Visualization method and device of industrial defect image, medium and computer equipment

Abstract

The invention discloses an industrial defect image visualization method and device, a storage medium and computer equipment, and the method comprises the steps: determining a plurality of defect images to be displayed and a target display dimension in response to a defect image display instruction; for each defect image, acquiring an image feature vector corresponding to the defect image, performing dimension reduction processing on the image feature vector according to the target display dimension, and determining a coordinate position of a data point corresponding to the defect image according to the image feature vector after the dimension reduction processing; and for the data point of each defect image, determining a first display position of the data point on an image display area based on the coordinate position of the data point, and displaying the defect image on the image display area in the form of a thumbnail image of the data point and the defect image according to the first display position.

Description

Technical Field

[0001] This application relates to the technical field of image analysis and visualization, and particularly to a visualization method and device for industrial defect images, a storage medium, and a computer device. Background Art

[0002] In the process of industrial manufacturing, quality control is a key link to ensure the performance and reliability of products. The wide application of intelligent manufacturing and automated detection technologies has made the defect detection of products on the production line more efficient and accurate.

[0003] After the detection equipment is put into operation, a large number of defect images will be generated through continuous processes such as feeding, image acquisition, and random inspection and re-inspection of detection results. Although currently, some simple statistical analyses are also carried out manually on these defect images, such as simple statistics of the quantity distribution of each defect category, the area, aspect ratio, and average gray value of the defects; or simple statistics based on the meta-information of the data, such as the factory area, product model, date, etc., the depth of these statistical analyses is insufficient, and it is difficult to intuitively see the correlation between different defect images. Summary of the Invention

[0004] In view of this, this application provides a visualization method and device for industrial defect images, a storage medium, and a computer device. By obtaining the image feature vectors corresponding to the defect images, the high-dimensional image feature vectors are reduced to a low-dimensional space, and the data points and thumbnail images corresponding to the defect images are displayed based on the image feature vectors in the low-dimensional space. Subsequently, statistical analyses are carried out on the defect images according to these data points. Since the image feature vectors include rich semantic information, the depth of the statistical analysis is increased. In addition, after reducing the high-dimensional image feature vectors to two or three dimensions for display, and the image feature vectors can well ensure that similar defects are closer in the feature space and different defects are farther apart, so that complex data relationships become visual and understandable, facilitating users to intuitively see the correlation between defect images, thereby helping the staff better understand and utilize the data.

[0005] According to one aspect of this application, a visualization method for industrial defect images is provided, including:

[0006] In response to a defect image display instruction, determine a plurality of defect images to be displayed and a target display dimension;

[0007] For each of the defect images, obtain the image feature vector corresponding to the defect image, perform dimensionality reduction processing on the image feature vector according to the target display dimension, and determine the coordinate position of the data point corresponding to the defect image based on the image feature vector after dimensionality reduction processing;

[0008] For each data point of the defect image, based on the coordinate position of the data point, determine the first display position of the data point on the image display area, and according to the first display position, display the defect image on the image display area in the form of the data point and a thumbnail image of the defect image.

[0009] According to another aspect of the present application, there is provided a visualization device for industrial defect images, including:

[0010] An instruction response module, configured to respond to a defect image display instruction, and determine a plurality of defect images to be displayed and a target display dimension;

[0011] A dimensionality reduction processing module, configured to, for each of the defect images, obtain an image feature vector corresponding to the defect image, perform dimensionality reduction processing on the image feature vector according to the target display dimension, and determine the coordinate position of the data point corresponding to the defect image according to the image feature vector after dimensionality reduction processing;

[0012] An image display module, configured to, for each data point of the defect image, based on the coordinate position of the data point, determine the first display position of the data point on the image display area, and according to the first display position, display the defect image on the image display area in the form of the data point and a thumbnail image of the defect image.

[0013] According to yet another aspect of the present application, there is provided a storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned visualization method for industrial defect images is implemented.

[0014] According to still another aspect of the present application, there is provided a computer device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, and when the processor executes the program, the above-mentioned visualization method for industrial defect images is implemented.

[0015] With the above technical solution, a visualization method and device for industrial defect images, a storage medium, and a computer device provided by the present application can determine which defect images the user wants to display in response to a defect image display instruction. At the same time, it can also determine the target display dimensions corresponding to these defect images. For each defect image, an image feature vector corresponding to this defect image can be obtained. The image feature vector is high-dimensional. In order to be able to implement visualization based on the image feature vector subsequently, these high-dimensional image feature vectors can be dimensionally reduced according to the target display dimension, so that the dimension of the dimensionally reduced image feature vector is consistent with the target display dimension. After dimensional reduction processing, each defect image corresponds to a data point, and the coordinate position of this data point can be represented by the dimensionally reduced image feature vector. Next, each data point is displayed on the image display area. Specifically, the first display position of the data point on the image display area can be determined according to the coordinate position of each data point. Subsequently, each data point is displayed on the image display area according to the corresponding first display position of each data point. In addition, a thumbnail image of the corresponding defect image can be displayed based on each data point, so that the user can intuitively identify the defect image represented by each data point. In the embodiment of the present application, by obtaining the image feature vector corresponding to the defect image, the high-dimensional image feature vector is dimensionally reduced to a low-dimensional space, and the data points and thumbnail images corresponding to the defect image are displayed based on the image feature vector in the low-dimensional space. Subsequently, statistical analysis is performed on the defect images according to these data points. Since the image feature vector includes rich semantic information, the depth of statistical analysis is increased. In addition, the high-dimensional image feature vector is dimensionally reduced to two-dimensional or three-dimensional and then displayed. The image feature vector can well ensure that similar defects are closer in the feature space and different defects are farther apart, so that complex data relationships become visual and understandable, facilitating the user to intuitively see the correlation relationships between defect images, thereby helping the staff better understand and utilize the data.

[0016] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the specific embodiments of the present application are hereinafter specifically exemplified. Brief Description of the Drawings

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0018] Figure 1 A flowchart showing a visualization method for industrial defect images provided by an embodiment of the present application is shown;

[0019] Figure 2 Shows a schematic flowchart of another method for visualizing industrial defect images provided by an embodiment of the present application;

[0020] Figure 3 Shows a schematic structural diagram of a device for visualizing industrial defect images provided by an embodiment of the present application;

[0021] Figure 4 Shows a schematic structural diagram of a device of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0022] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other.

