Deep learning based cyber security visual dashboard generation method

Abstract

The application discloses a network security visualization dashboard generation method based on deep learning, which comprises the following steps: obtaining a plurality of network security related dashboard pictures; labeling the plurality of dashboard pictures; preprocessing the plurality of dashboard pictures; training a built deep learning model through the labeled plurality of dashboard pictures; importing network security data into the trained deep learning model; outputting a plurality of alternative dashboard pictures by the deep learning model; and generating a color palette for color setting of the selected dashboard picture after a user determines a final dashboard picture from the plurality of alternative dashboard pictures. The network security visualization dashboard generation method based on deep learning can automatically generate a dashboard scheme capable of reflecting the association between data according to the network security related data imported by the user.

Description

Technical Field

[0001] This invention relates to a method for generating a network security visualization dashboard based on deep learning. Background Technology

[0002] With the rapid expansion of network scale and applications, cybersecurity threats are constantly increasing. Single cybersecurity protection technologies can no longer meet the needs, and visualization technology has emerged in this context. Data visualization technology enables people to perceive data information, quickly identify data patterns and differences and discover data anomalies, quickly identify and intuitively cluster data, and also quickly discover new attack patterns and predict attack trends.

[0003] Currently, most mainstream SIEM platforms focus primarily on information detection, extraction, and performance improvement in the area of ​​cybersecurity, while their visualization capabilities are still insufficient. Most only offer simple data listings, presenting data through basic chart types such as tables, line graphs, and bar charts. Furthermore, they fail to uncover the inherent relationships between visual data and present them in appropriate combinations.

[0004] At the same time, because the design process of data visualization requires a lot of professional knowledge, many non-professional users find it difficult to obtain ideal visualization results even if they obtain cybersecurity data, which greatly hinders the further popularization of data analysis scenarios and audiences. Summary of the Invention

[0005] This invention provides a method for generating a network security visualization dashboard based on deep learning to solve the aforementioned technical problems, specifically adopting the following technical solution:

[0006] A method for generating a cybersecurity visualization dashboard based on deep learning includes the following steps:

[0007] Obtain several dashboard images related to network security;

[0008] Annotate several dashboard images;

[0009] Preprocess several dashboard images;

[0010] The built deep learning model is trained using several labeled dashboard images;

[0011] Import cybersecurity data into a trained deep learning model;

[0012] The deep learning model outputs multiple alternative dashboard images;

[0013] After the user selects the final dashboard image from multiple alternatives, a color palette is generated for the user to set the colors of the selected dashboard image.

[0014] Furthermore, the specific method for annotating several dashboard images is as follows:

[0015] For each dashboard image, label it with chart type, chart title, chart size, chart position, and color information.

[0016] Furthermore, the specific method for preprocessing several dashboard images is as follows:

[0017] Fill the blank areas of the dashboard image with a color that has high contrast to the image.

[0018] Furthermore, the deep learning model is the Faster-RNN network model.

[0019] Furthermore, network security data includes LAP-type databases, OLTP-type databases, API data sources, and data files.

[0020] Furthermore, the specific method for the deep learning model to output multiple alternative dashboard images is as follows:

[0021] The deep learning model scores each dashboard image based on the input cybersecurity data and outputs the top n scored dashboard images as candidate dashboard images.

[0022] Furthermore, after the deep learning model outputs multiple alternative dashboard images, the deep learning-based method for generating cybersecurity visualization dashboards also includes:

[0023] By performing a similarity check, highly similar dashboard images from the multiple candidate dashboard images output by the deep learning model are deleted.

[0024] Furthermore, the generated color palette includes a recommended color palette and a full color palette.

[0025] Furthermore, the specific method for generating the recommended color palette is as follows:

[0026] Extract the primary color from the final dashboard image;

[0027] Extract auxiliary colors from the final dashboard image;

[0028] A recommended color palette is generated based on the extracted primary and secondary colors.

[0029] Furthermore, several of the obtained cybersecurity-related dashboard images included computer-generated images and hand-drawn sketches.

