Ceramic defect intelligent detection method, system and equipment based on deep learning and storage medium

Abstract

The invention provides a ceramic defect intelligent detection method, system and device based on deep learning and a storage medium. The method comprises the steps that a ceramic surface image is collected and preprocessed; performing feature extraction and classification on the preprocessed ceramic surface image by using a deep convolutional neural network to obtain a preliminary classification result; according to the preliminary classification result, performing defect positioning on the ceramic surface image by using a multi-scale feature fusion technology to generate a defect candidate area; performing secondary classification on the defect candidate area by applying a support vector machine to obtain a secondary classification result; and outputting a ceramic defect detection report according to the secondary classification result. According to the technical scheme provided by the invention, the efficiency, accuracy and robustness of ceramic defect detection are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of intelligent detection technology, and in particular to a method, system, device and storage medium for intelligent detection of ceramic defects based on deep learning. Background Art

[0002] With the development of the ceramic industry, the quality control of ceramic products has become particularly important. During the manufacturing process, various defects may occur on the ceramic surface, such as cracks, pores, color spots, etc. These defects directly affect the performance and appearance of the product.

[0003] There are already some ceramic defect detection methods based on machine vision on the market, which mainly use a combination of traditional image processing technology and machine learning methods. These solutions usually include steps such as image acquisition, preprocessing, feature extraction and classification.

[0004] However, these methods often rely on manually designed features and find it difficult to capture the detailed features of complex ceramic surfaces, which affects the detection accuracy. When processing a large number of ceramic products, it often takes a long time to complete the detection process, resulting in poor detection efficiency. Summary of the invention

[0005] The embodiments of the present application provide a method, system, device and storage medium for intelligent detection of ceramic defects based on deep learning, so as to solve the problems of low detection accuracy and poor detection efficiency in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a ceramic defect intelligent detection method based on deep learning, comprising:

[0007] Collect ceramic surface images and perform preprocessing;

[0008] Using a deep convolutional neural network to extract features and classify the preprocessed ceramic surface image to obtain a preliminary classification result;

[0009] Based on the preliminary classification results, multi-scale feature fusion technology is used to locate defects in the ceramic surface image to generate defect candidate areas;

[0010] Applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result;

[0011] A ceramic defect detection report is outputted according to the secondary classification result.

[0012] Optionally, the extracting and classifying the preprocessed ceramic surface image using a deep convolutional neural network includes:

[0013] Using a pre-trained deep convolutional neural network model as a feature extractor to extract multi-level features of the ceramic surface image;

[0014] The multi-level features are input into the fully connected layer for classification to obtain the preliminary classification result.

[0015] Optionally, the method of locating defects on the ceramic surface image using a multi-scale feature fusion technology according to the preliminary classification result includes:

[0016] Performing fusion processing on the feature maps of different scales in the preliminary classification results to enhance the detection capability of small-size defects;

[0017] The target detection algorithm is applied to analyze the fused feature map to generate the defect candidate area.

[0018] Optionally, applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result includes:

[0019] Calculating texture features and shape features within the defect candidate area;

[0020] Based on the texture features and the shape features, a support vector machine is used for training and classification to obtain the secondary classification result.

[0021] Optionally, calculating the texture features in the defect candidate area includes:

[0022] Obtaining the gray level co-occurrence matrix G(i,j|d,θ) within the defect candidate area;

[0023] The contrast, homogeneity and entropy of the gray level co-occurrence matrix are calculated by the following formula group;

[0024]

[0025]

[0026] Among them, G(i,j|d,θ) represents the gray level co-occurrence matrix, C represents contrast, H represents homogeneity, and E represents entropy, which supplements the meaning of each parameter;

[0027] Calculating texture features in the defect candidate region according to the gray level co-occurrence matrix and the contrast, homogeneity, and entropy of the gray level co-occurrence matrix;

[0028] Optionally, calculating the shape features in the defect candidate area includes:

[0029] The shape factor within the defect candidate area is calculated by the following formula group;

[0030]

[0031] Where S represents the shape factor, λ1 is the maximum eigenvalue in the main axis direction of the defect candidate area, and λ2 is the second largest eigenvalue in the main axis direction of the defect candidate area;

[0032] The shape feature in the defect candidate area is calculated according to the shape factor.