[0023] In this embodiment, a method for visualizing industrial defect images is provided. As Figure 1 shown, the method includes:

[0024] Step 101, in response to a defect image display instruction, determine a plurality of defect images to be displayed and a target display dimension.

[0025] A visualization method for industrial defect images provided by an embodiment of the present application can display multiple defect images in an intuitive and easy-to-understand manner, and at the same time facilitate viewing which defect images correspond to the same type of defect. When a user wants to display a defect image on the interface, a defect image display instruction can be generated by triggering a defect image display icon. Then, in response to the defect image display instruction, it can be determined which defect images the user wants to display. At the same time, the target display dimension corresponding to these defect images can also be determined. Among them, the defect images to be displayed can be determined based on user selection. For example, the defect type, generation time period, etc. of the defect images to be displayed are carried in the defect image display instruction. For example, if the defect type of the defect image to be displayed carried in the defect image display instruction is defect A, then the defect images marked as defect A are selected from the existing defect images as the defect images to be displayed; if the generation time period of the defect images to be displayed carried in the defect image display instruction is t1 to t2, then the defect images with generation times between t1 and t2 are selected from the existing defect images as the defect images to be displayed. In addition, when the user does not make any requirements, all existing defect images can be default selected as the defect images to be displayed, or the user can also select the defect images of all defect types as the defect images to be displayed. The target display dimension can also be determined based on user selection, specifically it can be two-dimensional or three-dimensional. When the user selects two-dimensional, the final display is in a two-dimensional manner; when the user selects three-dimensional, the final display is in a three-dimensional manner. In addition, when the user does not make any requirements, two-dimensional or three-dimensional can also be default selected as the target display dimension.

[0026] Step 102, for each of the defect images, obtain the image feature vector corresponding to the defect image, perform dimensionality reduction processing on the image feature vector according to the target display dimension, and determine the coordinate position of the data point corresponding to the defect image based on the image feature vector after dimensionality reduction processing.

[0027] In this embodiment, for each defective image, an image feature vector corresponding to this defective image can be obtained. Here, the image feature vector can be extracted through image processing or machine learning algorithms, and specifically can include multiple dimensions (for example, quantization values of features such as color, texture, shape, etc.). The image feature vector is high-dimensional. In order to be able to achieve visualization based on the image feature vector subsequently, these high-dimensional image feature vectors can be dimensionally reduced according to the target display dimension, so that the dimension of the dimensionally reduced image feature vector is consistent with the target display dimension. For example, if the target display dimension is two-dimensional, then the dimensionally reduced image feature vector is also two-dimensional. Among them, the dimensionality reduction process can be achieved through various methods, such as principal component analysis (PCA), t-SNE (t-Distributed Stochastic Neighbor Embedding), UMAP (Uniform Manifold Approximation and Projection), etc. These methods can reduce the complexity of the data while retaining important information in the original data.

[0028] After dimensionality reduction, each defective image corresponds to a data point. Assuming the target display dimension is two-dimensional, then after reducing the image feature vector to two dimensions, the dimensionally reduced image feature vector can be used as the data point corresponding to the defective image. Specifically, one element in the dimensionally reduced image feature vector can be used as the abscissa of the data point, and the other element can be used as the ordinate of the data point. For example, if the dimensionally reduced image feature vector is (x, y), then (x, y) can be directly used as the data point corresponding to the defective image, and (x, y) is also the coordinate position corresponding to this data point.

[0029] Step 103, for the data point of each defective image, based on the coordinate position of the data point, determine the first display position of the data point on the image display area, and according to the first display position, display the defective image on the image display area in the form of the data point and the thumbnail image of the defective image.

[0030] In this embodiment, next, each data point is displayed on the image display area. Specifically, the first display position of each data point on the image display area can be determined according to the coordinate position of each data point. Among them, the first display position can be directly determined based on the coordinate position of the data point, or can be determined by calculating according to the coordinate position of the data point through a certain algorithm. Subsequently, each data point is displayed on the image display area according to the first display position corresponding to each data point. In addition, the thumbnail image of the corresponding defective image can also be displayed based on each data point, so that the user can intuitively identify the defective image represented by each data point.

[0031] By applying the technical solution of this embodiment, in response to a defect image display instruction, it is determined which defect images the user wants to display. At the same time, the target display dimension corresponding to these defect images can also be determined. For each defect image, the image feature vector corresponding to this defect image can be obtained. The image feature vector is high-dimensional. In order to be able to realize visualization based on the image feature vector subsequently, the high-dimensional image feature vectors can be dimensionally reduced according to the target display dimension, so that the dimension of the dimensionally reduced image feature vector is consistent with the target display dimension. After dimensional reduction processing, each defect image corresponds to a data point, and the coordinate position of this data point can be represented by the dimensionally reduced image feature vector. Next, each data point is displayed on the image display area. Specifically, the first display position of the data point on the image display area can be determined according to the coordinate position of each data point. Subsequently, each data point is displayed on the image display area according to the corresponding first display position of each data point. In addition, a thumbnail image of the corresponding defect image can be displayed based on each data point, so that the user can intuitively identify the defect image represented by each data point. In the embodiment of the present application, by obtaining the image feature vector corresponding to the defect image, the high-dimensional image feature vector is dimensionally reduced to a low-dimensional space, and the data points and thumbnail images corresponding to the defect image are displayed based on the image feature vector in the low-dimensional space. Subsequently, statistical analysis is performed on the defect images according to these data points. Since the image feature vector includes rich semantic information, the depth of statistical analysis is increased. In addition, after the high-dimensional image feature vector is dimensionally reduced to two or three dimensions and displayed, and the image feature vector can well ensure that similar defects are closer in the feature space and different defects are farther apart, so that complex data relationships become visual and understandable, facilitating the user to intuitively see the correlation relationships between defect images, thereby helping the staff better understand and utilize the data.

