Vector map watermark embedding method and system

Abstract

The present invention relates to the field of geographic information security technology, and in particular to a vector map watermark embedding method and system. The method comprises: constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the model is an original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data; training the vector map learning model; inputting the original vector map into the trained vector map learning model, outputting a vector map containing geographic feature data, recorded as a first vector map; generating a watermark bit stream using a hash algorithm; and embedding the watermark bit stream into the geographic feature data of the first vector map using a strategy of dynamically adjusting the watermark embedding strength, thereby obtaining a vector map after watermark embedding, which can adapt to the structural characteristics of the vector map and has high robustness and transparency of watermark embedding, thereby realizing safe and reliable vector map data protection.

Description

Technical Field

[0001] The present invention relates to the field of geographic information security technology, and in particular to a vector map watermark embedding method and system. Background Art

[0002] Vector maps are important basic data in geographic information systems. Due to their accuracy and thoroughness, they have become information resources of great commercial value and strategic significance. They are widely used in various industries such as traffic navigation, urban planning and military applications. With the rapid development of Internet technology and mobile computing technology, the development of many geographic information system applications (such as Baidu Maps, Amap, Tencent Maps, etc.) has provided ordinary users with convenient access to vector maps. The emergence of some open geographic information system platforms (such as OpenStreetMap, etc.) has even provided ordinary users with the opportunity to participate in map making. However, this has inevitably led to the technical problem of the lack of effective security protection measures for vector maps, especially the lack of protection measures against illegal copying, tampering and unauthorized use of vector map data.

[0003] Digital watermarking technology is a copyright protection method that is widely used in multimedia data forms such as pixel-level images, audio and video, and has good effects. Therefore, more and more vector maps are beginning to use digital watermarking technology to protect vector map data. However, although the existing digital watermarking technology used in the field of vector maps (such as vector map data watermarking technology based on hash algorithms) can provide a certain degree of copyright protection, its robustness and efficiency are often greatly reduced when facing complex geometric transformation attacks, deep learning attacks, and structural challenges unique to vector data. These challenges have prompted researchers to explore new methods to improve the performance of vector map watermarking technology.

[0004] In recent years, the rapid development of deep learning technology has brought new opportunities to the field of digital watermarking, especially the powerful capabilities of deep neural networks (such as convolutional neural networks CNNs, residual networks ResNet, etc.) in image processing and feature extraction, making deep neural networks an ideal choice for optimizing watermark embedding and extraction processes. However, deep neural networks are mainly designed for pixel-level image data and are difficult to apply to the structural characteristics of vector map data. Specifically, vector map data has the structural characteristics of non-rasterization and geometric complexity, that is, vector map data is described by mathematical formulas and geometric objects, contains elements such as points, lines, and surfaces, has clear geographical meaning and structure, and has no pixelization restrictions. Therefore, when deep neural networks are used to process watermark embedding of vector map data, additional preprocessing steps are usually required, such as rasterization, and then pixel-level embedding is performed on the rasterized image. However, the precise geometric information of the vector map data will be pixelated during the rasterization process, which may not only lead to the loss of original geographic information, but also introduce errors, affecting the accuracy of the map and the accuracy of subsequent watermark embedding. In addition, rasterized data has a fixed resolution, which means that the rasterized map will not be able to accurately represent the details of the original vector map at different scales. At the same time, pixel-level embedding will also cause irregular changes in lines and shapes in the map, resulting in a decrease in map clarity, making it difficult to meet the requirements of vector map data for non-destructive watermark embedding to maintain the integrity and availability of the original map data. Summary of the invention

[0005] In response to the above technical problems, the present invention proposes a vector map watermark embedding method and system, which aims to embed vector map watermarks that are both adaptable to the structural characteristics of vector map data and have high robustness and transparency, thereby achieving more secure and reliable vector map data protection.

[0006] In a first aspect, the present application provides a vector map watermark embedding method, comprising the following steps:

[0007] Step 102, constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is the original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is the vector map containing the geographic feature data;

[0008] Step 104, randomly selecting a vector map as a training data set, and training the vector map learning model based on the training data set;

[0009] Step 106, inputting the original vector map into the trained vector map learning model, and outputting a vector map containing geographic feature data, which is recorded as a first vector map;

[0010] Step 108, using a hash algorithm to generate a watermark bit stream of the original vector map;

[0011] Step 110: embed the watermark bit stream into the geographic feature data of the first vector map by adopting a strategy of dynamically adjusting the watermark embedding strength, so as to obtain a vector map after watermark embedding.

[0012] In some embodiments, in step 110, the watermark bit stream is embedded into the geographic feature data of the first vector map using a strategy of dynamically adjusting the watermark embedding strength, including:

[0013] Step 202, dividing the first vector map into a plurality of sub-areas;

[0014] Step 204, calculating the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region;

[0015] Step 206: embed the watermark bit stream into the geographic feature data of each sub-region based on the watermark embedding strength of each sub-region.

[0016] In some embodiments, the geographic feature data includes point feature data, line feature data, and surface feature data. Then, in step 204, the watermark embedding strength of each sub-region is calculated based on the geographic feature data in the sub-region, including:

[0017] Determine the density of the sub-region based on the proportion of the number of feature points of the midpoint feature data, line feature data, and surface feature data in the sub-region;

[0018] Based on the density level of the sub-region, a first watermark embedding strength of the sub-region is determined, and the first watermark embedding strength is proportional to the density level of the sub-region.

[0019] In some embodiments, in step 204, the watermark embedding strength of each sub-region is calculated based on the geographic feature data in the sub-region, and further includes:

[0020] In each sub-region, a plurality of feature points in the geographic feature data are selected as embedding points according to the embedding strength of the first watermark;

[0021] Clustering all embedded points in the first vector map using a clustering algorithm to obtain a central reference point;

[0022] Calculate the distance from the embedded point in each sub-region to the central reference point;

[0023] Based on the distance from the embedding point in each sub-region to the central reference point, the second watermark embedding strength of each sub-region is determined, and the second watermark embedding strength is inversely proportional to the distance from the embedding point in each sub-region to the central reference point.

