Ladle air brick defect detection method and system based on image recognition

Abstract

The invention relates to the technical field of industrial detection, and provides a steel ladle air brick defect detection method and system based on image recognition, and the method comprises the steps: carrying out the preprocessing of a target steel ladle air brick image shot by an industrial camera, including edge feature extraction, cutting and enhancement, and obtaining a preprocessed image set; for the block corresponding to each preprocessed image, acquiring a multi-angle image set with different resolutions shot by a multi-angle sensor; inputting the multi-angle images into a defect feature extraction network to generate a feature image set, and integrating to obtain an integrated feature image; generating defect positioning and type information based on the integrated feature map and the defect identification model set; and finally, combining the initial image, the model set and the defect information to generate a defect detection result. According to the method, the defect detection precision is improved through multi-angle image fusion and model collaborative verification. According to the invention, the accuracy and reliability of steel ladle air brick defect detection can be improved, and the accuracy of defect positioning and classification is enhanced.

Description

Technical Field

[0001] This invention relates to the field of industrial testing technology, and more specifically, to a method and system for detecting defects in permeable bricks in steel ladles based on image recognition. Background Technology

[0002] In steel production, the quality inspection of permeable bricks in ladles is a crucial step in ensuring steel quality and production safety. Traditional methods for detecting defects in permeable bricks in ladles mainly rely on manual visual inspection or simple image analysis techniques. Manual visual inspection is not only inefficient but also susceptible to human error, leading to inconsistent results. Simple image analysis techniques typically only process images from a single perspective, making it difficult to comprehensively capture complex defect information on the surface of the permeable bricks, especially when dealing with multi-angle and multi-type defects, resulting in insufficient accuracy and reliability. Furthermore, existing technologies have limited accuracy in defect location and classification, failing to meet the requirements of high-precision production, and lack an effective verification mechanism for the inspection results, further affecting their reliability.

[0003] In the process of implementing the embodiments of the present invention, the inventors have discovered at least the following problems or defects in the prior art: On the one hand, traditional detection methods cannot efficiently and accurately identify a variety of complex defects in steel ladle permeable bricks, especially when multiple angles and types of defects coexist, the detection accuracy and reliability are difficult to guarantee; on the other hand, the prior art lacks the analysis of the spatial relationship between defects and the verification mechanism of the detection results, which may lead to misjudgment or omission of the detection results, and cannot meet the strict requirements of product quality and production efficiency in modern steel production. Summary of the Invention

[0004] This invention provides a method and system for detecting defects in permeable bricks in steel ladles based on image recognition.

[0005] In a first aspect of the present invention, a method for detecting defects in permeable bricks in steel ladles based on image recognition is provided, comprising:

[0006] The pre-acquired images of the target steel ladle permeable bricks are pre-processed to obtain a pre-processed image set, wherein the images of the target steel ladle permeable bricks are images taken by an industrial camera targeting the target steel ladle permeable bricks and whose resolution size meets the first condition.

[0007] For each block information in the block information set corresponding to the preprocessed image set, obtain a multi-angle image set with different resolution sizes that are captured by a multi-angle sensor for the area corresponding to the block information;

[0008] For each multi-angle image set in the obtained multi-angle image set group, each multi-angle image in the multi-angle image set is input into a pre-trained defect feature information extraction network to obtain a feature map set;

[0009] For each feature map set in the obtained feature map set group, the feature maps in the feature map set are integrated to obtain the integrated feature map;

[0010] For each integrated feature map in the obtained integrated feature map set, based on the integrated feature map and the defect identification model set, the location information and type information of each defect information in the defect information set corresponding to the integrated feature map are generated;

[0011] Based on the image of the target steel ladle permeable brick, the defect recognition model set, and the location and type information of each defect in the obtained defect information set, the defect detection result of the target steel ladle permeable brick is generated.

[0012] Furthermore, the step of preprocessing the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set includes: extracting edge feature information from the target steel ladle permeable brick image;

[0013] Based on the edge feature information, the image of the target steel ladle permeable brick is cropped and enhanced to obtain an enhanced image set;

[0014] The enhanced images that meet the third condition are selected from the enhanced image set and used as the preprocessed images to obtain the preprocessed image set.

[0015] Further, the step of generating the defect detection result of the target ladle permeable brick based on the target ladle permeable brick image, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set group includes: generating the location and type information of each initial defect information in the initial defect information set corresponding to the target ladle permeable brick image based on the target ladle permeable brick image and the defect recognition model set;

[0016] Based on the location and type information of each initial defect in the initial defect information set and the location and type information of each defect in the defect information set, the defect detection result of the target steel ladle permeable brick is generated.

[0017] Further, the step of generating the location information and type information of each initial defect in the initial defect information set corresponding to the target ladle permeable brick image based on the target ladle permeable brick image and the defect recognition model set includes: obtaining the priority of each defect recognition model in the defect recognition model set;

[0018] Select the defect identification model whose priority satisfies the fourth condition from the set of defect identification models, and use it as the target defect identification model;

[0019] The image of the target steel ladle permeable brick is input into a pre-trained target defect recognition model to obtain the location information of each first defect information in the first defect information set and the type information of each first defect information in the first defect information set.

[0020] The image of the target steel ladle permeable brick is input into each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each second defect information in the second defect information set, thereby obtaining the location information and type information of each second defect information in the second defect information set group;

[0021] For each location information of the first defect information, the following determination steps are performed: determine whether there is location information in the location information of the second defect information that is the same as the location information of the first defect information;

[0022] In response to the determination of existence, the location information that is the same as the location information of the first defect information is selected from the location information of each second defect information and used as the first target location information to obtain at least one first target location information;

[0023] Determine the number of each first target positioning information corresponding to the at least one first target positioning information;

[0024] In response to determining that the number is greater than or equal to a predetermined threshold, the location information of the first defect information is determined as the location information of the initial defect information, and the type information of the first defect information is determined as the type information of the corresponding initial defect information.