[0033] Optionally, calculating the texture features in the defect candidate area further includes:

[0034] The edge strength EI of the defect candidate area is calculated, wherein the edge strength is calculated by the following formula:

[0035]

[0036] Among them, I x (x,y) and I y (x,y) represents the first-order derivative in the x direction and y direction respectively;

[0037] Calculate the edge direction variation EDV of the defect candidate area, where:

[0038]

[0039] Where N represents the total number of pixels, I x (x,y) and I y (x,y) represents the first-order derivative in the x direction and y direction respectively;

[0040] The texture features in the defect candidate region are calculated according to the gray level co-occurrence matrix and the contrast, homogeneity, and entropy of the gray level co-occurrence matrix, as well as the edge strength and edge direction changes of the defect candidate region.

[0041] In a second aspect, the embodiment of the present application provides a ceramic defect intelligent detection system based on deep learning, comprising:

[0042] An image acquisition module is used to acquire ceramic surface images and perform preprocessing;

[0043] A feature extraction module, used to extract and classify the preprocessed ceramic surface image using a deep convolutional neural network to obtain a preliminary classification result;

[0044] A defect localization module, used to locate defects on the ceramic surface image based on the preliminary classification result by using multi-scale feature fusion technology to generate defect candidate areas;

[0045] A classification module, applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result;

[0046] An output module outputs a ceramic defect detection report according to the secondary classification result.

[0047] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the deep learning-based intelligent ceramic defect detection methods described in the first aspect.

[0048] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the deep learning-based intelligent detection method for ceramic defects as described in any one of the first aspects.

[0049] In the embodiment of the present application, a ceramic surface image is collected and preprocessed; a deep convolutional neural network is used to extract features and classify the preprocessed ceramic surface image to obtain a preliminary classification result; based on the preliminary classification result, a multi-scale feature fusion technology is used to locate defects in the ceramic surface image to generate a defect candidate area; a support vector machine is used to perform secondary classification on the defect candidate area to obtain the secondary classification result; and a ceramic defect detection report is output based on the secondary classification result. The technical solution provided by the present application improves the efficiency, accuracy and robustness of ceramic defect detection.

[0050] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 A flowchart of a ceramic defect intelligent detection method based on deep learning provided in an embodiment of the present application;

[0053] Figure 2 A schematic diagram of the structure of a ceramic defect intelligent detection system based on deep learning provided in an embodiment of the present application;

[0054] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0056] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0057] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0058] Figure 1 A flowchart of a ceramic defect intelligent detection method based on deep learning is provided for the embodiment of the present application, such as Figure 1 As shown, the method includes:

[0059] 101. Collect ceramic surface images and perform preprocessing;

[0060] In this step, you first need to capture an image of the ceramic surface, which is usually done through an industrial camera or other image acquisition device. The captured original image may contain noise, uneven lighting, and other problems, so it needs to be preprocessed to improve the accuracy and efficiency of subsequent processing. Preprocessing may include grayscale, denoising, brightness and contrast adjustment, image enhancement, and other operations.

[0061] In the embodiment of the present application, the ceramic surface image is collected and preprocessed in the following manner:

[0062] Image acquisition:

[0063] An industrial-grade CCD (Charge-Coupled Device, CCD) camera was used to capture surface images of ceramic samples from multiple angles to ensure that the images covered all critical areas.

[0064] Determine the appropriate shooting distance and angle based on the size and shape of the ceramic to ensure image quality.

[0065] Image preprocessing:

[0066] Grayscale: Convert a color image to a grayscale image, reducing the amount of data while retaining the main information of the image.

[0067] The following are examples:

[0068] If the original image has a resolution of 1024x768 pixels and uses the RGB color space, it is converted to a single-channel grayscale image.