[0024] In some embodiments, the geographic feature data includes point feature data, line feature data, and surface feature data. Then, in step 206, based on the watermark embedding strength of each sub-region, the watermark bit stream is embedded in the geographic feature data of each sub-region, including:

[0025] In each sub-region, a number of feature points in the geographic feature data are selected as embedding points according to the determined watermark embedding strength;

[0026] According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the feature points of the corresponding geographic feature data.

[0027] In some embodiments, according to the type of geographic feature data corresponding to the embedding point, embedding the watermark bit stream into the decimal places of the feature point coordinate value of the corresponding geographic feature data includes:

[0028] If the embedded point is point feature data, the watermark bit stream is embedded in the decimal place of the coordinate value of the point feature data.

[0029] If the embedded point is line feature data, the watermark bit stream is embedded in the decimal place of the feature point coordinate value of the line feature data.

[0030] If the embedded point is surface feature data, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the surface boundary points of the surface feature data.

[0031] In some embodiments, according to the type of geographic feature data corresponding to the embedding point, embedding the watermark bit stream into the decimal places of the feature point coordinate value of the corresponding geographic feature data includes:

[0032] Get the preset watermark embedding threshold;

[0033] In the process of embedding the watermark bit stream into the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data, it is determined in real time whether the difference between the embedded feature point coordinate values ​​and the original feature point coordinate values ​​exceeds the embedding threshold. If so, the part of the watermark bit stream exceeding the embedding threshold will no longer be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data. If not, the watermark bit stream will continue to be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data.

[0034] In some embodiments, according to the type of geographic feature data corresponding to the embedding point, embedding the watermark bit stream into the decimal places of the feature point coordinate value of the corresponding geographic feature data includes:

[0035] Get the preset watermark embedding threshold;

[0036] Dividing the watermark bit stream into several watermark bit groups bit by bit based on the embedding threshold of the watermark;

[0037] According to the type of geographic feature data corresponding to the embedded point, several watermark bit groups are embedded bit by bit in the decimal places of the feature point coordinate value of the corresponding geographic feature data, so that the difference between the feature point coordinate value after the watermark bit group is embedded and the original feature point coordinate value is lower than the embedding threshold.

[0038] In some embodiments, after step 110, the method further includes:

[0039] Step 111, using a composite loss function, based on the embedded vector map, to optimize the vector map learning model, the composite loss function includes an MSE Loss function, an SSIM Loss function and an Adversarial Loss function, wherein the MSE Loss function is used to measure the pixel-level difference between the vector map before and after watermark embedding, the SSIM Loss function is used to evaluate the overall visual quality of the vector map after watermark embedding, and the Adversarial Loss function is used to determine whether the vector map after watermark embedding is authentic.

[0040] In a second aspect, the present application provides a vector map watermark embedding system, comprising:

[0041] A model building module, which builds a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is an original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data;

[0042] A model training module randomly selects a vector map as a training data set, and trains the vector map learning model based on the training data set;

[0043] A first vector map generation module, inputting the original vector map into the trained vector map learning model, and outputting a vector map containing geographic feature data, which is recorded as a first vector map;

[0044] The watermark generation module uses a hash algorithm to generate the watermark bit stream of the original vector map;

[0045] The watermark embedding module adopts a strategy of dynamically adjusting the watermark embedding strength to embed the watermark bit stream into the geographic feature data of the first vector map to obtain the vector map after the watermark is embedded.

[0046] The beneficial technical effects of the present invention include at least:

[0047] 1. A vector map watermark embedding method and system is used to construct a vector map learning model for learning the complex geometric structure and key geographical features of vector map data, which provides a basis for embedding watermarks in these key geographical features in the future, thereby significantly reducing the impact of watermark embedding on the quality of vector map data. In the watermark embedding process, a hash algorithm and a strategy for dynamically adjusting the watermark embedding strength are introduced to make the coordinate transformation direction of the embedded point more random, the transformation amplitude controllable and smoother, so that the watermark embedding result is more consistent with the topological expression of the original vector map data, less likely to be detected or affect the use of data, and can adapt to the structural characteristics of vector map data, and have high robustness and transparency, so as to achieve safer and more reliable data protection. In addition, through the synergy of the vector map learning model and the hash algorithm, large-scale vector map data can be efficiently processed, and the practicality and reliability of watermark technology in practical applications can be significantly improved. This comprehensive solution not only achieves progress in the field of vector map data security protection, but also opens up a new direction for the development of geographic information system technology;

[0048] 2. According to different sub-regions of the vector map, the watermark embedding strength of each sub-region is dynamically adjusted based on the geographic feature data in the sub-region. For regions with rich geographic feature data, the embedding amount of the watermark bit stream can be appropriately increased, while for regions with less geographic feature data, the embedding amount of the watermark bit stream can be reduced. This avoids excessive distortion of the vector map data while maintaining the robustness of the watermark, meets the requirements of the vector map data for non-destructive watermark embedding to maintain the integrity and availability of the original map data, and can also optimize the use of computing resources to avoid wasting computing resources for a large number of watermark embeddings in map regions with single features;

[0049] 3. Based on the proportion of feature points of midpoint feature data, line feature data and surface feature data in the sub-region, the sub-regions with different densities are distinguished, and the watermark embedding strength is adaptively adjusted according to the different densities of the sub-regions. The watermark embedding of the vector map is finely regulated, which not only avoids the distortion of the original information of the low-density area of ​​the vector map due to watermark embedding, ensures the naturalness and practicality of the vector map, but also improves the concealment and anti-interference ability of the watermark in the high-density area of ​​the vector map, ensuring the balance between the robustness and transparency of the watermark;

[0050] 4. The embedding strength of the first watermark in different sub-areas is locally adjusted according to the distance from the embedding point to the central reference point, so that the embedding strength of the sub-area close to the central reference point can be appropriately increased, while the embedding strength of the sub-area far from the central reference point can be appropriately reduced. This can further maintain the smooth transition of the watermark in the entire vector map, better integrate with the vector map content, and avoid abrupt visual effects;

[0051] 5. By designing a specific composite loss function, the vector map learning model can take into account the transparency of the watermark, ensure the overall visual quality of the vector map, and take into account the robustness of the watermark, thereby achieving multi-objective optimization of the vector map learning model.

[0052] Other features and advantages of the present invention will be disclosed in detail in the following specific embodiments and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below in conjunction with the accompanying drawings:

[0054] Figure 1 This is a flow chart of a vector map watermark embedding method according to an embodiment of the present invention.

