marked as background areas, and the remaining pixels are classified as target candidate areas. Then, the grayscale features are combined for segmentation. First, a grayscale threshold is set. After the grayscale value is set, pixels in the target candidate region that are lower than the target grayscale lower limit are reclassified as background. Edge contour optimization is performed on the image after grayscale value segmentation, wherein the target candidate region is corrected based on the geometric normalization process. After the edge contour optimization is completed, interference regions are removed. This involves confirming the area and shape factor of each connected region, setting a reasonable size range for the target object, and identifying regions with too small an area or shapes that do not conform to the geometric characteristics of the sample as interference, marking them as background and removing them from the target region. After the interference area is removed, the separation results are verified. The separation results verification includes visual inspection and pixel ratio analysis. The visual inspection is to confirm that the target object covers the stone plastic floor sample and that there is no background residue or missing target area. The deviation between the edge and the actual boundary of the stone plastic floor sample is controlled within 2 pixels. The pixel ratio analysis is to count the ratio of the number of pixels in the target area to the total number of pixels in the image. The deviation between the sample size and the image size is less than 3%. If it is greater than 3%, the grayscale feature segmentation is re-performed. After the separation results are verified, an image of the separated stone-plastic flooring is obtained.

6. The computer vision-based method for grading raw materials for stone-plastic flooring according to claim 5, characterized in that, A sample dataset was constructed based on the selected hierarchical influence characteristics and the raw materials of SPC flooring, including: The basic information of the raw material samples of stone plastic flooring is summarized, including sample number, production batch, original grade label and the path of the collected multi-angle images, and a sample information summary table is established. For each stone plastic flooring raw material sample, match its corresponding graded influence characteristic data, and obtain the sample characteristic data of the stone plastic flooring raw material sample after the matching is completed; Based on the results of manual grading, the sample feature data are labeled with grade labels, and labeling rules are established to clarify the feature thresholds corresponding to each grade. The sample feature data with completed grade labels are divided into training set, validation set and test set. The quality of the divided dataset is verified. The quality verification is to verify whether the feature value of each sample feature data is within a reasonable range and whether the grade label is consistent with the actual situation of the sample feature data. The validated datasets are stored as structured files for training, validation, and test sets respectively. At the same time, a dataset description document is generated, which records the data source, partition ratio, feature meaning, and processing method. Finally, the sample dataset was constructed.

7. The computer vision-based method for grading raw materials for stone-plastic flooring according to claim 6, characterized in that, The model is trained on the constructed sample dataset, and a classification judgment is made based on the trained model, including: The model type is selected based on the size of the sample dataset, the type of features, and the requirements of the hierarchical task, including machine learning models or deep learning models. The training parameters are set according to the model type. The training parameters include basic parameters and regularization parameters. Then, the training parameters are tuned. The parameter tuning is carried out by testing different combinations within the preset parameter range through grid search or random search methods. The optimal parameter combination is selected with the hierarchical accuracy of the validation set as the indicator. After setting the training parameters, the model is trained using the training set data, and the training process is monitored in real time. The trained model is then evaluated using a separate test set, including hierarchical accuracy evaluation, category precision and recall evaluation, and confusion matrix analysis. The model after performance evaluation is classified and judged, in which the feature data of the sample to be tested is extracted first; The extracted feature data of the sample to be detected is input into the model after the performance evaluation is completed for model prediction. The model outputs the predicted level label through the learned features and hierarchical association rules, and records the model's confidence in the prediction results. The model prediction results are then validated, including confidence and consistency checks. The model prediction data that passed the verification in the verification results are associated and stored with the verification prediction level and sample information, and a classification judgment result is generated. Finally, the model training and classification judgment of the sample dataset are completed.

8. The method for grading raw materials of stone-plastic flooring based on computer vision according to claim 7, characterized in that, The results of the grading judgment are analyzed and output, and a feedback mechanism is provided, including: First, the results of the classification judgment are analyzed, including basic statistical analysis, misclassification analysis, and feature impact analysis. The analysis results are then visualized, including a grade distribution bar chart, a misclassification heatmap, and a feature correlation scatter plot, while generating production monitoring reports, model performance reports, and raw data reports. The generated report is sent to the corresponding display terminal for data display. Staff members view the data on the display terminal, perform manual verification, record the verification results, and generate a structured feedback form based on the verification results. The structured feedback table is used for classification and priority sorting, including feature recognition class, grade standard class and model stability class; Based on the classification and priority sorting results, the tasks are assigned to the corresponding staff for processing and execution. And record the processing results.

Images