[0032] In the embodiment of the present application, optionally, before step 101, the method further includes: for each of the defect images, cropping the defect image according to a preset cropping rule to obtain a first processed image centered on the defect area; based on the image attribute information of the first processed image, determining the target processing method corresponding to the first processed image, and reprocessing the first processed image according to the target processing method to obtain a second processed image, where when the image attribute information includes the image size, the target processing method includes image scaling, and / or when the image attribute information includes the storage space occupied, the target processing method includes image compression; annotating the defect area and the defect type in the second processed image to obtain the thumbnail image corresponding to the defect image.

[0033] In this embodiment, before displaying the defect image in the form of data points and thumbnail images, the thumbnail image can be generated first. First, for each defect image, it can be cropped according to a preset cropping rule. Here, the preset cropping rule can be a rule for cropping based on the position of the defect to ensure that the cropped image (i.e., the first processed image) is a regular-shaped image centered on the defect area. For example, the first processed image is a 512*512-sized image cropped centered on the defect area. It should be noted that the defect area may be an irregular area. Therefore, being centered on the defect area here can be the center determined based on the approximate position of the defect area or the center determined manually.

[0034] After that, based on the image attribute information of the first processed image, the corresponding target processing method is determined and reprocessed to obtain the second processed image. Among them, the image attribute information can include the size of the image, the storage space occupied, etc. Since the sizes and storage spaces occupied by different defect images are different, and the sizes of the defect areas included in them are also different, the sizes and storage spaces occupied by the cropped first processed images are actually different. For the convenience of subsequent processing, further, the sizes of these first processed images can be adjusted to be the same. For example, when the image size of the first processed image does not meet the specific size requirements (such as being too large or too small), the target processing method can include image scaling to adjust the image size to an appropriate range; when the image size of the first processed image meets the specific size requirements, the target processing method can be empty. At the same time, to save storage and transmission costs, further, the storage spaces occupied by these first processed images can be adjusted to an appropriate size. For example, when the storage space occupied by the first processed image is too large, the target processing method can include image compression to reduce the storage space occupied by the image.

[0035] After obtaining the second processed image, the defect area and defect type in the second processed image can be marked. The marking can include information such as the contour, position, and size of the defect area, as well as the text or symbol identification of the defect type. After the above cropping, reprocessing, and marking steps, the obtained image is the thumbnail image corresponding to the defect image. By cropping the defect image in the embodiment of the present application, unnecessary background information in the defect image can be removed, making the defect area more prominent and facilitating subsequent analysis and marking; by reprocessing, it can be ensured that the first processed image meets specific requirements in terms of size and storage space, facilitating subsequent storage, transmission, and display; by marking, the defect area and type can be clearly shown on the thumbnail image, facilitating subsequent quick browsing, identification, and analysis of the defect image.

[0036] In an embodiment of the present application, optionally, after "obtaining the thumbnail image corresponding to the defect image", the method further includes: for each of the thumbnail images, obtaining a target defect region according to the defect region marked in the thumbnail image, and inputting the target defect region into a pre-trained feature extraction model to obtain an image feature vector corresponding to the defect image.

[0037] In this embodiment, an image feature vector can also be obtained according to the thumbnail image. Specifically, for each thumbnail image, first, a target defect region can be obtained based on the defect region marked in the thumbnail image. Among them, when marking the defect region in the thumbnail image, methods such as segmentation contour marking and detection box marking can be used. For different marking methods, different target defect regions can be obtained. Subsequently, the obtained target defect region is input into a pre-trained feature extraction model. Through the feature extraction model, an image feature vector can be extracted from the target defect region. The image feature vector can be a vector containing multiple values, and each value represents the quantization value of a certain feature in the image, such as color, texture, shape, edge information, etc. Here, the feature extraction model can be pre-trained models such as CLIP (Contrastive Language-Image Pre-training) and DinoV2 (Self-Supervised Vision Transformers). These models are trained through unsupervised or self-supervised learning methods and do not rely on manually labeled samples. Through large-scale pre-training, these models can extract high-level semantic features of the defect image. It should be noted that the feature extraction model is not required here, as long as it can extract an image feature vector from the target defect region. By using the target defect region as the input of the feature extraction model to obtain the corresponding image feature vector in the embodiment of the present application, the interference of background information during the extraction of the image feature vector can be reduced, and the accuracy of feature extraction can be improved.

[0038] In an embodiment of the present application, optionally, "obtaining a target defect region according to the defect region marked in the thumbnail image" includes: determining the defect marking method corresponding to the defect region marked in the thumbnail image; if the defect marking method is segmentation contour marking, extracting the circumscribed rectangle region corresponding to the segmentation contour and using the circumscribed rectangle region as the target defect region; if the defect marking method is detection box marking, using the detection box as the target defect region.

[0039] In this embodiment, for different defect annotation methods, the method for determining the target defect area is also different. Common existing annotation methods include segmentation contour annotation and detection box annotation. Among them, segmentation contour annotation refers to depicting the precise boundary of a defect to form a closed contour; while detection box annotation is to roughly enclose the defect with a rectangular box.

[0040] If the defect annotation method is segmentation contour annotation, then the previously annotated defect segmentation contour can be extracted from the thumbnail image. Then, the corresponding circumscribed rectangular area is calculated based on the segmentation contour. The circumscribed rectangular area refers to the smallest rectangle that can completely contain the segmentation contour. Finally, this circumscribed rectangular area is used as the target defect area. The reason for doing this is that although the segmentation contour provides the precise boundary of the defect, the circumscribed rectangle is easier to process and calculate, and can also roughly contain all the defects. By extracting the circumscribed rectangular area of the segmentation contour in the embodiment of the present application, the subsequent processing steps can be simplified while ensuring that all defects are completely contained.

[0041] If the defect annotation method is detection box annotation, at this time, the previously annotated detection box can be directly extracted from the thumbnail image, and this detection box is used as the target defect area. Since the detection box itself is used to roughly enclose the defect, it can be directly used as the target defect area for subsequent processing without further calculation or conversion.

