Map scattering point data dynamic generation method and system based on generative adversarial network
By dynamically generating point data through generative adversarial networks, the problem of the inability to adjust the point layout in real time in existing technologies is solved, realizing the real-time nature and diversity of point data, and improving the intelligence level of security systems and public safety protection capabilities.
Patent Information
- Application Number
- CN202510982695.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-16
- Publication Date
- 2025-10-31
AI Technical Summary
Existing spotting technology cannot be dynamically adjusted in real time, lacks linkage with real-time data sources, has a slow response speed, a single layout mode, poor flexibility, and is difficult to adapt to complex environments and different security task requirements, and has a low level of intelligence.
Generative adversarial networks are used to generate point data. By collecting environmental features and historical case data, a condition generator model and a discriminator are used to judge the authenticity of the point data and iteratively train it to dynamically generate real-time point data.
It enables real-time generation and dynamic adaptability of data distribution points, improves the naturalness and rationality of distribution point layout, enhances the identification and coverage capabilities of high-risk areas, supports flexible adaptation to multiple scenarios, reduces manual planning costs, and improves the efficiency of security resource allocation and public safety assurance capabilities.
Smart Images
Figure CN120875375A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent security technology, and in particular to a method and system for dynamically generating map data based on generative adversarial networks. Background Technology
[0002] In fields such as security monitoring, urban management, and emergency response, the rational layout of deployment points (i.e., resource deployment locations, such as cameras, patrol points, and sensors) is key to improving system efficiency. Existing deployment point technologies include the following two types:
[0003] I. Traditional Static Spraying Technique
[0004] 1. Construction Description:
[0005] like Figure 1 As shown, the system is based on a Geographic Information System (GIS) platform; the locations of data points (such as cameras and patrol points) are manually preset; databases (such as PostGIS and MongoDB) are used to manage the data point location information; and the front-end system (such as a monitoring center map) retrieves the coordinates of the data points through an interface and renders and displays them.
[0006] 2. Component location and connection relationship:
[0007] The GIS map engine connects to the data point database; the data point database is manually updated periodically; the display system reads the database content in real time for visualization.
[0008] 3. Technological Process:
[0009] The application site planning is manually formulated; the application site locations are statically written into the database; and the results are queried and displayed through an API interface.
[0010] 4. Existing problems:
[0011] Because static databases are used to store location information, there is a lack of linkage mechanism with real-time data sources (such as real-time population flow and alarm event flow); updates to the locations usually require manual intervention or periodic batch updates, making it difficult to adjust the location layout in a timely manner according to dynamic environmental factors such as real-time changes in population flow, case distribution, and emergencies, resulting in slow response speed; the location layout pattern is monotonous and lacks natural diversity.
[0012] II. Heatmap-based Spreading Expansion Technique
[0013] 1. Construction Description:
[0014] like Figure 2 As shown, a heatmap is generated using user location and historical event points; additional data points are manually added or existing data points are adjusted based on the heatmap area; and a rule engine is used to set rules, such as "increase data points in densely populated areas".
[0015] 2. Component location and connection relationship:
[0016] The system includes a data collection module (collecting pedestrian / police information); a heatmap analysis module (generating hotspot areas); a rules engine module (setting and adjusting data distribution strategies based on hotspot areas); and a map engine for displaying results.
[0017] 3. Technological Process:
[0018] Collect event / people flow data in real time or periodically; generate heat maps; and apply manually set rules to generate new distribution point layouts.
[0019] 4. Existing problems:
[0020] Relying on pre-defined rules, rule engines are usually based on simple logic (such as "more points should be set up if there is a large flow of people"), which makes it impossible to capture complex environmental features. The point setting layout is highly repetitive, lacks flexibility, and is difficult to adapt to the actual needs of different geographical environments and different security tasks. This results in point distribution that is either too dense or too sparse. In addition, the rule engine is designed as a static rule set and lacks a dynamic parameter adjustment mechanism. Once point setting rules or templates are set, it is difficult to flexibly switch point layouts according to different application scenarios (such as major events, daily patrols, and emergency responses).