[0055] Figure 2 This is a flow chart of a method for calculating the watermark embedding strength of each sub-region according to an embodiment of the present invention.

[0056] Figure 3 This is a flow chart of a method for calculating the watermark embedding strength of each sub-region according to another embodiment of the present invention.

[0057] Figure 4 The figure is a schematic diagram of the principle of embedding vector data watermark of a curve in a vector map according to an embodiment of the present invention.

[0058] Figure 5 This is a schematic diagram of the structure of a vector map watermark embedding system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions of the embodiments of the present invention are explained and described below in conjunction with the drawings of the embodiments of the present invention, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the protection scope of the present invention.

[0060] In the following description, terms such as "inside", "outside", "up", "down", "left", "right", etc. that indicate directions or positional relationships are only used to facilitate the description of the embodiments and simplify the description, and do not indicate or imply that the referred device or element must have a specific direction, be constructed and operate in a specific direction. Therefore, it should not be understood as a limitation of the present invention.

[0061] This application embodiment provides a vector map watermark embedding method, please refer to the attached Figure 1 , including the following steps:

[0062] Step 102, constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is the original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data.

[0063] Among them, the deep neural network architecture including the encoder and the decoder includes but is not limited to U-Net, VGG-16 / 19, FCN, SegNet, Mask R-CNN, etc., which is not limited in this embodiment.

[0064] For example, taking the U-Net architecture including an encoder and a decoder as an example, the constructed vector map learning model includes:

[0065] a. Input layer: used to receive the original vector map data and convert it into a tensor form that can be processed by the model. Tensor refers to a concept in deep learning. Common tensors are in the form of high-dimensional arrays as data containers. In this embodiment, the original vector map data is converted into a tensor form for storage so that the model can perform subsequent calculations.

[0066] b. Encoder: It consists of multiple convolutional layers, each of which is followed by a ReLU activation function and BatchNormalization. The encoder is used to extract the geographic features of the original vector map and reduce overfitting.

[0067] c. Furthermore, in this embodiment, an attention module can be provided between the encoder and the decoder to weight the feature map so as to enhance the model's ability to capture key geographic features of the vector map.

[0068] d. Decoder part: used to reconstruct a vector map containing key geographic features, which also includes convolutional layers, upsampling layers, Batch Normalization and ReLU activation functions.

[0069] e. Output layer: Outputs a vector map containing geographic feature data for subsequent watermark embedding.

[0070] It is understood that in this embodiment, the geographic feature data of the vector map includes but is not limited to point feature data, line feature data, surface feature data and attribute information. Among them, the point feature data may at least include:

[0071] 1. Geographical location: coordinate points such as cities, villages, landmarks, etc.

[0072] 2. Control points: such as road intersections, river confluences, etc.

[0073] 3. Facility location: such as the coordinates of schools, hospitals, police stations, etc.

[0074] Line feature data may include at least:

[0075] 1. Roads: including expressways, provincial roads, county roads, urban roads, etc.

[0076] 2. Rivers: and other water systems, such as lakes, canals, etc.

[0077] 3. Administrative boundaries: national borders, provincial borders, municipal borders, county borders, etc.

[0078] 4. Public transportation routes: bus, subway, train and other routes.

[0079] The face feature data may include at least:

[0080] 1. Administrative region: the jurisdiction of administrative units such as provinces, cities, counties, and towns.

[0081] 2. Natural areas: natural landforms such as mountains, plains, lakes, etc.

[0082] 3. Land use: such as residential areas, industrial areas, agricultural areas, parks, etc.

[0083] 4. Special-purpose areas: such as military restricted areas, nature reserves, etc.

[0084] The attribute information may include at least:

[0085] 1. Name: such as place name, road name, building name, etc.

[0086] 2. Categories: such as settlements, water areas, roads, administrative divisions, etc.

[0087] 3. Attributes: such as road type (highway, national highway, etc.) and grade (first-level, second-level, etc.).

[0088] Vector map data exists in a geometric form, including elements such as points, lines, and surfaces, has clear geographical meanings and structures, and is not restricted by pixelation, which is essentially different from rasterized image data. To this end, this embodiment designs a network structure capable of processing such non-rasterized data based on a deep neural network including an encoder and a decoder, so that the model can learn the complex geometric structure and key geographical features of the vector map data. Since the geographical feature points extracted by the vector map learning model are specifically selected for each vector map data, they are more targeted than other algorithm rules. Therefore, they can have stronger robustness in the face of malicious attacks and unintentional changes, and provide a basis for the subsequent embedding of watermarks in these key geographical features, thereby significantly reducing the impact of watermark embedding on the quality of vector map data, and also helping to accurately restore the watermark by using these learned key geographical features during watermark detection.

[0089] Step 104: randomly select a vector map as a training data set, and train a vector map learning model based on the training data set.

[0090] Among them, the vector maps used as training data sets include maps of different scales and multiple types (city planning maps, topographic maps, traffic maps, etc.). This embodiment does not limit this to ensure that the model can learn the expression of geographical features of different scales and types, which helps to improve the vector map learning model's ability to extract key structural features.

[0091] Furthermore, the randomly selected vector maps may be preprocessed to standardize the coordinate system and unify the data format, and then the preprocessed vector maps may be used as training data sets.

[0092] Step 106: input the original vector map into the trained vector map learning model, and output a vector map containing geographic feature data, which is recorded as the first vector map.

[0093] Step 108: Generate a watermark bit stream of the original vector map using a hash algorithm.

[0094] Specifically, in this embodiment, the method of using the hash algorithm to generate the watermark bit stream of the original vector map includes:

[0095] (1) Hash function selection: A collision-resistant hash function, such as SHA-3, is selected to ensure that even if there are slight changes in the vector map feature data during the process of extracting geographic feature data during watermark detection, such as intentional or unintentional geometric transformations (such as scaling, rotation, translation, etc.), the hash value will be significantly different, thereby ensuring the uniqueness and security of the watermark.

[0096] (2) Watermark information encoding:

[0097] Encode metadata such as copyright information and timestamps into binary strings. For example, first, encode the copyright information into ASCII code, such as "©2024 CNCICO", and record the time when the watermark is embedded in the format of Unix timestamp. Then concatenate the above information and perform binary encoding to obtain the encoded watermark information.