[0042] In the embodiment of the present application, optionally, when the target display dimension is two-dimensional, in step 103, the “determining the first display position of the data point on the image display area based on the coordinate position of the data point” includes: setting the coordinate position of the data point as the coordinate positions of the first coordinate axis and the second coordinate axis, and obtaining a preset value and a random adjustment value, and determining the coordinate position of the third coordinate axis of the data point according to the preset value and the random adjustment value; determining the three-dimensional coordinates of the data point according to the coordinate position of the first coordinate axis, the coordinate position of the second coordinate axis, and the coordinate position of the third coordinate axis, marking the data point in the default three-dimensional coordinate system according to the three-dimensional coordinates, and projecting the data points in the default three-dimensional coordinate system according to a preset projection angle to obtain the projection coordinates of the data point; using the projection coordinates of the data point as the first display position of the data point on the image display area; correspondingly, in step 103, the “displaying the defect image in the form of the data point and the thumbnail image of the defect image on the image display area according to the first display position” includes: determining the first display position on the image display area, displaying the data point at the first display position, and using the data point as a reference point to display the thumbnail image according to the reference point.

[0043] In this embodiment, each defect image is represented as a data point in a low-dimensional space (two-dimensional or three-dimensional), and a thumbnail image is displayed based on each data point. When the target display dimension is two-dimensional, thumbnail images corresponding to relatively close data points may overlap with each other. Therefore, the first display position can be determined in the following manner to avoid overlap between different thumbnail images in two-dimensional display as much as possible.

[0044] First, the coordinate position of the data point is set to the coordinate position of the first coordinate axis (such as the X axis) and the coordinate position of the second coordinate axis (such as the Y axis). That is, the two values ​​in the image feature vector after dimensionality reduction corresponding to the data point are used as the coordinate values ​​of the two coordinate axes respectively. Next, a preset value is obtained, which represents a reference height or depth and is used to locate the data point on the third coordinate axis (such as the Z axis). For example, the preset value can be 1. At the same time, a random adjustment value is obtained, which is used to make a slight random adjustment to the data based on the preset value to increase the distribution diversity of the data point in the three-dimensional space. Subsequently, the coordinate position of the data point on the third coordinate axis is calculated based on the preset value and the random adjustment value. For example, the value obtained by subtracting the random adjustment value from the preset value can be used as the coordinate position of the data point on the third coordinate axis. At this point, the data point already has a complete three-dimensional coordinate (X, Y, Z), and the data point can be marked in a default three-dimensional coordinate system. This default three-dimensional coordinate system can be a virtual space used for projection calculation.

[0045] Further, according to the preset projection angle, the data points in the default three-dimensional coordinate system are projected, and the projection result is a two-dimensional coordinate, which is used as the first display position of the data point on the image display area. The embodiment of the present application can increase the distribution diversity of data points in three-dimensional space by introducing preset values ​​and random adjustment values, thereby effectively reducing the overlap between thumbnail images when projected onto a two-dimensional plane.

[0046] It should be noted that if the target display dimension is three-dimensional, the coordinate position of the data point can be directly used as the first display position of the data point on the image display area.

[0047] After determining the first display position of each data point, the data point can be displayed at the first display position, and the displayed data point is used as a reference point to display a thumbnail image according to the reference point. For example, the upper left corner or lower right corner of the thumbnail image is overlapped with the reference point to achieve the positioning of the thumbnail image in the image display area.

[0048] In an embodiment of the present application, optionally, before step 101, the method further includes: displaying a defect image analysis interface, where a defect image type determination area and a defect image display dimension determination area are displayed in the defect image analysis interface; step 101 includes: in response to a defect image display instruction, identifying a target type from the defect image type determination area, and identifying a target display dimension from the defect image display dimension determination area; obtaining defect images with the marked defect type being the target type from a target storage space as a plurality of defect images to be displayed.

[0049] In this embodiment, before displaying the data points and thumbnail images corresponding to the defect images, the user can first log in to the visualization analysis system. After logging in, the defect image analysis interface can be displayed. This interface is a window for the user to interact with the visualization analysis system. The defect image analysis interface may include a defect image type determination area and a defect image display dimension determination area. Among them, the defect image type determination area allows the user to select or input the defect image type they are interested in. This can be achieved through methods such as drop-down menus, check boxes, radio buttons, or text inputs. The user can select a specific defect type, such as scratches, stains, cracks, etc., according to their analysis needs or concerns. The defect image display dimension determination area allows the user to specify the display dimension of the defect image, such as two-dimensional, three-dimensional, etc., which helps the user to customize the way of image display according to their display needs.

[0050] When the user finishes inputting or selecting through the interface, a defect image analysis instruction can be generated. Then, the visualization analysis system can, in response to this defect image display instruction, identify the target type selected by the user from the defect image type determination area and identify the target display dimension specified by the user from the defect image display dimension determination area.

[0051] The visualization analysis system corresponds to a target storage space, in which a large number of defect images marked with defect types are stored. After determining the target type, defect images matching the target type can be screened and obtained from the target storage space, and these defect images are used as the defect images to be displayed. The embodiment of the present application provides a user-friendly visualization interaction interface, allowing the user to select and display specific defect images according to their needs, with relatively high flexibility.

[0052] In an embodiment of the present application, optionally, the defect image analysis interface further includes the image display area; after step 103, the method further includes: based on the received target operation instruction for the image display area, adjusting the display view of the image display area, and displaying the adjusted view in the image display area, where the target operation instruction is one of a zoom-in operation instruction, a zoom-out operation instruction, a pan operation instruction, and a rotation operation instruction; and / or, when it is detected that the mouse hovers over any data point in the image display area, magnifying and displaying the thumbnail image corresponding to the any data point; and / or, based on the received click instruction for any data point in the image display area, displaying the defect image corresponding to the any data point and the marked defect type in the defect image analysis interface.

[0053] In this embodiment, the defect image analysis interface of the above visualization interaction system may further include an image display area for displaying data points and thumbnail images. After the data points and thumbnail images that the user wants to view are displayed in the image display area, the user can further interact with the visualization interaction system.