[0021] Furthermore, the point-spreading logic relies excessively on surface-visible data (such as heat maps) and ignores implicit pattern recognition (such as predicting potential risks through historical cases); it also fails to introduce data mining or generative models to actively simulate the distribution of potential threats, lacks the ability to model possible future risks, and is difficult to cover unknown patterns or emergencies; the point-spreading adjustment strategy is based on preset rules rather than real-time learning, resulting in a lag between changes in heat maps and changes in point spreads; a large amount of manual intervention is still required to formulate point-spreading rules, and the level of intelligence is low. Summary of the Invention
[0022] To address the problems existing in the prior art, this invention provides a method for dynamically generating map point data based on generative adversarial networks, comprising:
[0023] Step S1: Collect environmental feature data and historical security case data under different application scenarios from security-related maps, and perform standardized encoding processing on the environmental feature data and the historical security case data to obtain input feature vectors;
[0024] Step S2: Input the input feature vector into the condition generator model to generate scattered data;
[0025] Step S3: Use a discriminator to determine the authenticity of the scattered data to obtain the authenticity probability, and determine whether the authenticity probability is less than a preset threshold.
[0026] If so, the parameters of the generator model are adjusted, and then the process returns to step S2 to iteratively train the condition generator model.
[0027] If not, the generator model is deployed to the security map system, and then real-time point data is dynamically generated based on the current environmental feature data associated with the security map system.
[0028] Preferably, in step S1, the security-related map is divided into multiple grids, and the environmental feature data and historical security case data are associated with the corresponding grids.
[0029] Preferably, the environmental feature data includes at least one of geographical environment data, population density data, and special protection target data.
[0030] Preferably, in step S1, the standardization encoding process includes:
[0031] The continuous values in the environmental feature data and the historical security case data are normalized and the discrete categories are one-hot encoded. Then, the normalized and one-hot encoded feature values are concatenated to form the input feature vector.
[0032] Preferably, in step S2, the conditional generation model predicts the probability of point placement within each grid based on the input feature vector, and outputs the point coordinates, point density, and point importance of the grid whose point placement probability is greater than a preset probability value as the point placement data.
[0033] Preferably, the application scenario is the input of external condition information to the condition generator model.
[0034] Preferably, in step S3, determining the authenticity probability of the scattered data includes:
[0035] The rationality of the distribution of the scattered points is assessed based on the coordinates of the points, the degree of matching between the scattered point density and the environmental characteristics is assessed based on the density of the points, and the spatial correlation between the scattered points and the key protection areas is assessed based on the importance of the points.
[0036] The probability of authenticity is calculated based on the rationality of the distribution of the scattered points, the degree of matching, and the spatial correlation.
[0037] Preferably, a feedback interface is provided for receiving security event data and fine-tuning the condition generator model online based on the security event data.
[0038] Preferably, the dynamic generation of real-time data distribution includes:
[0039] Dynamic triggering conditions are deployed in the security map system, wherein the dynamic triggering conditions are timed triggering and / or event-driven triggering.
[0040] This invention also provides a dynamic map point data generation system based on generative adversarial networks. Applying the above-described dynamic map point data generation method, the dynamic map point data generation system includes:
[0041] The data acquisition and processing module is used to collect environmental feature data and historical security case data under different application scenarios from security-related maps, and to perform standardized encoding processing on the environmental feature data and the historical security case data to obtain input feature vectors;
[0042] The data generation module is connected to the data acquisition and processing module and is used to input the input feature vector into the condition generator model to generate data.
[0043] The adversarial training optimization module, connected to the data generation module, is used to use a discriminator to determine the authenticity of the data and obtain the authenticity probability. When the authenticity probability is less than a preset threshold, the generator model is adjusted in terms of parameters and the conditional generator model is iteratively trained until the authenticity probability is not less than the preset threshold. Then, the generator model is deployed to the security map system, and real-time data is dynamically generated based on the current environmental feature data associated with the security map system.
[0044] The above technical solution has the following advantages or beneficial effects:
[0045] 1) Achieve dynamic real-time generation of data: Real-time data is generated dynamically through generative adversarial networks. New data can be intelligently generated within hundreds of milliseconds (average inference time of about 180ms) based on real-time changing environmental characteristics. Compared with the traditional static database data update cycle (usually 24 to 72 hours), the response speed is improved by more than 480 times, achieving real-time performance and dynamic adaptability of data.