[0069] Denoising: Use Gaussian filtering to remove random noise in the image.

[0070] Example: A grayscale image is filtered using a 3x3 Gaussian kernel with a standard deviation of 1.0.

[0071] Brightness and Contrast Adjustment: Adjust the overall brightness and contrast of the image to improve the quality of the image.

[0072] Example: Increase the brightness of an image by 20% and boost the contrast by 15%.

[0073] Image enhancement: Further improve the contrast of the image through techniques such as histogram equalization.

[0074] Embodiment: Apply the histogram equalization technology to enhance the image and improve the visibility of image details.

[0075] Through the above steps, a high-quality grayscale image of the ceramic surface can be obtained, which is ready for subsequent feature extraction and defect detection. For example, after preprocessing, the original color image is converted into a clear grayscale image, in which the details of the ceramic surface are more prominent, which is helpful for subsequent feature extraction and classification.

[0076] 102. Using a deep convolutional neural network to perform feature extraction and classification on the preprocessed ceramic surface image to obtain a preliminary classification result;

[0077] Optionally, the use of a deep convolutional neural network in step 102 to extract features and classify the preprocessed ceramic surface image includes: using a pretrained deep convolutional neural network model as a feature extractor to extract multi-level features of the ceramic surface image; inputting the multi-level features into a fully connected layer for classification to obtain the preliminary classification result.

[0078] In this step, we use the deep convolutional neural network (DCNN) to extract features and classify the preprocessed ceramic surface images. The deep convolutional neural network is a multi-layer neural network structure that can automatically learn complex features in images and perform efficient classification tasks. In this process, we use a pre-trained model to speed up the feature extraction process and use a fully connected layer to make the final classification decision.

[0079] Deep Convolutional Neural Network (DCNN): A multi-layer neural network structure that is specifically designed to process input data with a grid structure, such as images. It extracts local features and abstract features of images through structures such as convolutional layers and pooling layers.

[0080] Pre-trained models: Models that have been pre-trained using large datasets can be used as feature extractors in new tasks without having to be trained from scratch, thus saving time and computing resources.

[0081] Fully Connected Layer (FC layer): One or more layers in a neural network where each neuron is connected to all neurons in the previous layer. Fully connected layers are often used for classification tasks to map extracted features to classification labels.

[0082] In the embodiment of the present application, it is assumed that we have preprocessed a ceramic surface image. Next, we will use the pre-trained deep convolutional neural network model to extract image features and perform classification.

[0083] Selection of pre-trained model: VGG16 (Visual Geometry Group, VGG) model is selected as the pre-trained model. VGG16 is a deep convolutional neural network model widely used in image recognition tasks and has good feature extraction capabilities.

[0084] Feature extraction:

[0085] Model structure adjustment: remove the last classification layer (i.e. the last fully connected layer) of the VGG16 model, and retain the previous convolutional layer and pooling layer to form a feature extractor.

[0086] Feature extraction: The preprocessed ceramic surface image is input into the adjusted VGG16 model, and the model will automatically learn the multi-level features in the image, including edge, texture, color and other information.

[0087] Feature representation: Output a series of multi-level feature vectors that contain the key information of the ceramic surface image.

[0088] Classification:

[0089] Add fully connected layers: Add one or more fully connected layers after the feature extractor for classification.

[0090] Classification decision: The extracted multi-level features are input into the fully connected layer, and the classification decision is made through the trained weight parameters to output the preliminary classification results.

[0091] Assume that we are using the VGG16 model, which has 16 weight layers (including 13 convolutional layers and 3 fully connected layers) and has been pre-trained using the ImageNet dataset.

[0092] Input Image: Assume that the size of input image is 224x224 pixels.

[0093] Feature extraction: Remove the last fully connected layer of the VGG16 model and keep the previous convolutional layer and pooling layer, so that a feature extractor can be obtained. After inputting the image, a 4096-dimensional feature vector can be obtained.