[0054] First, adjustment of the display view in the image display area. Based on the received target operation instruction for the image display area, the visualization interaction system can adjust the display view of the image display area. These target operation instructions can be one of a zoom-in operation instruction, a zoom-out operation instruction, a pan operation instruction, and a rotation operation instruction. After the above adjustment is completed, the adjusted view can be displayed in the image display area for the user to view. In this way, the user can flexibly adjust the display content of the image display area, which is beneficial to improving the user experience.

[0055] Second, magnification on data point hover. When the visualization interaction system detects that the mouse hovers over any data point in the image display area, it can automatically magnify and display the thumbnail image corresponding to the data point. This can improve the convenience and intuitiveness of user interaction. The user does not need to perform additional clicks or operations. Just hovering the mouse over the data point can view more detailed thumbnail image information.

[0056] Third, display of detailed information on data point click. Based on the received click instruction for any data point in the image display area, the visualization interaction system can display the defect image corresponding to the data point and the marked defect type in the defect image analysis interface. This function provides users with a more in-depth data analysis and viewing function. The user only needs to click on the data point to view the complete defect image and defect type information corresponding to the data point, which helps the user better understand the relationship between the data point and the defect image.

[0057] By adding an interactive function to the image display area in the embodiments of the present application, the user experience and data analysis efficiency are improved. The user can adjust the size, position, and orientation of the display view according to needs to better view and analyze the thumbnail images of data points and defect images. Meanwhile, by hovering over and clicking on the data points, the user can quickly view more detailed thumbnail image information, defect images, and their type information.

[0058] In the embodiments of the present application, optionally, a defect type icon corresponding to the thumbnail image is displayed at a fixed position of each thumbnail image; after step 103, the method further includes: performing image recognition on the data points displayed in the image display area to obtain a plurality of data point clusters; for each data point cluster, identifying the defect type marked on the thumbnail image corresponding to each data point in the data point cluster, taking the defect type with the largest quantity in the data point cluster as the target defect type, and determining whether there is an abnormal defect type different from the target defect type among the defect types corresponding to the thumbnail images in the data point cluster. When there is such an abnormal defect type, highlighting the defect type icon of the thumbnail image corresponding to the abnormal defect type.

[0059] In this embodiment, a defect type icon corresponding to each thumbnail image can also be displayed at a fixed position of the thumbnail image, which can visually tell the user the defect type corresponding to the thumbnail image. Specifically, the visual interaction system can store the correspondence between the defect type and the defect type icon through a database or a mapping table. When the thumbnail image is loaded into the image display area, the visual interaction system can find and display the corresponding defect type icon on the thumbnail image according to this correspondence. Among them, for the convenience of the user to distinguish, different defect type icons can also be set for different defect types, such as different display colors, different display patterns, etc. In this way, when the user views the thumbnail images displayed in the image display area, the defect type marked on each thumbnail image can be quickly recognized.

[0060] After the data points and thumbnail images are displayed in the image display area, the visual interaction system can also use computer vision technology to perform image recognition on the data points displayed in the image display area to identify which data points belong to a data point cluster. In addition, for each data point cluster, the defect type marked on the thumbnail image corresponding to each data point therein can be identified, and the defect type with the largest quantity is found as the target defect type of the data point cluster. Then, it is statistically determined whether there is a defect type different from the target defect type in the data point cluster, and this defect type is called an abnormal defect type. If there is an abnormal defect type, then highlighting the defect type icons of the thumbnail images corresponding to these abnormal defect types (such as changing the color, adding a border, etc.).

[0061] In fact, assuming that the defective image is a sample image for training a defect detection model, when manually annotating the defect type of the sample image, incorrect annotation may occur during the annotation of the defect type corresponding to each thumbnail image. Such errors directly mislead the subsequent training of the defect detection model, resulting in a decline in the performance of the model. In the prior art, manual browsing and checking for annotation are inefficient and still prone to missing some incorrect annotations. In the embodiments of the present application, through the visualization of the low-dimensional space, if there is an obviously abnormal (i.e., prominently displayed) defect type icon in a cluster of data points, it indicates that the defect type annotation of the thumbnail image is incorrect. At this time, it is easy for humans to discover these abnormal defect types, and then directly check the annotation situation of the thumbnail images of these abnormal defect types. On the basis of improving the verification efficiency of the sample image, the verification accuracy can also be increased. In the embodiments of the present application, by observing the data distribution in the low-dimensional space, the clustering characteristics of various defects can be intuitively analyzed to assist in the detection of defect type classification and abnormal annotation. It should be noted that since the positions of the data points in the image display area are determined according to the image feature vectors of the defective images, the data points corresponding to the defective images of the same defect type are often close in the image display area and belong to the same data point cluster. This feature can assist the staff in quickly identifying the pattern samples with incorrect annotations.

[0062] In the embodiments of the present application, optionally, when the multiple defective images to be displayed are all the defective images in the target storage space, after step 103, the method further includes: obtaining the undetected images, and determining the thumbnail images and image feature vectors corresponding to the undetected images, where the undetected images are the images in which defects are not recognized by the trained defect detection model, and the defect detection model is trained based on all the defective images in the target storage space; determining the second display position of the data points corresponding to the undetected images on the image display area according to the image feature vectors, dynamically adding the data points and thumbnail images corresponding to the undetected images to the image display area based on the second display position, and adding new embedding marks.

[0063] In this embodiment, the user can select the defect type. Then, the defect images to be displayed are all the defect images of this defect type in the target storage space. If the user defaults to select "all types", then the defect images to be displayed are all the defect images in the target storage space. In the case where the defect images are sample images for training a defect detection model, if the data points and thumbnail images corresponding to all the defect images in the target storage space are displayed in the current image display area, it means that the data points and thumbnail images corresponding to all the image samples are displayed in the image display area at this time. At this time, the missed detection images can be obtained to check the reasons why the defects in these missed detection images are not recognized by the defect detection model. Here, a missed detection image is an image that is not recognized as having a defect by the defect detection model when applying the trained defect detection model for defect recognition. The defect detection model is trained based on all the defect images in the target storage space.