[0098] It can be understood that metadata embedded in the timestamp associates the watermark with a specific point in time, increasing the copyright traceability of the watermark, while also making the watermark information more resistant to tampering, because any subsequent geometric transformation may change the timestamp information, resulting in differences in the watermark information.

[0099] (3) Generate watermark bit stream: Use the SHA-3 algorithm to generate a hash value for the encoded watermark information, that is, obtain the watermark bit stream, which is convenient for subsequent embedding in the decimal places of the coordinate value of the point feature data, the offset of the line feature data, or the deformation of the surface feature data.

[0100] Furthermore, in this embodiment, the binary watermark information obtained in step (2) and the hash value obtained in step (3) can be subjected to an exclusive OR operation (XOR) to obtain a final watermark bit stream. This step further increases the complexity and security of the watermark bit stream. Even if the watermark bit stream is intercepted, it is difficult to decipher the original watermark information without the corresponding hash value.

[0101] This embodiment ensures that the watermark can still be accurately detected and restored through the selection and optimization of the hash algorithm even if the vector map data is attacked by geometric transformation, thereby ensuring the robustness of the watermark information.

[0102] Step 110: embed the watermark bit stream into the geographic feature data of the first vector map by adopting a strategy of dynamically adjusting the watermark embedding strength, so as to obtain a vector map after watermark embedding.

[0103] Furthermore, in this embodiment, in step 110, a strategy of dynamically adjusting the watermark embedding strength is adopted to embed the watermark bit stream into the geographic feature data of the first vector map, including:

[0104] Step 202, dividing the first vector map into a plurality of sub-areas;

[0105] Step 204: Calculate the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region.

[0106] Step 206: Based on the watermark embedding strength of each sub-region, the watermark bit stream is embedded into the geographic feature data of each sub-region to obtain a vector map after watermark embedding.

[0107] It can be understood that the watermark embedding strength in this embodiment is the number of watermark embeddings in each sub-region. A high watermark embedding strength in the current sub-region indicates that a large number of watermark bit streams are embedded in the current sub-region, and a low watermark embedding strength in the current sub-region indicates that a small number of watermark bit streams are embedded in the current sub-region.

[0108] This embodiment dynamically adjusts the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region according to the different sub-regions of the vector map. For the region with rich geographic feature data, the embedding amount of the watermark bit stream can be appropriately increased, while for the region with less geographic feature data, the embedding amount of the watermark bit stream can be reduced. Thus, while maintaining the robustness of the watermark, excessive distortion of the vector map data is avoided, the demand for non-destructive watermark embedding of the vector map data to maintain the integrity and availability of the original map data is met, and the use of computing resources is optimized to avoid wasting computing resources for a large amount of watermark embedding in a map region with a single feature.

[0109] For further information, please see the attached Figure 2 In step 204, based on the geographic feature data in the sub-region, the watermark embedding strength of each sub-region is calculated, including:

[0110] Step 302, determining the density of the sub-region based on the ratio of the number of feature points of the point feature data, line feature data and surface feature data in the sub-region;

[0111] Step 304: Determine the first watermark embedding strength of the sub-region based on the density of the sub-region, where the first watermark embedding strength is proportional to the density of the sub-region.

[0112] It can be understood that the density of a sub-region may include at least high density, medium density and low density. If a sub-region has more feature points of point feature data and line feature data and fewer feature points of surface feature data, it means that the sub-region may be a city center area, which is a high-density area. If a sub-region has more feature points of surface feature data and fewer feature points of point feature data and line feature data, it means that the sub-region may be an open natural area, which is a low-density area. The first watermark embedding strength is proportional to the density of the sub-region, which means that in the high-density city center area, due to the presence of a large number of details, the watermark embedding strength can be slightly stronger, while in the open natural area, the watermark embedding strength should be weakened to avoid obvious distortion of the landform.

[0113] In step 204 of this embodiment, the specific implementation method of calculating the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region may be:

[0114] First, the proportion of feature points in the center feature data, line feature data and surface feature data in each sub-area is calculated respectively, which is used as an indicator to measure the degree of density.

[0115] For example, taking the first vector map being divided into four sub-areas as an example, it is assumed that sub-area A includes 200 point feature data, feature points in 100 line feature data, and feature points in 50 surface feature data, and the total number of feature points is 350; sub-area B includes 50 point feature data, feature points in 150 line feature data, and feature points in 200 surface feature data, and the total number of feature points is 400; sub-area C includes 100 point feature data, feature points in 100 line feature data, and feature points in 100 surface feature data, and the total number of feature points is 300; sub-area D includes 75 point feature data, feature points in 75 line feature data, and feature points in 150 surface feature data, and the total number of feature points is 300.

[0116] Then the proportion of point feature data in sub-region A is: 200 / 350≈0.571, the proportion of feature points in line feature data is: 100 / 350≈0.286, and the proportion of feature points in surface feature data is: 50 / 350≈0.143. It can be seen that since the proportion of point feature data in sub-region A is greater than 0.5, we can consider sub-region A to be a high-density area;

[0117] The proportion of point feature data in sub-region B is: 50 / 400=0.125, the proportion of feature points in line feature data is: 150 / 400=0.375, and the proportion of feature points in surface feature data is: 200 / 400=0.5. It can be seen that since the proportion of feature points in surface feature data is equal to 0.5, we can consider sub-region B to be a low-density area;

[0118] The proportion of point feature data in sub-region C is: 100 / 300≈0.333, the proportion of feature points in line feature data is: 100 / 300≈0.333, and the proportion of feature points in surface feature data is: 100 / 300≈0.333. It can be seen that the proportion of feature points in point, line, and surface feature data is relatively balanced. We can consider sub-region C to be a medium-density area;

[0119] The proportion of point feature data in sub-area D is: 75 / 300=0.25, the proportion of feature points in line feature data is: 75 / 300=0.25, and the proportion of feature points in surface feature data is: 150 / 300=0.5. It can be seen that since the proportion of feature points in surface feature data is equal to 0.5, we can consider sub-area D to be a low-density area.