[0046] 2) Improve the naturalness and rationality of the point distribution layout: Through the adversarial training mechanism between the generator model and the discriminator, the generated point distribution data is improved by about 27% in terms of spatial distribution pattern and diversity index (based on spatial entropy assessment) compared with the traditional rule-based point distribution method. It is more in line with the actual threat distribution pattern and significantly improves the naturalness and spatial rationality of the point distribution.
[0047] 3) Enhance the identification and coverage of high-risk areas: By introducing generative adversarial learning, historical security case data and real-time environmental feature data are used to train the condition generator model to automatically learn and predict potential risk areas, thereby effectively covering hidden high-risk areas when applying heat maps, improving security and prevention capabilities. Test results show that, under the same number of heat map application points, the potential risk area coverage rate of the proposed solution is about 32.6% higher than that of the traditional heat map application method.
[0048] 4) Supports flexible adaptation to multiple scenarios: By introducing a condition input mechanism through the condition generator model, the condition generator model can dynamically adjust the point-spreading strategy according to different application scenario inputs (such as daily patrols, major event security, and emergency response). The time to switch point-spreading styles is less than 1 second, meeting diverse security needs.
[0049] 5) Reduce the cost of manual planning for security deployment: The automatic generation of the deployment layout reduces the reliance on professional planners and is expected to save about 40%-60% of deployment planning costs;
[0050] 6) Improve the efficiency of security resource allocation: Through more precise deployment, the same number of security resources can achieve more efficient prevention and control coverage, reducing the overall security investment cost;
[0051] 7) Enhance public safety protection capabilities: This invention can respond quickly and dynamically to environmental changes, improve the speed of perception and response to sudden safety incidents, and enhance the level of public safety protection;
[0052] 8) Promote the intelligent upgrade of security systems: By introducing generative AI models, promote the transformation of the security industry from rule-driven to data-driven and intelligent-driven, and help build smart cities and smart security systems. Attached Figure Description
[0053] Figure 1 This is a structural diagram of traditional static spot application technology;
[0054] Figure 2 This is a schematic diagram of the existing heatmap-based point-spreading technology.
[0055] Figure 3 A flowchart illustrating a method for dynamically generating map point data based on generative adversarial networks, as a preferred embodiment of the present invention.
[0056] Figure 4 A schematic diagram of a map point data dynamic generation system based on generative adversarial networks is shown in a preferred embodiment of the present invention. Detailed Implementation
[0057] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment; other embodiments that conform to the spirit of the present invention may also fall within the scope of the present invention.
[0058] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, a method for dynamically generating map point data based on generative adversarial networks is provided, such as... Figure 3 As shown, it includes:
[0059] Step S1: Collect environmental feature data and historical security case data from security-related maps for different application scenarios, and perform standardized encoding processing on the environmental feature data and historical security case data to obtain the input feature vector;
[0060] Step S2: Input the input feature vector into the condition generator model to generate scattered data;
[0061] Step S3: Use a discriminator to determine the authenticity of the scattered data to obtain the authenticity probability, and determine whether the authenticity probability is less than a preset threshold.
[0062] If so, the parameters of the generator model are adjusted, and then the process returns to step S2 to iteratively train the conditional generator model.
[0063] If not, the generator model will be deployed to the security map system, and then real-time point data will be dynamically generated based on the current environmental feature data associated with the security map system.
[0064] Specifically, in this embodiment, the aforementioned environmental feature data and historical security case data are basic data closely related to the point-based detection in the relevant map information, providing a multi-dimensional input feature basis for point-based detection generation.
[0065] The environmental characteristic data mainly includes:
[0066] • Geographical environmental data (such as building distribution, road structure, and area classification);
[0067] • Crowd density data (e.g., obtained based on communication data, Wi-Fi probes, camera thermal data, etc.);
[0068] • Data on special protection targets (such as government agencies, large public facilities, important units, transportation hubs, and areas where people gather).
[0069] Historical security case data mainly includes records of public security cases and high-incidence incidents, and mainly involves case distribution, event types, and time characteristics.
[0070] Furthermore, the collected data is spatially encoded in a rasterized form. That is, in step S1, the security-related map is divided into multiple grids, and environmental feature data and historical security case data are associated with the corresponding grids.