[0025] Further, the step of generating the location information and type information of each defect information in the defect information set corresponding to the integrated feature map based on the integrated feature map and the defect recognition model set includes: inputting the integrated feature map into a pre-trained target defect recognition model to obtain the location information and type information of each third defect information in the third defect information set;

[0026] The integrated feature map is input to each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each fourth defect information in the fourth defect information set, thereby obtaining the location information and type information of each fourth defect information in the fourth defect information set group;

[0027] For each location information of the third defect information, the following determination steps are performed: determine whether there is location information in the location information of the fourth defect information that is the same as the location information of the third defect information;

[0028] In response to the determination of existence, the location information that is the same as the location information of each third defect information is selected from the location information of each fourth defect information and used as the second target location information to obtain at least one second target location information;

[0029] Determine the number of each second target positioning information corresponding to the at least one second target positioning information;

[0030] In response to determining that the number is greater than or equal to the predetermined threshold, the location information of the third defect information is determined as the location information of the defect information, and the type information of the third defect information is determined as the type information of the corresponding defect information.

[0031] Further, the step of generating the defect detection result of the target steel ladle permeable brick based on the location information and type information of each initial defect information in the initial defect information set and the location information and type information of each defect information in the defect information set group includes: for each initial defect information in the initial defect information, performing the following generation steps to generate the defect detection result corresponding to the initial defect information: determining the location information of the initial defect information;

[0032] Determine the block information corresponding to the location information;

[0033] Determine the defect information set in the defect information set group corresponding to the block information as the target defect information set;

[0034] Determine whether there is any defect information in the target defect information set that has the same geographical location as the initial defect information;

[0035] In response to the determination of existence, a detection result is generated that indicates the presence of the defect corresponding to the initial defect information on the positioning information in the target steel ladle permeable brick;

[0036] Based on the obtained test results, the defect detection results of the target steel ladle permeable brick are generated.

[0037] Furthermore, the method also includes:

[0038] Obtain the location information of each defect corresponding to the defect detection results;

[0039] Based on the location information of each defect, the relative position information corresponding to each defect is generated;

[0040] Based on the relative position information corresponding to each defect, the spatial relationship between every two defects in each defect is determined, and a spatial relationship set is obtained.

[0041] The defect detection results are verified based on the spatial relationship set.

[0042] In a second aspect of the invention, an image recognition-based defect detection system for permeable bricks in steel ladles is provided, comprising:

[0043] The preprocessing unit is configured to perform image preprocessing on the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set, wherein the target steel ladle permeable brick image is an image taken by an industrial camera targeting the target steel ladle permeable brick and whose resolution size meets the first condition.

[0044] The acquisition unit is configured to acquire, for each block information in the block information set corresponding to the preprocessed image set, a multi-angle image set captured by a multi-angle sensor for the region corresponding to the block information, with a resolution size satisfying different second conditions;

[0045] The input unit is configured to input each multi-angle image in the obtained multi-angle image set into a pre-trained defect feature information extraction network to obtain a feature map set for each multi-angle image set in the obtained multi-angle image set group.

[0046] The integration unit is configured to integrate the individual feature maps in each feature map set in the obtained feature map set group to obtain an integrated feature map;

[0047] The first generation unit is configured to generate, for each integrated feature map in the obtained integrated feature map set, location information and type information of each defect information in the defect information set corresponding to the integrated feature map, based on the integrated feature map and the defect identification model set;

[0048] The second generation unit is configured to generate the defect detection result of the target steel ladle permeable brick based on the image of the target steel ladle permeable brick, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set.

[0049] In a third aspect of the invention, an electronic device is provided, comprising: at least one processor, a memory, and an input / output unit; wherein the memory is used to store a computer program, and the processor is used to invoke the computer program stored in the memory to perform the method described in any one of the first aspects.

[0050] In a fourth aspect of the invention, a computer-readable storage medium is provided, comprising instructions that, when executed on a computer, cause the computer to perform the method described in any one of the first aspects.

[0051] The embodiments of the present invention have at least the following beneficial effects:

[0052] (1) This invention uses multiple steps such as image preprocessing, multi-angle image acquisition, feature map integration and multi-model recognition to comprehensively analyze the defects of steel ladle permeable bricks from different dimensions, reducing the misjudgment that may be caused by a single perspective or a single model. Specifically, it integrates the features extracted from different angles to form a more comprehensive feature representation. At the same time, it utilizes the advantages of different models to improve the accuracy of defect location and classification, effectively reduce misjudgment and improve the reliability of detection.

[0053] (2) The process of generating the defect detection results of the target steel ladle permeable brick in this invention is based on the initial defect information and the defect information set obtained by multi-angle image analysis. By comprehensively analyzing the defect information from different sources, the misjudgment that may be caused by a single information source is effectively avoided, and the comprehensiveness and accuracy of the detection results are ensured.

[0054] (3) The present invention determines the spatial relationship between defects based on the relative position information of the defects, and verifies the test results accordingly, thereby improving the reliability of the test results and providing strong support for the quality control of steel ladle permeable bricks.

[0055] (4) The system provided by this invention can improve detection efficiency and reduce labor costs. The automated image acquisition and processing process reduces the time and effort required for manual inspection, while the multi-model collaborative recognition and result verification mechanism reduces the need for manual re-inspection, making the entire defect detection process more efficient and faster, better adaptable to the quality inspection needs in large-scale production, and helping to improve production efficiency and product quality stability. Attached Figure Description

[0056] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0057] Figure 1 This is a flowchart illustrating a defect detection method for permeable bricks in steel ladles based on image recognition, according to an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of a defect detection system for permeable bricks in steel ladles based on image recognition, provided in an embodiment of the present invention.