[0064] For the missed detection images, by obtaining the thumbnail images and image feature vectors of the aforementioned existing defect images, the thumbnail image and image feature vector corresponding to the missed detection image are determined. Then, according to the image feature vector of the missed detection image, the second display position of the data point corresponding to the missed detection image on the current image display area is determined. The method for determining the second display position is the same as the method for determining the first display position of the data points of the aforementioned defect images. Subsequently, the data points and thumbnail images corresponding to the missed detection image are dynamically added to the image display area. At the same time, a new embedding mark can also be added. This mark can be an icon, a color mark, or other visual indications used to distinguish the newly added missed detection images from the previously displayed defect images.

[0065] In fact, during the training and debugging of the defect detection model, some missed detection situations may occur, that is, defects not detected by the model. In the embodiment of the present application, by embedding new missed detection images into the existing image display area, the relationship between the new missed detection images and the original defect images can be intuitively analyzed. If there are few similar original defect images around the new missed detection image, it indicates that there are few image samples similar to the new missed detection image during the training of the defect detection model. If there are similar original defect images around the new missed detection image, it may indicate problems such as model training and parameter settings. Through these visualization interaction systems, it can help users analyze potential reasons for missed detection and subsequent improvement directions.

[0066] In an embodiment of the present application, optionally, after step 103, the method further includes: in response to a data export instruction, obtaining a first display position, an image feature vector, and a thumbnail image corresponding to each data point displayed on the image display area, as well as a target export format, and generating a target export file according to the target export format based on the first display position, the image feature vector, and the thumbnail image of each data point on the image display area; and / or generating a defect image display report based on the display result of the current image display area, and storing the defect image display report according to a preset storage path.

[0067] In this embodiment, the user can export the information currently displayed on the image display area to an external file or system for further analysis, archiving, or sharing. Therefore, the visual interaction system can respond to a data export instruction triggered by the user. This data export instruction can be triggered by the user by clicking a button, selecting a menu item, or performing other interaction operations. To generate the export file, further, the visual interaction system can obtain the first display position, the image feature vector, and the thumbnail image corresponding to each data point displayed on the image display area. These pieces of information together constitute the content of the export file. At the same time, the target export format desired by the user can also be obtained, such as common data formats like CSV, Excel, JSON, XML, etc., depending on the user's needs. After that, based on the obtained first display position, image feature vector, and thumbnail image of each data point, an export file is generated according to the target export format specified by the user. This file will contain the relevant information of all the data points displayed on the image display area and be organized in the format expected by the user.

[0068] In addition to data export, the visual interaction system can also automatically generate a report containing the display result of the current image display area each time. This report can be a document in formats such as PDF, HTML, or others, used to display the defect image and its related information. After that, the generated defect image display report is stored according to a preset storage path. This path can be a directory in the local file system or a location on the network, depending on the system configuration and user preferences.

[0069] The visual interaction system in the embodiment of the present application can respond to different user needs, support both data export and report generation, and support multiple export formats and report formats; the user can trigger the data export or report generation function through simple interaction operations without complex settings or configurations, enhancing the practicality and user experience of the system.

[0070] Further, as a refinement and extension of the specific implementation manner of the above embodiment, to fully illustrate the specific implementation process of this embodiment, a defect image analysis interface of a visual interaction system is provided, asFigure 2 As shown in, among which, Figure 2 The upper left corner is the function area, including a defect image type determination area (i.e., the "category" box in the figure) and a defect image display dimension determination area (i.e., the "dimension" box in the figure). In addition, it also includes a dimensionality reduction method determination area (i.e., the "method" box). Each area can correspond to a drop-down menu, and the user can click on the drop-down menu and select a dimensionality reduction method, display dimension, and defect category. Among them, dimensionality reduction methods can adopt non-linear dimensionality reduction techniques such as UMAP and t-SNE. These methods can map high-dimensional features to two-dimensional or three-dimensional space while preserving the local structure of the data, and other dimensionality reduction methods can also be adopted. In addition, the function area can also include a file name search box, and the user can search according to the file name to quickly display relevant information such as the defect image corresponding to the file name.

[0071] The lower left corner is the detailed information display area and the entire defect display area. Among them, the detailed information display area can be used to display the defect type corresponding to any data point, and the entire defect display area can be used to display the original defect image corresponding to any data point.

[0072] The right side is the main display area, that is, the image display area, which can be used to display the data point distribution of the defect image after dimensionality reduction. The position is the first display position determined according to the coordinates after dimensionality reduction. Each square is a thumbnail image, and different defect categories are displayed in the upper left corner of each thumbnail image through different colors or textures ( Figure 2 textures are used in). The user can perform operations such as zooming, rotating, panning, hovering for preview, and clicking to view details in this area. When the mouse hovers over a certain data point, the thumbnail image corresponding to that data point can be enlarged and displayed. When the user clicks on a certain data point, the detailed information display area in the lower left corner can display detailed information such as the defect type corresponding to that data point, and the entire defect display area can display the original defect image corresponding to that data point. Thumbnail image A in the image display area illustrates the function of this visual interaction system for mislabeling analysis. If the defect type icons of thumbnail image A and the surrounding similar thumbnail images are inconsistent, it is possible that the defect type of thumbnail image A is mislabeled. When the user discovers this situation, they can click on the data point corresponding to thumbnail image A to view the detailed information and further check and correct the label. Thumbnail image B illustrates the function of this visual interaction system for missed detection cause analysis. This figure is a visualization of the missed detection thumbnail image embedded in the training set. From Figure 2It can be seen that there are no similar samples around the thumbnail image B, indicating that there are no similar samples in the training samples. Through this interactive visualization, users can be helped to analyze potential reasons for missed detections and subsequent improvement directions. In the embodiment of the present application, the image feature vectors after dimensionality reduction are displayed in a two-dimensional or three-dimensional space through a visualization interaction system. At the same time, clicking on a certain data point in the low-dimensional space can display its detailed information, including the original defective image, defect type, etc., which can provide users with a comprehensive analysis perspective.