[0094] Fully connected layer: Add a new fully connected layer after the feature extractor. Assuming that we need to classify three types of defects on the ceramic surface (no defects, cracks, bubbles), the number of output nodes of the fully connected layer is 3.

[0095] Classification: The output of the fully connected layer is normalized by the Softmax function to obtain the probability of each defect type, and the category with the highest probability is taken as the preliminary classification result.

[0096] Suppose we have a preprocessed ceramic surface image, and after passing through the VGG16 feature extractor, we get a 4096-dimensional feature vector. This feature vector is input into a fully connected layer with 3 output nodes, and the output vector obtained is [0.1, 0.8, 0.1], which means that the image is most likely to belong to the "crack" category (assuming that the category index 0 is no defect, 1 is crack, and 2 is bubble).

[0097] Through the above steps, we get the preliminary classification results of the ceramic surface image, which will be used for the next step of defect location and further classification optimization.

[0098] 103. Based on the preliminary classification result, use multi-scale feature fusion technology to locate defects in the ceramic surface image and generate defect candidate areas;

[0099] Optionally, the method in step 103 uses a multi-scale feature fusion technique to locate defects in the ceramic surface image based on the preliminary classification result, including: fusing feature maps of different scales in the preliminary classification result to enhance the detection capability of small-sized defects; and applying a target detection algorithm to analyze the fused feature map to generate the defect candidate area.

[0100] This step mainly involves two concepts: multi-scale feature fusion technology and target detection algorithm. Multi-scale feature fusion technology aims to enhance the detection ability of defects of different sizes, especially small defects, by combining feature maps at different levels. The target detection algorithm analyzes the fused feature map to locate potential defect areas and generate defect candidate areas.

[0101] Multi-scale feature fusion technology: In deep learning, feature maps at different layers can usually capture information at different scales. Shallow feature maps often contain more detailed information, while deep feature maps can capture more abstract features. Multi-scale feature fusion technology can simultaneously utilize information of different scales by fusing feature maps at different levels, thereby improving the detection capability of small-sized defects.

[0102] Object detection algorithms: These algorithms can locate and identify specific target objects in an image. Common object detection algorithms include but are not limited to the R-CNN series (such as FastR-CNN, Faster R-CNN), YOLO (You Only LookOnce), SSD (Single Shot Multi Box Detector), etc. These algorithms locate and classify objects by sliding windows on images and generating candidate regions.

[0103] In the embodiment of the present application, it is assumed that we have obtained the preliminary classification results of the ceramic surface image and hope to further accurately locate the defect position in the image. We can take the following steps:

[0104] Feature map fusion:

[0105] Feature map acquisition: Get feature maps from different layers of a deep convolutional neural network. For example, get feature maps from the conv3_3, conv4_3, and conv5_3 layers of VGG16.

[0106] Feature map fusion: Feature maps of different levels are fused together by upsampling or skip connection. For example, the feature map of the conv3_3 layer is upsampled to the same resolution as conv5_3, and then added to the feature map of the conv5_3 layer.

[0107] Object Detection:

[0108] Candidate region generation: Apply target detection algorithms, such as Faster R-CNN, to process the fused feature maps and generate a series of candidate regions.

[0109] Candidate region screening: Screen the generated candidate regions, for example, using non-maximum suppression (NMS) to remove overlapping candidate regions.

[0110] The following is the specific implementation:

[0111] Feature map fusion:

[0112] Feature map acquisition: Get the feature maps of conv3_3, conv4_3, and conv5_3 layers from the VGG16 model.

[0113] Feature map fusion: The feature map of the conv3_3 layer is upsampled to the same resolution as conv5_3, and then added element-by-element to the feature map of the conv5_3 layer. Similarly, the feature map of the conv4_3 layer is also upsampled to the same resolution and added to the fused feature map above to obtain the fused feature map.

[0114] Object Detection:

[0115] Candidate region generation: Using the Faster R-CNN algorithm, a series of candidate regions are first generated through the Region Proposal Network (RPN), which cover potential defect locations.