[0059] Figure 3 A schematic diagram of the structure of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0060] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.

[0061] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0062] It should be noted that the number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0063] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating a defect detection method for permeable bricks in steel ladles based on image recognition, according to an embodiment of the present invention. Figure 1 As shown, a defect detection method for permeable bricks in steel ladles based on image recognition includes:

[0064] S1 performs image preprocessing on the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set, wherein the target steel ladle permeable brick image is an image taken by an industrial camera targeting the target steel ladle permeable brick and whose resolution size meets the first condition.

[0065] S2 For each block information in the block information set corresponding to the preprocessed image set, obtain a multi-angle image set with different resolution sizes that are captured by a multi-angle sensor for the area corresponding to the block information.

[0066] S3 For each multi-angle image set in the obtained multi-angle image set group, input each multi-angle image in the multi-angle image set into the pre-trained defect feature information extraction network to obtain a feature map set;

[0067] S4 For each feature map set in the obtained feature map set group, integrate the feature maps in the feature map set to obtain the integrated feature map;

[0068] S5 For each integrated feature map in the obtained integrated feature map set, based on the integrated feature map and the defect identification model set, generate the location information and type information of each defect information in the defect information set corresponding to the integrated feature map;

[0069] S6 generates the defect detection result of the target steel ladle permeable brick based on the image of the target steel ladle permeable brick, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set.

[0070] It should be noted that this invention proposes a defect detection method for steel ladle permeable bricks based on image recognition. The method first preprocesses the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set. Here, the target steel ladle permeable brick image refers to an image captured by an industrial camera targeting the target steel ladle permeable brick, and the resolution of this image needs to meet a first condition, namely, achieving a certain level of clarity, so that defects can be accurately identified subsequently. The preprocessed image set is used for more precise analysis and processing of image data. Next, for each block information in the block information set corresponding to the preprocessed image set, a multi-angle image set captured by a multi-angle sensor targeting the corresponding area of ​​the block information is obtained. The resolution of these images meets different second conditions, meaning that images captured from different angles may have different resolution requirements to adapt to defect feature capture from different perspectives. Then, each multi-angle image in each multi-angle image set is input into a pre-trained defect feature information extraction network to obtain a feature map set. This process utilizes a deep learning network to extract key features from the image for subsequent defect identification. Then, the feature maps in each feature map set are integrated to obtain an integrated feature map. This integration process aims to fuse features extracted from different perspectives to form a more comprehensive feature representation. Finally, based on the integrated feature map and the defect recognition model set, the location and type information of each defect in the defect information set are generated. Combined with the target steel ladle permeable brick image, the defect recognition model set, and the location and type information from the defect information set, the final defect detection result is generated. This process involves multiple steps and the collaborative work of multiple models, aiming to improve the accuracy and reliability of defect detection.

[0071] Specifically, the image of the target steel ladle permeable brick is obtained through an industrial camera. An industrial camera is a high-resolution imaging device capable of capturing detailed information about the surface of the steel ladle permeable brick. The first condition is that the image resolution should meet a certain standard, such as at least 1920×1080 pixels, to ensure that the image contains sufficient detail for subsequent analysis. A multi-angle sensor refers to a device capable of photographing the steel ladle permeable brick from different angles. The second condition is that the resolution of images taken from different angles may differ. For example, an image taken from the front may have a resolution of 1920×1080 pixels, while an image taken from the side may have a resolution of 1280×720 pixels. This is because images taken from different angles have different uses and therefore different resolution requirements. The defect feature information extraction network is a pre-trained deep learning model, such as a convolutional neural network (CNN), that can automatically extract key features from the image, such as edges and textures. These features are crucial for defect identification. The feature map set is a set of images output by this network, with each image representing features of the input image at different levels. Feature map integration refers to fusing these feature maps, for example, through weighted averaging or stitching, to form an integrated feature map containing more comprehensive information. The defect recognition model set is a group of models used to identify defects. Each model may be trained for different types of defects or images from different angles. They can output the defect's location information (such as the defect's coordinates in the image) and type information (such as cracks, holes, etc.) based on the input feature map. The final defect detection result is derived by comprehensively considering the location and type information from the target ladle permeable brick image, the defect recognition model set, and the defect information set. It can accurately indicate whether a defect exists on the ladle permeable brick, as well as the location and type of the defect.

[0072] Preferably, a convolutional neural network (CNN) architecture, such as ResNet or VGG, can be used to construct the defect feature extraction network. These networks extract features from the input image through multiple convolutional and pooling layers. The size and number of convolutional kernels in each layer can be adjusted according to actual needs. For example, the first layer can have a kernel size of 3×3 and a number of 64 kernels to extract low-level features; the kernel size and number of subsequent layers can be gradually increased to extract deeper features. During network training, the input parameters include a large number of labeled images of steel-clad permeable bricks. The labeling information includes the location and type of defects. The weights of the network are optimized using a backpropagation algorithm, enabling the network to accurately extract defect features. For the feature map integration step, a weighted average method can be used, assigning different weights according to the importance of each feature map. For example, feature maps taken from the front are given higher weights, while feature maps taken from the side are given lower weights. The weight values ​​can be determined experimentally to ensure that the integrated feature maps better reflect the characteristics of the defects. In constructing a defect identification model set, models can be trained separately for different types of defects. For example, one model can be specifically used to identify cracks, while another model can be used to identify holes. Each model can employ different algorithms, such as Support Vector Machines (SVM) or decision trees. The input parameters are the integrated feature maps, and the output is the location and type information of the defect. In this way, the accuracy and specificity of defect identification can be improved.

[0073] In some embodiments, the step of preprocessing the image of the target steel ladle permeable brick to obtain a preprocessed image set includes: extracting edge feature information of the target steel ladle permeable brick image;

[0074] Based on the edge feature information, the image of the target steel ladle permeable brick is cropped and enhanced to obtain an enhanced image set;

[0075] The enhanced images that meet the third condition are selected from the enhanced image set and used as the preprocessed images to obtain the preprocessed image set.