[0120] Then, based on the density of the sub-region, the first watermark embedding strength of the sub-region is determined. The first watermark embedding strength here can be set according to the actual density of the vector map, and this embodiment does not limit this. The first watermark embedding strength is proportional to the density of the sub-region, indicating that a higher first watermark embedding strength is assigned to a high-density sub-region, that is, a larger number of watermark bit streams are embedded in the high-density sub-region. For example, the first watermark embedding strength for the high-density sub-region is 50%, which means that in the high-density sub-region, 50 out of every 100 feature points of geographic feature data will be selected as embedding points. It also indicates that a lower first watermark embedding strength is assigned to a low-density sub-region, that is, a smaller number of watermark bit streams are embedded in the low-density sub-region, for example, 15%, which means that in the low-density sub-region, 15 out of every 100 feature points of geographic feature data will be selected as embedding points.

[0121] This embodiment distinguishes sub-regions of different densities based on the proportion of the number of feature points of midpoint feature data, line feature data and surface feature data in the sub-region, adaptively adjusts the watermark embedding strength according to the different densities of the sub-regions, and finely regulates the watermark embedding of the vector map, thereby avoiding the distortion of the original information in the low-density area of ​​the vector map due to watermark embedding, ensuring the naturalness and practicality of the vector map, and improving the concealment and anti-interference ability of the watermark in the high-density area of ​​the vector map, ensuring the balance between the robustness and transparency of the watermark.

[0122] Furthermore, the geographic feature data includes point feature data, line feature data and surface feature data. Then, in step 206, based on the watermark embedding strength of each sub-region, the watermark bit stream is embedded into the geographic feature data of each sub-region respectively, including:

[0123] In each sub-region, a number of feature points in the geographic feature data are selected as embedding points according to the determined watermark embedding strength;

[0124] According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the feature points of the corresponding geographic feature data.

[0125] Among them, in each sub-area, several feature points in the geographic feature data can be randomly selected as embedding points according to the determined watermark embedding strength, or several feature points in the geographic feature data can be selected as embedding points using algorithms such as dynamic programming, which is not limited in this embodiment.

[0126] It can be understood that the feature points of geographic feature data represent key position points of point feature data, line feature data and surface feature data with specific meaning or representativeness on the vector map. The point feature data itself is a feature point. There can be several feature points in both line feature data and surface feature data. The feature points of surface feature data are located on the surface boundary of the surface feature data. The selection of feature points of geographic feature data is mainly based on not affecting the accuracy of the vector map, and this embodiment does not limit this.

[0127] Preferably, in this embodiment, according to the type of geographic feature data corresponding to the embedding point, embedding the watermark bit stream into the decimal place of the feature point coordinate value of the corresponding geographic feature data includes:

[0128] If the embedded point is point feature data, the watermark bit stream is embedded in the decimal place of the coordinate value of the point feature data.

[0129] If the embedded point is line feature data, the watermark bit stream is embedded in the decimal place of the feature point coordinate value of the line feature data.

[0130] If the embedded point is surface feature data, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the surface boundary points of the surface feature data.

[0131] Specifically, as the point feature data of the embedded point, the watermark bit stream is embedded in the decimal places of the coordinate value of the point feature data. For example, for the coordinates of the point feature data (51.5074, -0.1278), the first few decimal places of the coordinate value of the point feature data can be kept unchanged, and the watermark bit stream can be embedded in the last few decimal places of the coordinate value of the point feature data. In this way, even if the watermark bit stream is embedded, the overall appearance and accuracy of the vector map will not be greatly affected;

[0132] The line feature data as embedded points usually represent roads, rivers, etc. The watermark bit stream can be embedded in the decimal place of the coordinate value of a feature point on the line feature data, and the coordinate of a feature point on the line segment can be slightly adjusted to achieve a slight offset of the line feature data that is almost imperceptible visually. At the same time, the watermark bit stream is included in the digital representation of the line segment.

[0133] The surface feature data as embedded points usually represent closed areas such as industrial areas and lakes. The watermark bit stream can be embedded in the decimal places of the coordinate values ​​of the surface boundary points of the surface feature data. That is, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of a feature point on the surface boundary, so that the shape of the entire surface undergoes extremely subtle changes. At the same time, the watermark bit stream is included in the digital representation of the surface shape.

[0134] This embodiment avoids significant impact on the visual presentation of the vector map and ensures the accuracy of the vector map data by embedding the watermark bit stream into the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data according to the type of geographic feature data corresponding to the embedding point.

[0135] It is understandable that when designing a vector map watermark embedding method, a watermark detection method is usually included to verify the existence and integrity of the watermark in the vector map. To this end, in another embodiment of the present specification, after step 110, the following steps are also included:

[0136] Step 112, attacking the vector map after the watermark is embedded, and recording the attacked vector map after the watermark is embedded as a second vector map.

[0137] Among them, the attack operations performed on the vector map after watermark embedding in this embodiment are similar to the attack operations performed on the vector map in the prior art to detect watermarks, including but not limited to geometric transformation attacks, deep learning attacks, illegal copying and tampering, etc., which will not be repeated in this embodiment.

[0138] Step 114, inputting the second vector map into the trained vector map learning model, and outputting the second vector map containing geographic feature data;

[0139] Step 116, calculating the difference between the coordinate values ​​of the feature points of the geographic feature data of the second vector map and the coordinate values ​​of the feature points of the corresponding geographic feature data of the first vector map.

[0140] It can be understood that the feature point coordinate values ​​of the geographic feature data of the second vector map are the feature point coordinate values ​​that already contain watermark information in the second vector map, and the difference between the feature point coordinate values ​​of the geographic feature data of the second vector map and the feature point coordinate values ​​of the corresponding geographic feature data of the first vector map is the watermark information of the feature point in the second vector map corresponding to the first vector map.

[0141] Step 118, compare the difference with the watermark bit stream to obtain a watermark integrity detection result.

[0142] Among them, comparing the difference with the watermark bit stream is to perform non-blind watermark detection on the watermark information of the feature points extracted from the second vector map and the watermark bit stream of the original vector map. According to the comparison result, it can be detected whether the watermark has been tampered with or damaged, that is, the change in watermark shape caused by the attack operation is identified, and the integrity of the watermark is confirmed. Subsequently, the watermark information of the feature points extracted from the second vector map can also be restored according to the watermark bit stream of the original vector map.