[0116] Candidate area screening: Screen the generated candidate areas, for example, use NMS to remove overlapping candidate areas and leave the most likely defect locations.

[0117] Assume that we have obtained preliminary classification results and extracted the feature maps of conv3_3, conv4_3, and conv5_3 layers from the VGG16 model. We follow the steps below:

[0118] Feature map fusion:

[0119] The resolution of the conv3_3 feature map is assumed to be 56x56, the resolution of the conv4_3 feature map is assumed to be 28x28, and the resolution of the conv5_3 feature map is assumed to be 14x14.

[0120] The conv3_3 feature map is upsampled to 28x28 and then added to the conv4_3 feature map.

[0121] The above fused feature map is then upsampled to 14x14 and added to the conv5_3 feature map to obtain the final fused feature map.

[0122] Object Detection:

[0123] Use the RPN of the Faster R-CNN algorithm to generate candidate regions. Suppose 100 candidate regions are generated.

[0124] NMS is used to screen out the 50 most likely candidate regions as defect candidate regions.

[0125] Through the above steps, we are able to generate a series of candidate regions that may contain defects, providing a basis for subsequent accurate classification.

[0126] 104. Apply a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result;

[0127] Optionally, the applying of a support vector machine to perform secondary classification on the defect candidate area in step 104 to obtain the secondary classification result includes: calculating texture features and shape features within the defect candidate area; and training and classifying using a support vector machine based on the texture features and the shape features to obtain the secondary classification result.

[0128] The step of calculating the texture features in the defect candidate area includes:

[0129] Obtaining a gray level co-occurrence matrix G(i, j|d, θ) within the defect candidate area;

[0130] The contrast, homogeneity and entropy of the gray level co-occurrence matrix are calculated by the following formula group;

[0131]

[0132] Among them, G(i, j|d, θ) represents the gray level co-occurrence matrix, C represents contrast, H represents homogeneity, and E represents entropy, which supplements the meaning of each parameter;

[0133] Calculating texture features in the defect candidate region according to the gray level co-occurrence matrix and the contrast, homogeneity, and entropy of the gray level co-occurrence matrix;

[0134] The step of calculating the shape features in the defect candidate area includes:

[0135] The shape factor within the defect candidate area is calculated by the following formula group;

[0136]

[0137] Where S represents the shape factor, λ1 is the maximum eigenvalue in the main axis direction of the defect candidate area, and λ2 is the second largest eigenvalue in the main axis direction of the defect candidate area;

[0138] The shape feature in the defect candidate area is calculated according to the shape factor.

[0139] The step of calculating the texture features in the defect candidate area further includes:

[0140] The edge strength EI of the defect candidate area is calculated, wherein the edge strength is calculated by the following formula:

[0141]

[0142] Among them, I x (x, y) and I y (x, y) represents the first-order derivative in the x-direction and y-direction respectively;

[0143] Calculate the edge direction variation EDV of the defect candidate area, where:

[0144]

[0145] Where N represents the total number of pixels, I x (x, y) and I y (x, y) represents the first-order derivative in the x-direction and y-direction respectively;

[0146] The texture features in the defect candidate region are calculated according to the gray level co-occurrence matrix and the contrast, homogeneity, and entropy of the gray level co-occurrence matrix, as well as the edge strength and edge direction changes of the defect candidate region.

[0147] This step mainly includes two parts: calculating the texture features and shape features in the defect candidate area, and using these features for secondary classification of the support vector machine (SVM). The key concepts involved here include gray-level co-occurrence matrix, contrast, homogeneity, entropy, edge strength, edge direction change, and support vector machine.

[0148] Gray-Level Co-occurrence Matrix (GLCM): It is a commonly used texture feature description method that records the spatial relationship between pixels of different gray levels in an image.

[0149] Contrast: It measures the difference in grayscale values ​​in local areas of an image.

[0150] Homogeneity: Measures the proximity of pixels with similar grayscale values.