[0076] It should be noted that the image preprocessing step for the target steel ladle permeable brick image in this invention is to improve image quality and usability, thereby providing a better foundation for subsequent defect detection. The purpose of image preprocessing is to enhance key structures in the image by extracting edge feature information, and then cropping and enhancing the image based on these edge features, ultimately selecting images that meet specific conditions as the preprocessed image set. Here, edge feature information refers to parts of the image with significant changes in brightness or color, typically corresponding to the outline of an object or surface discontinuities; these features are crucial for defect identification. Through cropping and enhancement operations, irrelevant background information in the image can be removed, while highlighting areas that may contain defects, thereby improving the efficiency and accuracy of detection.

[0077] Specifically, the image preprocessing process includes the following key steps. First, the edge feature information of the target steel ladle permeable brick image is extracted. This can be achieved by applying edge detection algorithms, such as the Sobel operator or the Canny edge detection algorithm. These algorithms can detect the gradient of brightness changes in the image, thereby locating the edge position. Next, the image is cropped based on the extracted edge feature information. The purpose of cropping is to remove areas in the image that are irrelevant to defect detection, such as the background, thereby reducing the amount of data for subsequent processing and improving processing speed. The cropping range can be determined based on the location of the edge features; for example, the cropping area can be set as the smallest rectangular area containing the edge features. Then, the cropped image is enhanced. Enhancement operations can include contrast adjustment, sharpening, etc., to improve the visibility of defect features in the image. Finally, images that meet the third condition are selected from the enhanced image set as preprocessed images. The third condition can be image sharpness, contrast, or other quality indicators. For example, a sharpness threshold can be set, and only images with a sharpness higher than the threshold are included in the preprocessed image set.

[0078] Preferably, for edge feature extraction, the Canny edge detection algorithm can be used. This algorithm detects edges in an image through multiple steps, including noise removal, gradient calculation, non-maximum suppression, and hysteresis thresholding. When applying the Canny algorithm, two threshold parameters need to be set: a low threshold and a high threshold. The low threshold is used to detect weak edges, and the high threshold is used to detect strong edges. These two thresholds can be adjusted according to the characteristics of the actual image. For example, for a high-resolution image of a steel ladle permeable brick, the low threshold can be set to 50, and the high threshold can be set to 150. When cropping the image, the size and position of the cropping region can be determined based on the location of the edge features. For example, the cropping region can be set as a rectangular area extending a certain number of pixels beyond the edge features. The number of pixels extended can be determined based on the expected size of the defect. For image enhancement, histogram equalization can be used to adjust the image contrast. This method improves image quality by changing the gray-level distribution of the image, making the contrast more uniform. When screening preprocessed images, image sharpness evaluation metrics can be used. For example, the image sharpness can be evaluated by calculating the variance of the Laplacian operator of the image. Only when the variance is greater than a certain set threshold is the image considered to meet the third condition.

[0079] In some embodiments, generating the defect detection result of the target ladle permeable brick based on the target ladle permeable brick image, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set group includes: generating the location and type information of each initial defect information in the initial defect information set corresponding to the target ladle permeable brick image based on the target ladle permeable brick image and the defect recognition model set;

[0080] Based on the location and type information of each initial defect in the initial defect information set and the location and type information of each defect in the defect information set, the defect detection result of the target steel ladle permeable brick is generated.

[0081] It should be noted that the process of generating defect detection results for the target ladle permeable brick mentioned in this invention is based on the image of the target ladle permeable brick, the defect recognition model set, and the location and type information in the defect information set. Here, the defect detection result refers to the output information after comprehensively judging whether the ladle permeable brick has defects, the location of the defects, and the type of the defects. The purpose of this step is to integrate information from multiple sources to improve the accuracy and reliability of defect detection. By combining initial defect information and defect information obtained from multi-angle image analysis, the defect situation of the ladle permeable brick can be more comprehensively evaluated, thereby providing a basis for subsequent quality control and maintenance.

[0082] Specifically, initial defect information refers to defect information obtained through preliminary analysis of the target steel ladle permeable brick image, including defect location and type information. Location information refers to the coordinate position of the defect in the image, such as (x, y) coordinates in pixels; type information is the classification of the defect, such as cracks, holes, wear, etc. The defect information set refers to the collection of defect information obtained through multi-angle image analysis, containing defect information identified from images taken from different angles. The integration of this information is accomplished by comparing the location and type information in the initial defect information and the defect information set. For example, if the location of a defect in the initial defect information is the same as or close to the location of a defect in the defect information set, the two defects can be considered the same, thus further confirming the existence of the defect and determining its type.

[0083] Preferably, the process of generating defect detection results can be further refined. For example, when comparing the initial defect information with the location information in the defect information set, a tolerance range can be set, such as an error range in pixels. If the location information of two defects is within this tolerance range, they are considered to be the same defect. Furthermore, a voting mechanism can be used to determine the defect type. For example, if multiple defect recognition models agree on the type of a certain defect, the type of the defect can be determined with greater confidence. Simultaneously, to improve the reliability of the detection results, the defect information can be further verified. For example, if a defect is detected in images from multiple angles, the confidence level for that defect can be increased. Finally, the defect detection results generated based on this comprehensive information can be output in the form of a report, including detailed information such as the location, type, and confidence level of the defect, providing users with comprehensive and accurate defect detection results.

[0084] In some embodiments, generating the location information and type information of each initial defect in the initial defect information set corresponding to the target ladle permeable brick image based on the target ladle permeable brick image and the defect recognition model set includes: obtaining the priority of each defect recognition model in the defect recognition model set;

[0085] Select the defect identification model whose priority satisfies the fourth condition from the set of defect identification models, and use it as the target defect identification model;

[0086] The target steel ladle permeable brick image is input into a pre-trained target defect recognition model to obtain the location information and type information of each first defect information in the first defect information set.