[0030] The advantage of this invention lies in the fact that the deep learning-based network security visualization dashboard generation method can automatically generate a dashboard solution that can reflect the relationship between data based on network security-related data sources imported by the user. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for generating a network security visualization dashboard based on deep learning, according to the present invention. Detailed Implementation

[0033] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0034] like Figure 1 The diagram illustrates a deep learning-based method for generating a cybersecurity visualization dashboard, comprising the following steps: S1: Obtaining several cybersecurity-related dashboard images. S2: Annotating the dashboard images. S3: Preprocessing the dashboard images. S4: Training a pre-built deep learning model using the annotated dashboard images. S5: Importing cybersecurity data into the trained deep learning model. S6: The deep learning model outputs multiple candidate dashboard images. S7: After the user selects the final dashboard image from the candidate images, a color palette is generated for the user to set the colors of the selected image. Through these steps, the deep learning-based cybersecurity visualization dashboard generation method of this application can automatically generate a dashboard display scheme that reflects the relationships between data based on the cybersecurity-related data sources imported by the user. The following details these steps.

[0035] For step S1: Obtain several dashboard images related to network security.

[0036] Most existing dashboards offer convenient chart initialization templates and flexible customization options. However, dashboards contain multiple charts, and their design should be determined based on data, analysis tasks, and aesthetic factors, resulting in a wide variety of dashboard designs. This diversity makes it difficult to cover all situations with a limited number of templates. Novices lack the ability to modify them as needed and struggle to choose a dashboard that best presents their data.

[0037] This application includes a corpus of images containing over 8,000 dashboard instances, including computer-generated instances and hand-drawn sketches scraped from the internet. All images are standardized to 1920x1080 pixels, covering more than 95% of the needs of cybersecurity scenarios.

[0038] For step S2: Label several dashboard images.

[0039] Dashboards scraped from the internet were automatically annotated by extracting information from their stored files. Dashboards and sketches lacking sufficient information were manually annotated.

[0040] Dashboards typically consist of multiple visual charts. The annotations include several parts: chart type, chart title, chart size, chart position, and color information. Chart types include, but are not limited to, pie charts, geographic coordinate charts, basic line charts, area charts, and horizontal bar charts. Chart size specifies the length and width of each chart. Chart position indicates the chart's location relative to the dashboard image, using the overall image as a coordinate system to record the relative position of each chart type. For example, using the top-left corner of the overall image as the origin, the x-axis and y-axis coordinates of the top-left corner of each chart type are recorded. Color information includes the overall hue of the dashboard image and the color of each chart type. These five parts of the dataset are combined into a quintuple, serving as the data foundation for training the model.

[0041] For step S3: preprocess several dashboard images.

[0042] To more efficiently and accurately identify dashboard images, preprocessing is necessary. Since the white space around the image can affect recognition efficiency, the white space should be filled with a color of high contrast to the image. The contrast ratio can be selected based on actual needs.

[0043] For step S4: Train the built deep learning model using several labeled dashboard images.

[0044] The deep learning model used in this application is based on an optimized Faster-RNN network. By learning the features of different chart types, it can effectively identify various types of charts in the dashboard. It is understandable that adaptive optimizations can be made based on the Faster-RNN network.

[0045] For step S5: Import cybersecurity data into the trained deep learning model.

[0046] This application supports importing from various data sources, including OLAP databases, OLTP databases, API data sources, and data files. The data originates from relevant SIEM (Security Information and Event Management) tools.

[0047] For step S6: The deep learning model outputs multiple alternative dashboard images.

[0048] The specific method for a deep learning model to output multiple alternative dashboard images is as follows:

[0049] The deep learning model scores each dashboard image based on the input cybersecurity data and outputs the top n scored dashboard images as candidate images. n can be set as needed.

[0050] Preferably, after the model outputs the top n dashboard images with the highest scores, a similarity test is performed to remove highly similar dashboard images from the multiple candidate dashboard images output by the deep learning model, resulting in several final dashboard options for the user to choose from.

[0051] For step S7: After the user selects the final dashboard image from multiple alternative dashboard images, a color palette is generated for the user to set the colors of the selected dashboard image.