[0151] Entropy: Measures the uncertainty of gray values ​​in the gray-level co-occurrence matrix.

[0152] Edge Intensity (EI): Measures the strength of edges in an image.

[0153] Edge Direction Variation (EDV): Measures the change in edge direction in an image.

[0154] Shape Factor (SF): measures the shape characteristics of the defect candidate area.

[0155] Support Vector Machine (SVM): A supervised learning model used for classification and regression analysis, especially for classifying data in high-dimensional space.

[0156] In the embodiment of the present application, assuming that we have obtained the defect candidate regions, we need to calculate the texture features and shape features in these candidate regions and perform secondary classification using a support vector machine. The following are the specific implementation steps:

[0157] Calculate texture features:

[0158] Construction of gray-level co-occurrence matrix: Given a defect candidate area, calculate the gray-level co-occurrence matrix G(i, j|d, θ), where i, j are gray levels, d is the distance, and θ is the angle.

[0159] Calculation of contrast, homogeneity and summation:

[0160] Contrast ratio C = ∑ i，j (ij) 2 G(i, j|d, θ);

[0161] Homogeneity

[0162] Entropy E = -∑ i，j G(i,j∣d,θ)logG(i,j∣d,θ);

[0163] Calculation of edge strength and edge direction changes:

[0164] Edge intensity EI = ∑ x，y |I x (x,y)|+|I y (x, y)|;

[0165] Edge direction change

[0166] Calculate the shape factor: Calculate the shape factor S based on the maximum eigenvalue λ1 and the second largest eigenvalue λ2 in the main axis direction of the candidate region.

[0167] in,

[0168] The following secondary classification is performed:

[0169] Support Vector Machine Training: Using texture features and shape features as input, a support vector machine is trained to distinguish between defect and non-defect regions.

[0170] Classification result output: Classify each defect candidate area according to the trained support vector machine model to obtain the final secondary classification result.

[0171] Assuming we have a defect candidate region with a size of 100×100 pixels, we follow the steps below:

[0172] Construction of gray-level co-occurrence matrix:

[0173] Assuming that the distance d = 1 and the angle θ = 0°, the gray-level co-occurrence matrix G(i, j|1, 0°) is constructed.

[0174] Calculation of contrast, homogeneity and summation:

[0175] Assume contrast C = 0.01, homogeneity H = 0.85, and entropy E = 0.23.

[0176] Calculation of edge strength and edge direction changes:

[0177] Assume that the edge intensity EI=500 and the edge direction variation EDV=0.12.

[0178] Calculate the shape factor:

[0179] Assume that the largest eigenvalue λ1=1200 and the second largest eigenvalue λ2=800.

[0180]

[0181] Support vector machine training and classification:

[0182] Assume that we use a pre-trained SVM model and the input features include contrast, homogeneity, entropy, edge strength, edge direction change, and shape factor.

[0183] Assume that after SVM classification, the defect candidate area is marked as a defect.

[0184] Through the above steps, we can effectively perform secondary classification on the defect candidate areas and improve the accuracy of detection.

[0185] 105. Output a ceramic defect detection report according to the secondary classification result.

[0186] In this step, a ceramic defect inspection report is generated based on the secondary classification results obtained previously. This report will summarize all the inspection results, indicate which areas have defects, and may include information such as the type, location, and size of the defects. In addition, the report may also provide an assessment of the severity of the defects and recommended follow-up measures.

[0187] In the embodiment of the present application, after completing the secondary classification of the defect candidate area, the system will generate a detailed ceramic defect detection report. The following is an embodiment of the specific steps:

[0188] Summarize the classification results: Collect all the regional information confirmed as defects through secondary classification, including the location, type (such as cracks, bubbles, etc.) and size of the defects.

[0189] Generate test report: Create a test report based on the summarized information. The report will include but not limited to the following:

[0190] Marking of defect locations: Mark each area confirmed as a defect on the original ceramic surface image.