[0087] The image of the target steel ladle permeable brick is input into each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each second defect information in the second defect information set, thereby obtaining the location information and type information of each second defect information in the second defect information set group;

[0088] For each location information of the first defect information, the following determination steps are performed: determine whether there is location information in the location information of the second defect information that is the same as the location information of the first defect information;

[0089] In response to the determination of existence, the location information that is the same as the location information of the first defect information is selected from the location information of each second defect information and used as the first target location information to obtain at least one first target location information;

[0090] Determine the number of each first target positioning information corresponding to the at least one first target positioning information;

[0091] In response to determining that the number is greater than or equal to a predetermined threshold, the location information of the first defect information is determined as the location information of the initial defect information, and the type information of the first defect information is determined as the type information of the corresponding initial defect information.

[0092] It should be noted that the process of generating initial defect information corresponding to the target steel ladle permeable brick image mentioned in this invention is achieved by combining multiple models from a defect recognition model set. The purpose of this process is to leverage the advantages of different models to improve the accuracy of defect localization and classification. Specifically, firstly, the priority of each model in the defect recognition model set needs to be obtained; the priority reflects the reliability or importance of the model in identifying a specific defect. Then, one or more models are selected as target defect recognition models according to the priority, and the target steel ladle permeable brick image is input into these models to obtain preliminary defect information. In addition, the image is also input into other non-target models to obtain more reference information. By comparing the output results of these models, consistent defect information is filtered out, thereby determining the initial defect information. This method can effectively reduce false positives and improve the reliability of detection.

[0093] Specifically, the defect recognition model set is a collection of multiple models trained for different defect types or detection angles. Priority refers to the order in which each model identifies a specific defect; for example, a model specifically designed for crack detection might have a higher priority for crack detection. The target defect recognition model is selected from the defect recognition model set and used for initial defect detection. Models whose priority meets the fourth condition are selected as target models. The fourth condition can be a priority higher than a set threshold, such as a priority greater than 0.8 (assuming the priority range is 0 to 1). After inputting the target steel ladle permeable brick image into the target defect recognition model, a first defect information set is obtained, containing the location information (e.g., the coordinate position of the defect in the image) and type information (e.g., cracks, holes, etc.) of each defect. Simultaneously, the image is also input into other models in the defect recognition model set besides the target model, resulting in a second defect information set. Next, by comparing the location information in the first and second defect information sets, consistent defect information is filtered out. For example, if the location information of a certain defect exists in the outputs of multiple models, the location information of that defect is considered reliable. The predetermined threshold refers to the minimum number of models used to determine consistency. For example, if the location information of a defect is present in the output of at least 3 models, the location information of the defect is considered valid.

[0094] Preferably, the defect identification model can be constructed using deep learning methods, such as convolutional neural networks (CNNs). The model's input parameters are images of the target steel ladle permeable bricks, and the image resolution and size need to meet the model's input requirements. During model training, a large number of labeled steel ladle permeable brick images can be used as training data, with annotation information including the location and type of defects. The model training process includes forward propagation and backpropagation, adjusting the model's weights by optimizing the loss function to ensure accurate defect identification. Priority settings can be determined based on the model's performance on the validation set; for example, higher accuracy on the validation set corresponds to higher priority. When filtering consistent defect information, the comparison process can be further refined. For example, for the location information of each first defect, not only can it be compared to see if it has the same points as the location information of a second defect, but the similarity of the location information can also be considered, allowing for a certain error range to improve the flexibility and accuracy of the filtering.

[0095] In some embodiments, generating the location information and type information of each defect information in the defect information set corresponding to the integrated feature map based on the integrated feature map and the defect recognition model set includes: inputting the integrated feature map into a pre-trained target defect recognition model to obtain the location information of each third defect information in the third defect information set and the type information of each third defect information in the third defect information set;

[0096] The integrated feature map is input to each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each fourth defect information in the fourth defect information set, thereby obtaining the location information and type information of each fourth defect information in the fourth defect information set group;

[0097] For each location information of the third defect information, the following determination steps are performed: determine whether there is location information in the location information of the fourth defect information that is the same as the location information of the third defect information;

[0098] In response to the determination of existence, the location information that is the same as the location information of each third defect information is selected from the location information of each fourth defect information and used as the second target location information to obtain at least one second target location information;

[0099] Determine the number of each second target positioning information corresponding to the at least one second target positioning information;

[0100] In response to determining that the number is greater than or equal to the predetermined threshold, the location information of the third defect information is determined as the location information of the defect information, and the type information of the third defect information is determined as the type information of the corresponding defect information.

[0101] It should be noted that the process of generating defect information from the integrated feature map mentioned in this invention is implemented using a set of defect recognition models. This process aims to improve the accuracy and reliability of defect detection through the collaborative work of multiple models. Specifically, the integrated feature map is first input into the target defect recognition model to obtain preliminary defect information, including the location and type of the defect. Then, the same feature map is input into other non-target models to obtain more reference information. By comparing the outputs of these models, consistent defect information is selected to determine the final defect information. This method can effectively reduce false positives and ensure the accuracy of the detection results.

[0102] Specifically, the integrated feature map refers to the image feature representation after multi-angle image feature extraction and integration. It integrates image features taken from different angles, and can more comprehensively reflect the surface condition of the steel ladle permeable brick. The target defect recognition model is selected from the defect recognition model set and is used for preliminary defect detection. It is usually selected based on the model priority or specific task requirements. After inputting the integrated feature map into the target defect recognition model, a third defect information set is obtained, which contains the location information (such as the coordinate position of the defect in the image) and type information (such as cracks, holes, etc.) of each defect. At the same time, this feature map is input into other models in the defect recognition model set other than the target model to obtain a fourth defect information set. Next, by comparing the location information in the third defect information set and the fourth defect information set, consistent defect information is filtered out. For example, if the location information of a certain defect exists in the output of multiple models, the location information of the defect is considered reliable. The predetermined threshold refers to the minimum number of models used to judge consistency. For example, if the location information of a certain defect exists in the output of at least 3 models, the location information of the defect is considered valid.