[0143] Steps 112 to 118 in this embodiment establish a set of fast and accurate watermark detection and verification processes, so that even if the vector map data is attacked by geometric transformation or deep learning, the integrity and validity of the watermark can be confirmed and restored through the joint verification of the vector map learning model and the hash algorithm. In this way, the copyright and integrity of the vector map data are effectively protected.

[0144] In another embodiment of this specification, please refer to the attached Figure 3 Compared with the above-mentioned embodiment, in step 204, the present embodiment calculates the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region, and further includes the following steps:

[0145] Step 306: In each sub-region, several feature points in the geographic feature data are selected as embedding points according to the first watermark embedding strength.

[0146] Step 308: cluster all embedded points in the first vector map using a clustering algorithm to obtain a central reference point.

[0147] Among them, the clustering algorithm for clustering all embedded points in the first vector map to obtain the central reference point in this embodiment is similar to the clustering algorithm for clustering feature points to determine the central reference point in the prior art, such as K-means clustering algorithm, DBSCAN clustering algorithm, OPTICS clustering algorithm, Gaussian mixture model clustering algorithm, etc., which will not be repeated in this embodiment.

[0148] Step 310, calculating the distance from the embedded point in each sub-region to the central reference point.

[0149] In this embodiment, the distance from the embedded point in each sub-region to the central reference point can be calculated using Euclidean distance, Manhattan distance or other suitable distance measurement methods, which is not limited in this embodiment.

[0150] Step 312, based on the distance from the embedding point in each sub-region to the central reference point, determine the embedding strength of the second watermark in each sub-region, and the embedding strength of the second watermark is inversely proportional to the distance from the embedding point in each sub-region to the central reference point.

[0151] Among them, based on the distance from the embedding point in each sub-region to the central reference point, the first watermark embedding strength of the sub-region is adjusted to the second watermark embedding strength, and the second watermark embedding strength here can be set according to the actual distance between each sub-region and the central reference point in the vector map, which is not limited in this embodiment. The second watermark embedding strength is inversely proportional to the distance from the embedding point in each sub-region to the central reference point, indicating that the closer the embedding point in the sub-region is to the central reference point, the greater the second watermark embedding strength of the sub-region, and the farther the embedding point in the sub-region is from the central reference point, the smaller the second watermark embedding strength of the sub-region.

[0152] This embodiment locally adjusts the embedding strength of the first watermark in different sub-areas according to the distance from the embedding point to the central reference point, so that the embedding strength of the sub-area close to the central reference point can be appropriately increased, while the embedding strength of the sub-area far from the central reference point can be appropriately reduced. In this way, the smooth transition of the watermark in the entire vector map can be further maintained, and it can be better integrated with the vector map content to avoid abrupt visual effects.

[0153] In another embodiment of the present specification, compared with the foregoing embodiment, this embodiment embeds the watermark bit stream into the decimal place of the feature point coordinate value of the corresponding geographic feature data according to the type of the geographic feature data corresponding to the embedding point, and further includes:

[0154] Get the preset watermark embedding threshold;

[0155] In the process of embedding the watermark bit stream into the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data, it is determined in real time whether the difference between the embedded feature point coordinate values ​​and the original feature point coordinate values ​​exceeds the embedding threshold. If so, the part of the watermark bit stream exceeding the embedding threshold will no longer be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data. If not, the watermark bit stream will continue to be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data.

[0156] It can be understood that the preset watermark embedding threshold in this embodiment determines the maximum amplitude of watermark embedding. This embedding threshold can be adjusted and set according to the characteristics and requirements of the vector map data, and this embodiment does not limit this.

[0157] Optionally, if the difference between the coordinate value of the embedded feature point and the coordinate value of the original feature point exceeds the embedding threshold, the watermark embedding of the feature point of the geographic feature data may be abandoned.

[0158] For example, see the attached Figure 4 , a curve vector data in a vector map is as follows Figure 4As shown in the upper middle curve, the curve vector data contains feature points from a total of 5 line feature data, M1-M5. After analyzing the watermark embedding strength of the current sub-region, two feature points, M2 and M4, are selected as embedding points. The watermark bit stream W is generated based on the hash algorithm according to the curve content and serial number. In the process of embedding the watermark bit stream W into the decimal places of the coordinate values ​​of M2 and M4, M2' and M4' are obtained. Then, the change amplitude of the coordinates of the embedded feature points and the original feature point coordinates is judged. If the change amplitude of M2' (i.e., the value of M2'-M2) exceeds the preset embedding threshold (i.e., the maximum value of the difference between M'-M), and the change amplitude of M4' (i.e., the value of M4'-M4) does not exceed the preset embedding threshold, the watermark embedding of point M2 is abandoned, and only the watermark embedding of point M4 is continued. Assuming that the watermark bit stream W has just been embedded at this time, the curve vector data after the watermark bit stream is finally obtained as shown in the figure. Figure 4 As shown in the lower middle curve, the dotted line content is the situation of fine-tuning the curve after the watermark is embedded at point M4 in the curve.

[0159] In the watermark embedding process, this embodiment monitors the change amplitude of each embedding point coordinate in real time. If the change amplitude exceeds the preset embedding threshold, the watermark embedding operation will be restricted to prevent negative impact on the visual quality of the vector map data and improve the transparency of the watermark. At the same time, it can assist the vector map learning model in identifying geographic features and avoid the situation in practice where the model cannot identify feature points due to excessive change amplitude of watermark embedding.

[0160] In another embodiment of the present specification, an alternative to the previous embodiment is proposed. This embodiment embeds the watermark bit stream into the decimal place of the coordinate value of the feature point of the corresponding geographic feature data according to the type of the geographic feature data corresponding to the embedding point, and further includes:

[0161] Get the preset watermark embedding threshold;

[0162] Dividing the watermark bit stream bit by bit into a number of watermark bit groups based on the embedding threshold of the watermark;

[0163] According to the type of geographic feature data corresponding to the embedded point, several watermark bit groups are embedded bit by bit in the decimal places of the feature point coordinate value of the corresponding geographic feature data, so that the difference between the feature point coordinate value after the watermark bit group is embedded and the original feature point coordinate value is lower than the embedding threshold.

[0164] It can be understood that, in this embodiment, the watermark embedding strength is determined by segmenting the watermark bit stream and confirming a number of watermark bit groups based on the watermark embedding threshold. The higher the watermark embedding strength, the more watermark bit groups are embedded.