[0191] Defect Type: Clearly indicate the type of each defect.

[0192] Defect Size: Provides metric information such as the area or diameter of each defect region.

[0193] Defect Severity Assessment: Assess the overall defect severity based on the number, type, and size of defects.

[0194] Recommended Actions: Based on the assessment results, recommendations are made as to whether further action or investigation is needed.

[0195] Assume that we have completed the inspection of the ceramic surface image and obtained the secondary classification results. Now we need to generate a test report based on these results. The following are the specific steps:

[0196] Summarize the classification results:

[0197] After secondary classification, we found that there are three defect areas on the ceramic surface, namely:

[0198] Defect 1: Crack, located in the upper left corner of the image, size 10mmx5mm.

[0199] Defect 2: Bubble, located slightly to the right of the center of the image, with a size of 3 mm.

[0200] Defect 3: Scratches, located in the lower right corner of the image, with a length of 15 mm.

[0201] Generate a test report:

[0202] Create a Word document or PDF file as the inspection report.

[0203] On the first page of the report, attach an image of the original ceramic surface and mark the three defect areas mentioned above in the image.

[0204] Detailed information about each defect is listed in the report, including type, location and size.

[0205] Provide a table or list that summarizes all defect information for quick review.

[0206] Based on the number and type of defects, rate the overall defect severity. For example, if there are only a few small defects, then the severity may be low.

[0207] Based on the evaluation results, recommended measures are given. For example, if there are many or large defects, it is recommended to repair or scrap the ceramic part.

[0208] The following is a sample report:

[0209] Ceramic defect inspection report

[0210] Testing date: August 30, 2024

[0211] Test results:

[0212] Defect 1: Crack location: upper left corner Size: 10mmx5mm

[0213] Defect 2: Bubble location: center right Size: 3mm

[0214] Defect 3: Scratch location: lower right corner Length: 15mm

[0215] Defect severity assessment:

[0216] Since three different types of defects were detected, although the individual defects were not large in size, considering the number and distribution, further inspection or repair is recommended.

[0217] Suggested actions:

[0218] It is recommended to conduct a detailed inspection of the ceramic parts and consider whether they need to be repaired according to the specific situation. If they cannot be repaired, it is recommended to scrap them.

[0219] Through the above steps, we can generate a detailed ceramic defect detection report to help manufacturers promptly detect and deal with defects in ceramic products.

[0220] Figure 2 A structural schematic diagram of a ceramic defect intelligent detection system based on deep learning is provided for an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0221] An image acquisition module 21 is used to acquire ceramic surface images and perform preprocessing;

[0222] A feature extraction module 22 is used to extract and classify the preprocessed ceramic surface image using a deep convolutional neural network to obtain a preliminary classification result;

[0223] A defect localization module 23 is used to locate defects on the ceramic surface image based on the preliminary classification result using a multi-scale feature fusion technique to generate a defect candidate area;

[0224] A classification module 24, applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result;

[0225] The output module 25 outputs a ceramic defect detection report according to the secondary classification result.

[0226] Figure 2 The ceramic defect intelligent detection system based on deep learning can perform Figure 1 The implementation principle and technical effect of the ceramic defect intelligent detection method based on deep learning described in the illustrated embodiment will not be repeated. The specific manner in which each module and unit performs operations in the ceramic defect intelligent detection system based on deep learning in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0227] In one possible design, Figure 2 The ceramic defect intelligent detection system based on deep learning of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0228] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0229] The processing component 32 is used to: collect ceramic surface images and perform preprocessing; use a deep convolutional neural network to extract features and classify the preprocessed ceramic surface images to obtain preliminary classification results; based on the preliminary classification results, use multi-scale feature fusion technology to locate defects in the ceramic surface images and generate defect candidate areas; use a support vector machine to perform secondary classification on the defect candidate areas to obtain the secondary classification results; and output a ceramic defect detection report based on the secondary classification results.

[0230] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0231] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0232] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0233] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0234] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0235] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0236] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a method for intelligent detection of ceramic defects based on deep learning.