[0103] Preferably, the target defect recognition model can be constructed using deep learning methods, such as convolutional neural networks (CNNs). The model's input parameters are the integrated feature maps, and the size and number of channels of the feature maps need to meet the model's input requirements. During model training, a large number of labeled feature maps of steel ladle permeable bricks can be used as training data, with annotation information including the location and type of the defect. The model training process includes forward propagation and backward propagation, adjusting the model's weights by optimizing the loss function to ensure accurate defect recognition. Priority settings can be determined based on the model's performance on the validation set; for example, higher accuracy on the validation set corresponds to higher priority. When filtering consistent defect information, the comparison process can be further refined. For example, for the location information of each third defect, not only can it be compared to see if it has the same points as the location information of the fourth defect, but the similarity of the location information can also be considered, allowing for a certain error range to improve the flexibility and accuracy of the filtering. Furthermore, for defect type judgment, a voting mechanism can be used, where multiple models statistically analyze the judgment results for the same defect type, selecting the type that appears most frequently as the final defect type.

[0104] In some embodiments, generating the defect detection result of the target steel ladle permeable brick based on the location information and type information of each initial defect information in the initial defect information set and the location information and type information of each defect information in the defect information set group includes: for each initial defect information in the initial defect information, performing the following generation steps to generate the defect detection result corresponding to the initial defect information: determining the location information of the initial defect information;

[0105] Determine the block information corresponding to the location information;

[0106] Determine the defect information set in the defect information set group corresponding to the block information as the target defect information set;

[0107] Determine whether there is any defect information in the target defect information set that has the same geographical location as the initial defect information;

[0108] In response to the determination of existence, a detection result is generated that indicates the presence of the defect corresponding to the initial defect information on the positioning information in the target steel ladle permeable brick;

[0109] Based on the obtained test results, the defect detection results of the target steel ladle permeable brick are generated.

[0110] It should be noted that the process of generating defect detection results for the target steel ladle permeable brick mentioned in this invention is based on a set of defect information obtained from initial defect information and multi-angle image analysis. The purpose of this process is to improve the accuracy and reliability of defect detection by comprehensively analyzing defect information from different sources. Specifically, for each initial defect information, its exact location on the target steel ladle permeable brick needs to be determined, and combined with the defect information obtained from multi-angle image analysis, it is determined whether a corresponding defect exists at that location. If it exists, a detection result characterizing the defect is generated, and finally, all detection results are summarized to form a complete defect detection report. This method can effectively avoid misjudgments that may arise from a single information source, ensuring the comprehensiveness and accuracy of the detection results.

[0111] Specifically, initial defect information refers to defect information obtained through preliminary analysis of the target steel ladle permeable brick image, including defect location information (such as the defect's coordinate position in the image) and type information (such as cracks, holes, etc.). Block information refers to the region division information corresponding to the preprocessed image set, used to determine the specific location of the defect on the steel ladle permeable brick. Defect information set group refers to the set of defect information obtained through multi-angle image analysis, which includes defect information identified from images taken from different angles. Target defect information set refers to the multi-angle image analysis results corresponding to the initial defect information, that is, the set of defect information with the same geographical location as the initial defect information. The process of generating defect detection results includes: first, determining the location information of the initial defect information; then, finding the corresponding block information based on the location information; next, extracting the target defect information set corresponding to the block information from the defect information set group; and finally, determining whether there is defect information with the same geographical location as the initial defect information in the target defect information set. If so, generating a detection result characterizing the defect, such as recording the defect's location, type, and severity.

[0112] Preferably, the process of generating defect detection results can be further refined. For example, when determining the location information of the initial defect information, a sub-pixel-level precision positioning algorithm can be used to improve the accuracy of positioning. When determining whether there is defect information in the target defect information set that has the same geographical location as the initial defect information, a tolerance range can be set, such as an error range in pixels, allowing for a certain positioning deviation to improve the flexibility of detection. In addition, for the determination of defect type, the type information in the initial defect information and the target defect information set can be combined, and a majority voting mechanism or a confidence weighting mechanism can be used to select the type that appears most frequently or has the highest confidence as the final defect type. The final generated defect detection results can be output in the form of a structured report, including detailed information such as the location, type, severity, and suggested repair measures of the defect, providing clear guidance for the maintenance and repair of steel ladle permeable bricks.

[0113] In some embodiments, the method further includes:

[0114] Obtain the location information of each defect corresponding to the defect detection results;

[0115] Based on the location information of each defect, the relative position information corresponding to each defect is generated;

[0116] Based on the relative position information corresponding to each defect, the spatial relationship between every two defects in each defect is determined, and a spatial relationship set is obtained;

[0117] The defect detection results are verified based on the spatial relationship set.

[0118] It should be noted that the process of verifying defect detection results mentioned in this invention is achieved by analyzing the spatial relationships between defects. The purpose of this process is to further verify the accuracy and rationality of the detection results. Specifically, firstly, relative positional information between each defect is generated based on its location information. Then, by analyzing this relative positional information, the spatial relationships between defects are determined, such as distances and angles between defects. Finally, the detection results are verified based on these spatial relationships to determine whether the results conform to actual physical laws and defect distribution characteristics. For example, if two defects cannot possibly occur simultaneously in reality, but the detection results show their presence, then it can be considered that the detection results may be misjudged. In this way, the reliability of defect detection results can be effectively improved.