[0073] Further, as Figure 1 a specific implementation of the method, the embodiment of the present application provides a visualization device for industrial defect images, as Figure 3 shown. The device includes:

[0074] An instruction response module, configured to respond to a defective image display instruction, and determine a plurality of defective images to be displayed and a target display dimension;

[0075] A dimensionality reduction processing module, configured to, for each of the defective images, obtain an image feature vector corresponding to the defective image, perform dimensionality reduction processing on the image feature vector according to the target display dimension, and determine the coordinate position of the data point corresponding to the defective image according to the image feature vector after dimensionality reduction processing;

[0076] An image display module, configured to, for the data point of each defective image, determine a first display position of the data point on the image display area based on the coordinate position of the data point, and display the defective image in the form of the data point and a thumbnail image of the defective image on the image display area according to the first display position.

[0077] Optionally, the device further includes a thumbnail image generation module; the thumbnail image generation module is configured to:

[0078] Before responding to the defective image display instruction and determining a plurality of defective images to be displayed and a target display dimension, for each of the defective images, crop the defective image according to a preset cropping rule to obtain a first processed image centered on the defective area;

[0079] Based on the image attribute information of the first processed image, determine a target processing method corresponding to the first processed image, and perform reprocessing on the first processed image according to the target processing method to obtain a second processed image, where when the image attribute information includes the image size, the target processing method includes image scaling, and / or when the image attribute information includes the storage space occupied, the target processing method includes image compression;

[0080] Label the defect area and defect type in the second processed image to obtain a thumbnail image corresponding to the defect image.

[0081] Optionally, the device further includes an image feature vector generation module; the image feature vector generation module is used for:

[0082] After obtaining the thumbnail image corresponding to the defect image, for each thumbnail image, obtain a target defect area according to the defect area marked in the thumbnail image, and input the target defect area into a pre-trained feature extraction model to obtain an image feature vector corresponding to the defect image.

[0083] Optionally, the image feature vector generation module is used for:

[0084] Determine the defect annotation method corresponding to the defect area marked in the thumbnail image;

[0085] If the defect annotation method is segmentation contour annotation, extract the circumscribed rectangle area corresponding to the segmentation contour, and use the circumscribed rectangle area as the target defect area;

[0086] If the defect annotation method is detection box annotation, use the detection box as the target defect area.

[0087] Optionally, when the target display dimension is two-dimensional, the image display module is used for:

[0088] Set the coordinate position of the data point to the coordinate positions of the first coordinate axis and the second coordinate axis, and obtain a preset value and a random adjustment value. According to the preset value and the random adjustment value, determine the coordinate position of the third coordinate axis of the data point;

[0089] According to the coordinate position of the first coordinate axis, the coordinate position of the second coordinate axis, and the coordinate position of the third coordinate axis, determine the three-dimensional coordinates of the data point, mark the data point in the default three-dimensional coordinate system according to the three-dimensional coordinates, and project the data points in the default three-dimensional coordinate system according to a preset projection angle to obtain the projection coordinates of the data point;

[0090] Use the projection coordinates of the data point as the first display position of the data point on the image display area;

[0091] Correspondingly, the image display module is further used for:

[0092] Determine the first display position on the image display area, display the data point at the first display position, and use the data point as a reference point to display the thumbnail image according to the reference point.

[0093] Optionally, the device further includes an interface display module; the interface display module is configured to:

[0094] Before determining a plurality of defect images to be displayed and a target display dimension in response to a defect image display instruction, display a defect image analysis interface, wherein a defect image type determination area and a defect image display dimension determination area are shown in the defect image analysis interface;

[0095] The instruction response module is configured to:

[0096] In response to a defect image display instruction, identify a target type from the defect image type determination area and identify a target display dimension from the defect image display dimension determination area;

[0097] Obtain, from a target storage space, defect images with the marked defect type being the target type as the plurality of defect images to be displayed.

[0098] Optionally, the defect image analysis interface further includes the image display area; the device further includes an image display area operation module; the image display area operation module is configured to:

[0099] After displaying the defect images in the form of the data points and thumbnail images of the defect images on the image display area, based on a received target operation instruction for the image display area, adjust a display view of the image display area and display an adjusted view on the image display area, wherein the target operation instruction is one of a zoom-in operation instruction, a zoom-out operation instruction, a pan operation instruction, and a rotation operation instruction; and / or when it is monitored that the mouse hovers over any one of the data points in the image display area, magnify and display the thumbnail image corresponding to the any one of the data points; and / or based on a received click instruction for any one of the data points in the image display area, display, in the defect image analysis interface, the defect image corresponding to the any one of the data points and the marked defect type.

[0100] Optionally, a defect type icon corresponding to the thumbnail image is displayed at a fixed position of each of the thumbnail images; the device further includes an anomaly analysis module; the anomaly analysis module is configured to:

[0101] After displaying the defect images in the form of the data points and thumbnail images of the defect images on the image display area, perform image recognition on the data points displayed on the image display area to obtain a plurality of data point clusters;

[0102] For each cluster of data points, identify the defect types corresponding to the thumbnail image markers of each data point in the data point cluster, use the defect type with the largest quantity in the data point cluster as the target defect type, and determine whether there are abnormal defect types different from the target defect type among the defect types corresponding to the thumbnail images in the data point cluster. When there are such abnormal defect types, highlight the defect type icons of the thumbnail images corresponding to the abnormal defect types.

[0103] Optionally, when the multiple defect images to be displayed are all the defect images in the target storage space, the device further includes a missed detection analysis module; the missed detection analysis module is configured to:

[0104] After displaying the defect images in the form of the data points and the thumbnail images of the defect images on the image display area, obtain missed detection images, and determine the thumbnail images and image feature vectors corresponding to the missed detection images, where the missed detection images are images for which defects are not recognized by a trained defect detection model, and the defect detection model is trained based on all the defect images in the target storage space;

[0105] According to the image feature vectors, determine the second display positions of the data points corresponding to the missed detection images on the image display area, and dynamically add the data points and thumbnail images corresponding to the missed detection images to the image display area based on the second display positions, and add new embedding markers.