[0165] Among them, according to the type of geographic feature data corresponding to the embedded point, the implementation method of embedding several watermark bit groups bit by bit into the decimal places of the feature point coordinate value of the corresponding geographic feature data can be: for fine-tuning the watermark embedding coordinates of the point feature data, the watermark bit groups can be embedded into the decimal places of the coordinate values ​​of each point feature data according to the order of division; for the watermark embedding offset of the line feature data or the watermark embedding deformation of the surface feature data, the watermark bit stream can be embedded in a dispersed manner according to similar rules, and this embodiment will not be repeated here.

[0166] Compared with the technical solution of real-time monitoring of the change amplitude of each embedded point coordinate in the previous embodiment, the technical solution in this embodiment is less complex. The embedding threshold of the watermark is taken into consideration when dividing the watermark bit group, and combined with the embedding of the decimal places of the feature point coordinate values ​​of the geographic feature data, the change amplitude of each embedded point coordinate is directly controlled within the range of the embedding threshold to prevent negative impact on the visual quality of the vector map data.

[0167] In summary, the vector map watermark embedding method proposed in multiple embodiments of this specification constructs a vector map learning model for learning the complex geometric structure and key geographical features of vector map data, providing a basis for embedding watermarks in these key geographical features in the future, thereby significantly reducing the impact of watermark embedding on the quality of vector map data, and in the watermark embedding process, introducing a hash algorithm and a strategy for dynamically adjusting the watermark embedding strength, so that the coordinate transformation direction of the embedded point is more random, the transformation amplitude is controllable and smoother, so that the watermark embedding result is more consistent with the topological expression of the original vector map data, less likely to be detected or affect the use of data, and can adapt to the structural characteristics of vector map data, and has high robustness and transparency, so as to achieve safer and more reliable data protection. In addition, through the synergy of the vector map learning model and the hash algorithm, the vector map watermark embedding method proposed in multiple embodiments of this specification can efficiently process large-scale vector map data, significantly improve the practicality and reliability of watermark technology in practical applications, and this comprehensive solution not only achieves progress in the field of vector map data security protection, but also opens up a new direction for the development of geographic information system technology.

[0168] In another embodiment of the present specification, after step 110, the following further steps are included:

[0169] Step 111, using a composite loss function, based on the embedded vector map, to optimize the vector map learning model, the composite loss function includes an MSE Loss function, an SSIM Loss function, and an Adversarial Loss function, wherein the MSE Loss function is used to measure the pixel-level difference between the vector map before and after watermark embedding, the SSIM Loss function is used to evaluate the overall visual quality of the vector map after watermark embedding, and the Adversarial Loss function is used to determine whether the vector map after watermark embedding is authentic.

[0170] Among them, the MSE Loss function, namely the Mean Squared Error loss function, is a loss function used in machine learning to evaluate the difference between the model prediction value and the true value in regression problems. This embodiment uses the MSELoss function to measure the pixel-level difference of the vector map data before and after the watermark is embedded, which is used to maintain the transparency of the watermark. This loss function encourages the model to reduce the changes at the pixel level, so that the impact of the watermark embedding on the vector map data is as small as possible, and the transparency of the watermark is improved. The SSIM Loss function, namely the Structural Similarity Index, is an indicator for evaluating two sets of similarity data. It is widely used in image quality assessment, especially in the fields of image processing and computer vision. In this embodiment, by optimizing the SSIM loss function, the model can embed the watermark while ensuring the overall visual quality of the vector map. In this embodiment, the Adversarial Loss function determines whether the vector map after the watermark is embedded is real through the adversarial network, which is used to ensure the robustness of the watermark. Through the training of adversarial loss, the model learns not only how to generate high-quality vector maps, but also how to keep the watermark detectable after being attacked.

[0171] In this embodiment, corresponding weights are set for MSE, SSIM and adversarial loss functions respectively, and then the loss values ​​obtained by each loss function are weighted and summed to obtain a total loss value, and the network parameters of the encoder and decoder are continuously updated with the total loss value, so that the neural network of the vector map learning model gradually converges, thereby optimizing the vector map learning model, wherein the selection of weights for MSE, SSIM and adversarial loss functions depends on the requirements for transparency and robustness of specific application scenarios of vector maps, which are not limited in this embodiment. For example, the implementation method of optimizing the vector map learning model based on the embedded vector map using a composite loss function can be: the loss values ​​calculated based on the three loss functions are added with the same weight ratio to obtain a total loss value, and the network parameters of the encoder and decoder are continuously updated with the total loss value, so that the neural network of the vector map learning model gradually converges.

[0172] By designing a specific composite loss function, the vector map learning model in this embodiment can take into account the transparency of the watermark, ensure the overall visual quality of the vector map, and take into account the robustness of the watermark, thereby achieving multi-objective optimization of the vector map learning model.

[0173] On the other hand, the present application embodiment also provides a vector map watermark embedding system, please refer to the attached Figure 5 ,include:

[0174] Model building module 1, constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is an original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data;

[0175] Model training module 2, randomly selecting a vector map as a training data set, and training the vector map learning model based on the training data set;

[0176] A first vector map generation module 3 inputs the original vector map into the trained vector map learning model, and outputs a vector map containing geographic feature data, which is recorded as a first vector map;

[0177] The watermark generation module 4 uses a hash algorithm to generate a watermark bit stream of the original vector map;

[0178] The watermark embedding module 5 adopts a strategy of dynamically adjusting the watermark embedding strength to embed the watermark bit stream into the geographic feature data of the first vector map to obtain a vector map after the watermark is embedded.

[0179] It can be understood that the technical concept of a vector map watermark embedding system provided in this embodiment is similar to the technical concept of the aforementioned vector map watermark embedding method, and this embodiment will not be repeated here.

[0180] The above description is only the preferred embodiment disclosed in this application and the description of the technical principle used. Those skilled in the art should understand that the scope of protection involved in this disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but also should cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this disclosure (but not limited to) to form a technical solution.