[0237] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0238] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0239] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0240] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A ceramic defect intelligent detection method based on deep learning, characterized in that: include: Collect ceramic surface images and perform preprocessing; Using a deep convolutional neural network to extract features and classify the preprocessed ceramic surface image to obtain a preliminary classification result; Based on the preliminary classification results, multi-scale feature fusion technology is used to locate defects in the ceramic surface image to generate defect candidate areas; Applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result; A ceramic defect detection report is outputted according to the secondary classification result.

2. The method according to claim 1, characterized in that The method of using a deep convolutional neural network to extract features and classify the preprocessed ceramic surface image includes: Using a pre-trained deep convolutional neural network model as a feature extractor to extract multi-level features of the ceramic surface image; The multi-level features are input into the fully connected layer for classification to obtain the preliminary classification result.

3. The method according to claim 1, characterized in that The method of locating defects on the ceramic surface image using a multi-scale feature fusion technology based on the preliminary classification result includes: Performing fusion processing on the feature maps of different scales in the preliminary classification results to enhance the detection capability of small-size defects; The target detection algorithm is applied to analyze the fused feature map to generate the defect candidate area.

4. The method according to claim 1, characterized in that: The applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result includes: Calculating texture features and shape features within the defect candidate area; Based on the texture features and the shape features, a support vector machine is used for training and classification to obtain the secondary classification result.

5. The method according to claim 4, characterized in that Calculating the texture features in the defect candidate area includes: Obtaining the gray level co-occurrence matrix G(i,j|d,θ) within the defect candidate area; The contrast, homogeneity and entropy of the gray level co-occurrence matrix are calculated by the following formula group; Among them, G(i,j|d,θ) represents the gray level co-occurrence matrix, C represents contrast, H represents homogeneity, and E represents entropy, which supplements the meaning of each parameter; The texture features in the defect candidate region are calculated according to the gray level co-occurrence matrix and the contrast, homogeneity and entropy of the gray level co-occurrence matrix.

6. The method according to claim 4, characterized in that Calculating the shape features in the defect candidate area includes: The shape factor within the defect candidate area is calculated by the following formula group; Where S represents the shape factor, λ1 is the maximum eigenvalue in the main axis direction of the defect candidate area, and λ2 is the second largest eigenvalue in the main axis direction of the defect candidate area; The shape feature in the defect candidate area is calculated according to the shape factor.

7. The method according to claim 5, characterized in that Calculating the texture features in the defect candidate area also includes: The edge strength EI of the defect candidate area is calculated, wherein the edge strength is calculated by the following formula: Among them, I x (x,y) and I y (x,y) represents the first-order derivative in the x direction and y direction respectively; Calculate the edge direction variation EDV of the defect candidate area, where: Where N represents the total number of pixels, I x (x,y) and I y (x,y) represents the first-order derivative in the x direction and y direction respectively; The texture features in the defect candidate region are calculated according to the gray level co-occurrence matrix and the contrast, homogeneity, and entropy of the gray level co-occurrence matrix, as well as the edge strength and edge direction changes of the defect candidate region.

8. A ceramic defect intelligent detection system based on deep learning, characterized in that: include: An image acquisition module is used to acquire ceramic surface images and perform preprocessing; A feature extraction module, used to extract and classify the preprocessed ceramic surface image using a deep convolutional neural network to obtain a preliminary classification result; A defect localization module, used to locate defects on the ceramic surface image based on the preliminary classification result by using multi-scale feature fusion technology to generate defect candidate areas; A classification module, applying a support vector machine to perform secondary classification on the defect candidate area to obtain the secondary classification result; An output module outputs a ceramic defect detection report according to the secondary classification result.

9. An electronic device, comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the deep learning-based intelligent detection method for ceramic defects as described in any one of claims 1 to 7.

10. A computer storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the deep learning-based intelligent detection method for ceramic defects described in any one of claims 1 to 7 is implemented.

Images