[0119] Specifically, defect location information refers to the exact position of the defect on the permeable brick of the ladle, usually represented by coordinates, such as (x, y) coordinates. Relative position information refers to the positional relationship between two defects, which can be obtained by calculating the Euclidean or Manhattan distance between the coordinates of the two defects. The spatial relationship set refers to the collection of relative positional information between all defects, reflecting the spatial distribution characteristics of the defects. When determining the spatial relationships between defects, some rules can be set according to the actual application scenario. For example, certain types of defects cannot appear simultaneously within a certain distance, or the distribution pattern of certain defects conforms to a specific geometry. By analyzing these spatial relationships, the reasonableness of the detection results can be judged. For example, if the detection results show that the distance between two defects is less than a certain set threshold, but according to practical experience, these two defects cannot be so close, then the detection results may be misjudged.

[0120] Preferably, the process of generating relative position information can take the following steps: First, calculate the Euclidean distance between each pair of defects, using the formula:

[0121]

[0122] in, and These are the coordinates of two defects. Then, reasonable distance thresholds are set according to the actual application scenario. For example, for certain types of defects, a minimum distance threshold of 10 mm and a maximum distance threshold of 100 mm are set. If the distance between two defects is less than the minimum threshold or greater than the maximum threshold, the spatial relationship between the two defects is considered unreasonable, and the detection result is corrected or marked as suspicious. Furthermore, angle information can be introduced to further analyze the spatial relationship between defects. For example, the angle between the two defects and the center point of the steel ladle permeable brick is calculated. If the angle does not conform to the expected distribution pattern, the accuracy of the detection result can be further verified. Through these refined steps, the spatial relationship between defects can be analyzed more comprehensively, thereby more effectively verifying the defect detection results.

[0123] The above-described embodiments of the present invention have the following beneficial effects: The present invention can effectively improve the identification accuracy of surface defects in steel ladle permeable bricks by multi-angle image acquisition and feature fusion processing, combined with multi-resolution image analysis. Preprocessing methods such as edge feature extraction and image enhancement can optimize the quality of the original image; the use of multi-sensor collaborative acquisition and feature map integration technology can comprehensively capture defect features from different perspectives; and the collaborative verification mechanism based on the defect identification model set can reduce the false detection rate through comparison of multiple model results, thereby ensuring the accuracy of defect localization and classification.

[0124] Furthermore, this invention can further enhance the reliability of detection results through spatial relationship verification and multi-level defect information matching. Cross-validation of initial defect information with multi-angle detection results can eliminate isolated misjudgments; combined with analysis of the relative positions of defects, the logical rationality of defect distribution can be verified; and finally, by generating detection results through comprehensive judgment, more comprehensive data support can be provided for the quality assessment of steel ladle permeable bricks, meeting the high-precision requirements of industrial testing.

[0125] like Figure 2 As shown in some embodiments, a defect detection system for permeable bricks in steel ladles based on image recognition is provided. The system includes:

[0126] The preprocessing unit 201 is configured to perform image preprocessing on the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set, wherein the target steel ladle permeable brick image is an image taken by an industrial camera targeting the target steel ladle permeable brick and whose resolution size meets the first condition.

[0127] The acquisition unit 202 is configured to acquire, for each block information in the block information set corresponding to the preprocessed image set, a multi-angle image set captured by a multi-angle sensor for the area corresponding to the block information, with a resolution size satisfying different second conditions;

[0128] The input unit 203 is configured to input each multi-angle image in the obtained multi-angle image set into a pre-trained defect feature information extraction network to obtain a feature map set for each multi-angle image set in the obtained multi-angle image set group.

[0129] The integration unit 204 is configured to integrate the individual feature maps in each feature map set in the obtained feature map set group to obtain an integrated feature map.

[0130] The first generation unit 205 is configured to generate, for each integrated feature map in the obtained integrated feature map set, location information and type information of each defect information in the defect information set corresponding to the integrated feature map, based on the integrated feature map and the defect identification model set;

[0131] The second generation unit 206 is configured to generate the defect detection result of the target steel ladle permeable brick based on the target steel ladle permeable brick image, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set.

[0132] It is understood that the modules described in the image recognition-based defect detection system for permeable bricks in steel ladles are similar to those in the reference system. Figure 1The steps described in the image recognition-based defect detection method for permeable bricks in steel ladles correspond to each other. Therefore, the operations, features, and beneficial effects described above for the image recognition-based defect detection method for permeable bricks in steel ladles are also applicable to an image recognition-based defect detection system for permeable bricks in steel ladles and the modules contained therein, and will not be repeated here.

[0133] The following is for reference. Figure 3 The diagram illustrates a structural schematic of an electronic device 300 suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The terminal device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0134] like Figure 3 As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0135] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0136] Furthermore, the storage medium in the embodiments of this application stores program instructions capable of implementing all the above methods. These program instructions can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0137] The above description is merely an explanation of some preferred embodiments of the present invention and the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments of the present invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for defect detection of permeable bricks in steel ladles based on image recognition, characterized in that, include: The pre-acquired images of the target steel ladle permeable bricks are pre-processed to obtain a pre-processed image set, wherein the images of the target steel ladle permeable bricks are images taken by an industrial camera targeting the target steel ladle permeable bricks and whose resolution size meets the first condition. For each block information in the block information set corresponding to the preprocessed image set, obtain a multi-angle image set with different resolution sizes that are captured by a multi-angle sensor for the area corresponding to the block information; For each multi-angle image set in the obtained multi-angle image set group, each multi-angle image in the multi-angle image set is input into a pre-trained defect feature information extraction network to obtain a feature map set; For each feature map set in the obtained feature map set group, the feature maps in the feature map set are integrated to obtain the integrated feature map; For each integrated feature map in the obtained integrated feature map set, based on the integrated feature map and the defect identification model set, the location information and type information of each defect information in the defect information set corresponding to the integrated feature map are generated; Based on the image of the target steel ladle permeable brick, the defect recognition model set, and the location and type information of each defect in the obtained defect information set, the defect detection result of the target steel ladle permeable brick is generated.