[0052] In the previous process, the overall style of the candidate dashboards and the styles of each chart type have been largely determined. However, users' perception of color varies, and different usage scenarios require different color tones. The candidate dashboards generated based on the model have already provided the highest-rated color combinations. Ideally, a color palette should also be generated, allowing users to choose recommended color tones or to use their creativity to adjust the colors according to different scenarios.

[0053] Preferably, the generated color palette includes a recommended color palette and a full-color palette. The full-color palette is fixed and is directly called from a pre-defined template.

[0054] The recommended color palette needs to be generated in real time based on the dashboard image ultimately selected by the user.

[0055] The specific method for recommending a color palette is as follows:

[0056] The primary color is extracted from the final dashboard image. Specifically, the color distribution ratios in the candidate dashboard images are extracted, and each identified color is labeled with three CIELAB channels and their proportions, denoted as C = (x, y, z, p). The primary color is then determined sequentially.

[0057] Auxiliary colors are extracted from the final dashboard image. Other significant elements of the dashboard include chart colors, background colors, and text colors. The color with the largest proportion in the candidate dashboard is selected as the background color. A greedy search in the CIELAB color space is used to generate the text color based on the background color. The text color is defined as a color with high contrast to the background color. Using the Web Content Accessibility Guidelines (WCAG 2.02) as the contrast definition, a greedy algorithm is used to select the text color from the entire color space, with a contrast ratio greater than 7:1 with the background color. As recommended by WCAG, this contrast ratio ensures good readability.

[0058] A recommended color palette is generated based on the extracted primary and secondary colors. After determining the primary color of the candidate dashboard, the remaining colors are clustered using the DBSCAN algorithm. For each cluster, the color with the largest proportion in the cluster is selected. Finally, the color of the cluster center is used to generate a color palette, which serves as the recommended color palette that closely matches the overall color tone of the candidate dashboard.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for generating a network security visualization dashboard based on deep learning, characterized in that, Includes the following steps: Obtain several dashboard images related to network security; Annotate several dashboard images; The specific method for annotating several dashboard images is as follows: For each dashboard image, label it with chart type, chart title, chart size, chart position, and color information; Preprocess several dashboard images; The built deep learning model is trained using several labeled dashboard images; Import cybersecurity data into the trained deep learning model. The cybersecurity data includes LAP databases, OLTP databases, API data sources, and data files. The deep learning model outputs multiple alternative dashboard images; The specific method by which the deep learning model outputs multiple candidate dashboard images is as follows: The deep learning model scores each dashboard image based on the input cybersecurity data and outputs the top n scored dashboard images as candidate dashboard images. The specific method by which the deep learning model outputs multiple candidate dashboard images is as follows: The deep learning model scores each dashboard image based on the input cybersecurity data and outputs the top n dashboard images as candidate dashboard images. After the user selects the final dashboard image from multiple alternatives, a color palette is generated for the user to set the colors of the selected dashboard image.

2. The method for generating a network security visualization dashboard based on deep learning according to claim 1, characterized in that, The specific method for preprocessing several dashboard images is as follows: Fill the blank areas of the dashboard image with a color that has high contrast to the image.

3. The method for generating a network security visualization dashboard based on deep learning according to claim 1, characterized in that, The deep learning model is the Faster-RNN network model.

4. The method for generating a network security visualization dashboard based on deep learning according to claim 1, characterized in that, After the deep learning model outputs multiple candidate dashboard images, the deep learning-based cybersecurity visualization dashboard generation method further includes: By performing a similarity check, highly similar dashboard images from the multiple candidate dashboard images output by the deep learning model are deleted.

5. The method for generating a network security visualization dashboard based on deep learning according to claim 1, characterized in that, The generated color palette includes a recommended color palette and a full color palette.

6. The method for generating a network security visualization dashboard based on deep learning according to claim 5, characterized in that, The specific method for generating a recommended color palette is as follows: Extract the primary color from the final dashboard image; Extract auxiliary colors from the final dashboard image; A recommended color palette is generated based on the extracted primary and secondary colors.

7. The method for generating a network security visualization dashboard based on deep learning according to claim 1, characterized in that, The acquired dashboard images related to cybersecurity included both computer-generated images and hand-drawn sketches.

Images