[0181] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

Claims

1. A vector map watermark embedding method, characterized in that: The following steps are involved: Step 102, constructing a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is the original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is the vector map containing the geographic feature data; Step 104, randomly selecting a vector map as a training data set, and training the vector map learning model based on the training data set; Step 106, inputting the original vector map into the trained vector map learning model, and outputting a vector map containing geographic feature data, which is recorded as a first vector map; Step 108, using a hash algorithm to generate a watermark bit stream of the original vector map; Step 110, using a strategy of dynamically adjusting the watermark embedding strength to embed the watermark bit stream into the geographic feature data of the first vector map, to obtain a vector map after the watermark is embedded; Wherein, step 110 includes: Step 202, dividing the first vector map into a plurality of sub-areas; Step 204, calculating the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region; Step 206, embedding the watermark bit stream into the geographic feature data of each sub-region respectively based on the watermark embedding strength of each sub-region; Furthermore, the geographic feature data includes point feature data, line feature data and surface feature data, and step 204 includes: Determine the density of the sub-region based on the proportion of the number of feature points of the midpoint feature data, line feature data, and surface feature data in the sub-region; Determining a first watermark embedding strength of the sub-region based on the density of the sub-region, wherein the first watermark embedding strength is proportional to the density of the sub-region; In each sub-region, a plurality of feature points in the geographic feature data are selected as embedding points according to the embedding strength of the first watermark; Clustering all embedded points in the first vector map using a clustering algorithm to obtain a central reference point; Calculate the distance from the embedded point in each sub-region to the central reference point; Based on the distance from the embedding point in each sub-region to the central reference point, the second watermark embedding strength of each sub-region is determined, and the second watermark embedding strength is inversely proportional to the distance from the embedding point in each sub-region to the central reference point.

2. A vector map watermark embedding method as claimed in claim 1, characterized in that: The geographic feature data includes point feature data, line feature data and surface feature data. In step 206, based on the watermark embedding strength of each sub-region, the watermark bit stream is embedded in the geographic feature data of each sub-region, including: In each sub-region, a number of feature points in the geographic feature data are selected as embedding points according to the determined watermark embedding strength; According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the feature points of the corresponding geographic feature data.

3. A vector map watermark embedding method as claimed in claim 2, characterized in that: According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the feature point coordinate value of the corresponding geographic feature data, including: If the embedded point is point feature data, the watermark bit stream is embedded in the decimal place of the coordinate value of the point feature data. If the embedded point is line feature data, the watermark bit stream is embedded in the decimal place of the feature point coordinate value of the line feature data. If the embedded point is surface feature data, the watermark bit stream is embedded in the decimal places of the coordinate values ​​of the surface boundary points of the surface feature data.

4. A vector map watermark embedding method as claimed in claim 2, characterized in that: According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the feature point coordinate value of the corresponding geographic feature data, including: Get the preset watermark embedding threshold; In the process of embedding the watermark bit stream into the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data, it is determined in real time whether the difference between the embedded feature point coordinate values ​​and the original feature point coordinate values ​​exceeds the embedding threshold. If so, the part of the watermark bit stream exceeding the embedding threshold will no longer be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data. If not, the watermark bit stream will continue to be embedded in the decimal places of the feature point coordinate values ​​of the corresponding geographic feature data.

5. A vector map watermark embedding method as claimed in claim 2, characterized in that: According to the type of geographic feature data corresponding to the embedding point, the watermark bit stream is embedded in the decimal places of the feature point coordinate value of the corresponding geographic feature data, including: Get the preset watermark embedding threshold; Dividing the watermark bit stream into several watermark bit groups bit by bit based on the embedding threshold of the watermark; According to the type of geographic feature data corresponding to the embedded point, several watermark bit groups are embedded bit by bit in the decimal places of the feature point coordinate value of the corresponding geographic feature data, so that the difference between the feature point coordinate value after the watermark bit group is embedded and the original feature point coordinate value is lower than the embedding threshold.

6. A vector map watermark embedding method as claimed in claim 1, characterized in that: After step 110, the method further includes: Step 111, using a composite loss function, based on the embedded vector map, to optimize the vector map learning model, the composite loss function includes an MSE Loss function, an SSIM Loss function and an Adversarial Loss function, wherein the MSE Loss function is used to measure the pixel-level difference between the vector map before and after watermark embedding, the SSIM Loss function is used to evaluate the overall visual quality of the vector map after watermark embedding, and the Adversarial Loss function is used to determine whether the vector map after watermark embedding is authentic.

7. A vector map watermark embedding system, characterized in that: include: A model building module, which builds a vector map learning model based on a deep neural network architecture including an encoder and a decoder, wherein the input of the vector map learning model is an original vector map, the output of the encoder is the geographic features of the original vector map, and the output of the decoder is a vector map containing geographic feature data; A model training module randomly selects a vector map as a training data set, and trains the vector map learning model based on the training data set; A first vector map generation module, inputting the original vector map into the trained vector map learning model, and outputting a vector map containing geographic feature data, which is recorded as a first vector map; The watermark generation module uses a hash algorithm to generate the watermark bit stream of the original vector map; A watermark embedding module, which embeds the watermark bit stream into the geographic feature data of the first vector map by adopting a strategy of dynamically adjusting the watermark embedding strength, so as to obtain a vector map after the watermark is embedded; The watermark embedding module is used to perform the following steps: Step 202, dividing the first vector map into a plurality of sub-areas; Step 204, calculating the watermark embedding strength of each sub-region based on the geographic feature data in the sub-region; Step 206, embedding the watermark bit stream into the geographic feature data of each sub-region respectively based on the watermark embedding strength of each sub-region; Furthermore, the geographic feature data includes point feature data, line feature data and surface feature data, and step 204 includes: Determine the density of the sub-region based on the proportion of the number of feature points of the midpoint feature data, line feature data, and surface feature data in the sub-region; Determining a first watermark embedding strength of the sub-region based on the density of the sub-region, wherein the first watermark embedding strength is proportional to the density of the sub-region; In each sub-region, a plurality of feature points in the geographic feature data are selected as embedding points according to the embedding strength of the first watermark; Clustering all embedded points in the first vector map using a clustering algorithm to obtain a central reference point; Calculate the distance from the embedded point in each sub-region to the central reference point; Based on the distance from the embedding point in each sub-region to the central reference point, the second watermark embedding strength of each sub-region is determined, and the second watermark embedding strength is inversely proportional to the distance from the embedding point in each sub-region to the central reference point.

Images