2. The method according to claim 1, characterized in that, The step of preprocessing the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set includes: extracting edge feature information of the target steel ladle permeable brick image; Based on the edge feature information, the image of the target steel ladle permeable brick is cropped and enhanced to obtain an enhanced image set; The enhanced images that meet the third condition are selected from the enhanced image set and used as the preprocessed images to obtain the preprocessed image set.

3. The method according to claim 1, characterized in that, The step of generating the defect detection result of the target ladle permeable brick based on the target ladle permeable brick image, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set group includes: generating the location and type information of each initial defect information in the initial defect information set corresponding to the target ladle permeable brick image based on the target ladle permeable brick image and the defect recognition model set; Based on the location and type information of each initial defect in the initial defect information set and the location and type information of each defect in the defect information set, the defect detection result of the target steel ladle permeable brick is generated.

4. The method according to claim 3, characterized in that, The step of generating the location information and type information of each initial defect in the initial defect information set corresponding to the target steel ladle permeable brick image based on the target steel ladle permeable brick image and the defect recognition model set includes: obtaining the priority of each defect recognition model in the defect recognition model set; Select the defect identification model whose priority satisfies the fourth condition from the set of defect identification models, and use it as the target defect identification model; The target steel ladle permeable brick image is input into a pre-trained target defect recognition model to obtain the location information and type information of each first defect information in the first defect information set. The image of the target steel ladle permeable brick is input into each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each second defect information in the second defect information set, thereby obtaining the location information and type information of each second defect information in the second defect information set group; For each location information of the first defect information, the following determination steps are performed: determine whether there is location information in the location information of the second defect information that is the same as the location information of the first defect information; In response to the determination of existence, the location information that is the same as the location information of the first defect information is selected from the location information of each second defect information and used as the first target location information to obtain at least one first target location information; Determine the number of each first target positioning information corresponding to the at least one first target positioning information; In response to determining that the number is greater than or equal to a predetermined threshold, the location information of the first defect information is determined as the location information of the initial defect information, and the type information of the first defect information is determined as the type information of the corresponding initial defect information.

5. The method according to claim 4, characterized in that, The step of generating the location information and type information of each defect information in the defect information set corresponding to the integrated feature map and the defect recognition model set includes: inputting the integrated feature map into a pre-trained target defect recognition model to obtain the location information and type information of each third defect information in the third defect information set; The integrated feature map is input to each defect recognition model in the defect recognition model set except for the target defect recognition model, so as to output the location information and type information of each fourth defect information in the fourth defect information set, thereby obtaining the location information and type information of each fourth defect information in the fourth defect information set group; For each location information of the third defect information, the following determination steps are performed: determine whether there is location information in the location information of the fourth defect information that is the same as the location information of the third defect information; In response to the determination of existence, the location information that is the same as the location information of each third defect information is selected from the location information of each fourth defect information and used as the second target location information to obtain at least one second target location information; Determine the number of each second target positioning information corresponding to the at least one second target positioning information; In response to determining that the number is greater than or equal to the predetermined threshold, the location information of the third defect information is determined as the location information of the defect information, and the type information of the third defect information is determined as the type information of the corresponding defect information.

6. The method according to claim 3, characterized in that, The step of generating the defect detection result of the target steel ladle permeable brick based on the location information and type information of each initial defect information in the initial defect information set and the location information and type information of each defect information in the defect information set group includes: for each initial defect information in the initial defect information, performing the following generation steps to generate the defect detection result corresponding to the initial defect information: determining the location information of the initial defect information; Determine the block information corresponding to the location information; Determine the defect information set in the defect information set group corresponding to the block information as the target defect information set; Determine whether there is any defect information in the target defect information set that has the same geographical location as the initial defect information; In response to the determination of existence, a detection result is generated that indicates the presence of the defect corresponding to the initial defect information on the positioning information in the target steel ladle permeable brick; Based on the obtained test results, the defect detection results of the target steel ladle permeable brick are generated.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the location information of each defect corresponding to the defect detection results; Based on the location information of each defect, the relative position information corresponding to each defect is generated; Based on the relative position information corresponding to each defect, the spatial relationship between every two defects in each defect is determined, and a spatial relationship set is obtained; The defect detection results are verified based on the spatial relationship set.

8. A defect detection system for permeable bricks in steel ladles based on image recognition, characterized in that, include: The preprocessing unit is configured to perform image preprocessing on the pre-acquired target steel ladle permeable brick image to obtain a preprocessed image set, wherein the target steel ladle permeable brick image is an image taken by an industrial camera targeting the target steel ladle permeable brick and whose resolution size meets the first condition. The acquisition unit is configured to acquire, for each block information in the block information set corresponding to the preprocessed image set, a multi-angle image set captured by a multi-angle sensor for the region corresponding to the block information, with a resolution size satisfying different second conditions; The input unit is configured to input each multi-angle image in the obtained multi-angle image set into a pre-trained defect feature information extraction network to obtain a feature map set for each multi-angle image set in the obtained multi-angle image set group. The integration unit is configured to integrate the individual feature maps in each feature map set in the obtained feature map set group to obtain an integrated feature map. The first generation unit is configured to generate, for each integrated feature map in the obtained integrated feature map set, location information and type information of each defect information in the defect information set corresponding to the integrated feature map, based on the integrated feature map and the defect identification model set; The second generation unit is configured to generate the defect detection result of the target steel ladle permeable brick based on the image of the target steel ladle permeable brick, the defect recognition model set, and the location and type information of each defect information in the obtained defect information set.

9. An electronic device, characterized in that, include: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images