[0106] Optionally, the device further includes a file export module; the file export module is configured to:

[0107] After displaying the defect images in the form of the data points and the thumbnail images of the defect images on the image display area, in response to a data export instruction, obtain the first display positions, image feature vectors, and thumbnail images corresponding to each data point displayed on the image display area, and a target export format, and generate a target export file according to the first display positions, image feature vectors, and thumbnail images of each data point on the image display area in accordance with the target export format; and / or,

[0108] Generate a defect image display report based on the display result of the current image display area, and store the defect image display report according to a preset storage path.

[0109] It should be noted that for other corresponding descriptions of the various functional units involved in the visualization device for industrial defect images provided in the embodiments of the present application, reference can be made to Figures 1 to 2 the corresponding descriptions in the method, which will not be elaborated here.

[0110] The embodiments of the present application also provide a computer device, which can specifically be a personal computer, a server, a network device, etc. For example, Figure 4 As shown, the computer device includes a bus, a processor, a memory, and a communication interface, and may also include an input / output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements the steps in the method embodiments.

[0111] Those skilled in the art can understand that Figure 4 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0112] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium can be non-volatile or volatile, and stores a computer program. When the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0113] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0114] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0115] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0116] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0117] The above-described embodiments merely represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for visualizing industrial defect images, characterized in that: include: In response to the defect image display instruction, determining a plurality of defect images to be displayed and a target display dimension; For each defect image, an image feature vector corresponding to the defect image is obtained, and the image feature vector is subjected to dimensionality reduction processing according to the target display dimension, and the coordinate position of the data point corresponding to the defect image is determined according to the image feature vector after dimensionality reduction processing; For each data point of the defect image, the first display position of the data point on the image display area is determined based on the coordinate position of the data point, and according to the first display position, the defect image is displayed on the image display area in the form of a thumbnail image of the data point and the defect image.

2. The method according to claim 1, characterized in that Before determining the plurality of defect images to be displayed and the target display dimensions in response to the defect image display instruction, the method further includes: For each of the defect images, cropping the defect image according to a preset cropping rule to obtain a first processed image centered on the defect area; Based on the image attribute information of the first processed image, determine a target processing mode corresponding to the first processed image, and reprocess the first processed image according to the target processing mode to obtain a second processed image, wherein when the image attribute information includes image size, the target processing mode includes image scaling, and / or when the image attribute information includes occupied storage space, the target processing mode includes image compression; The defect area and the defect type are marked in the second processed image to obtain a thumbnail image corresponding to the defect image.

3. The method according to claim 2, characterized in that After obtaining the thumbnail image corresponding to the defective image, the method further includes: For each of the thumbnail images, a target defect area is obtained according to the defect area marked in the thumbnail image, and the target defect area is input into a pre-trained feature extraction model to obtain an image feature vector corresponding to the defect image.

4. The method according to claim 3, characterized in that The step of obtaining a target defect area according to the defect area marked in the thumbnail image includes: Determining a defect marking method corresponding to the defect area marked in the thumbnail image; If the defect marking method is segmentation contour marking, extracting a circumscribed rectangular area corresponding to the segmentation contour, and using the circumscribed rectangular area as the target defect area; If the defect marking method is detection frame marking, the detection frame is used as the target defect area.

5. The method according to claim 1, characterized in that When the target display dimension is two-dimensional, determining the first display position of the data point on the image display area based on the coordinate position of the data point includes: The coordinate position of the data point is set as the coordinate position of the first coordinate axis and the coordinate position of the second coordinate axis, and a preset value and a random adjustment value are obtained, and the coordinate position of the data point on the third coordinate axis is determined according to the preset value and the random adjustment value; Determine the three-dimensional coordinates of the data point according to the coordinate position of the first coordinate axis, the coordinate position of the second coordinate axis, and the coordinate position of the third coordinate axis, mark the data point according to the three-dimensional coordinates in a default three-dimensional coordinate system, and project the data point in the default three-dimensional coordinate system according to a preset projection angle to obtain the projection coordinates of the data point; Using the projection coordinates of the data point as the first display position of the data point on the image display area; Accordingly, displaying the defect image in the form of the data point and a thumbnail image of the defect image on the image display area according to the first display position includes: The first display position is determined on the image display area, the data point is displayed at the first display position, and the data point is used as a reference point, and the thumbnail image is displayed according to the reference point.

6. The method according to claim 1, characterized in that Before determining the plurality of defect images to be displayed and the target display dimensions in response to the defect image display instruction, the method further includes: Displaying a defect image analysis interface, wherein the defect image analysis interface displays a defect image type determination area and a defect image display dimension determination area; The step of determining, in response to the defect image display instruction, a plurality of defect images to be displayed and a target display dimension comprises: In response to the defect image display instruction, determining a region recognition target type from the defect image type, and determining a region recognition target display dimension from the defect image display dimension; Defect images whose defect types are marked as the target type are acquired from the target storage space as a plurality of defect images to be displayed.

7. The method according to claim 6, characterized in that The defect image analysis interface further includes the image display area; after displaying the defect image in the form of the data points and a thumbnail image of the defect image on the image display area, the method further includes: Based on the received target operation instruction for the image display area, adjusting the display view of the image display area, and displaying the adjusted view in the image display area, wherein the target operation instruction is one of a zoom-in operation instruction, a zoom-out operation instruction, a translation operation instruction, and a rotation operation instruction; and / or, When it is detected that the mouse is hovering over any data point in the image display area, the thumbnail image corresponding to any data point is enlarged and displayed; and / or, Based on the received click instruction for any data point in the image display area, the defect image corresponding to the any data point and the marked defect type are displayed in the defect image analysis interface.

8. A device for visualizing industrial defect images, characterized in that: include: An instruction response module, used to determine a plurality of defect images to be displayed and a target display dimension in response to a defect image display instruction; A dimension reduction processing module is used to obtain, for each defect image, an image feature vector corresponding to the defect image, perform dimension reduction processing on the image feature vector according to the target display dimension, and determine the coordinate position of the data point corresponding to the defect image according to the image feature vector after the dimension reduction processing; An image display module is used to determine, for each data point of the defect image, a first display position of the data point on an image display area based on the coordinate position of the data point, and display the defect image in the image display area in the form of a thumbnail image of the data point and the defect image according to the first display position.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images