[0071] To ensure that different types of data can be used as effective input to the neural network model, this embodiment performs standardized encoding processing on the collected heterogeneous data to form a unified feature vector, which is used as input to the conditional generator model. Specifically, this includes:
[0072] The continuous values in the environmental feature data and historical security case data are normalized and the discrete categories are one-hot encoded. Then, the normalized and one-hot encoded feature values are concatenated to form the input feature vector.
[0073] Continuous values are normalized, such as population density being normalized to the [0,1] interval. Discrete categories are processed using one-hot encoding, such as region usage labeling. These features are then concatenated into a unified input feature vector for use in the subsequent input conditional generator model.
[0074] In a preferred embodiment of the present invention, in step S2, the conditional generation model predicts the probability of point placement within each grid based on the input feature vector, and outputs the point coordinates, point density, and point importance of grids with point placement probabilities greater than a preset probability value as point placement data.
[0075] Specifically, in this embodiment, the condition generator model is mainly used to intelligently generate a point-scattering layout that meets actual needs based on different environmental characteristics. The condition generator model employs a conditional generative adversarial network (cGAN), preferably constructed using a multi-layer convolutional neural network (CNN) or an attention mechanism network (Transformer-based) to ensure the capture of spatial distribution characteristics; it supports conditional generation (Conditional GAN, cGAN) and can dynamically control the point-scattering style based on input features. Here, the point coordinates are latitude and longitude or raster IDs, the point density is the number of points per unit area (i.e., the number of points scattered in each region), and the point importance is used to distinguish between general and critical points, thereby differentiating between ordinary and high-priority points.
[0076] This invention is not limited to specific types of generative adversarial network architectures, such as GAN, cGAN, WGAN-GP, etc., and can be adapted for application.
[0077] In a preferred embodiment of the present invention, the application scenario is the input of external condition information of the condition generator model. The application scenario can be set and adjusted according to needs, including but not limited to daily patrols, security for major events, and emergency response.
[0078] Furthermore, the discriminator takes both real and generated point-scattering data as input and uses a binary classification task to determine the authenticity and reasonableness of the generated point-scattering data, whether it conforms to the real point-scattering distribution pattern, thereby guiding the generator model to improve the authenticity, reasonableness, and diversity of the point-scattering results. The discriminator employs a convolutional network (CNN) or multi-scale discriminator design to determine the authenticity probability of the point-scattering data, including:
[0079] The rationality of the distribution of the application points is assessed based on the coordinates of the application points; the degree of matching between the application point density and environmental characteristics is assessed based on the density of the application points; and the spatial correlation between the application points and key protection areas is assessed based on the importance of the application points.
[0080] The probability of authenticity is calculated based on the rationality of the distribution of the scattered points, the degree of matching, and the spatial correlation.
[0081] In other words, the discriminator's task is to determine whether the data points originate from real historical data point distributions and to evaluate the rationality of the data point layout, including spatial uniformity, hotspot coverage, and the rationality of blank areas. The discriminator outputs the probability of authenticity as a training feedback signal.
[0082] The following sections explain the rationality of the distribution of spray points, the degree of matching between the spray point density and environmental characteristics, the spatial correlation between the spray points and key protection areas, and the calculation method for the probability of accuracy:
[0083] 1. Assessment of the rationality of the distribution of the scattered points
[0084] Objective: To determine whether the sprinkled points exhibit a reasonable spatial distribution structure, avoiding dense accumulation or abnormal voids.
[0085] 1.1 Evaluation Method:
[0086] Based on Ripley's K function or mean nearest neighbor distance (NND):
[0087] Formula (NND):
[0088]
[0089] Where Dobs is the average nearest neighbor distance of the observation points;
[0090] Drand is the average nearest neighbor distance for randomly scattered points in the same region with the same number of points.
[0091] When R1≈1, it represents a random distribution; when R1<1, it represents clustering; and when R1>1, it represents uniformity.
[0092] 1.2 Calculation of Reasonableness Score:
[0093] Normalization is performed using the Gaussian function:
[0094]
[0095] Wherein, S1 is the score for the rationality of the distribution of the scattered points;
[0096] μ is the ratio of the empirically optimal distribution (e.g., 1.0);
[0097] σ is a hyperparameter that adjusts the sensitivity of the rating.
[0098] 2. The degree of matching between the scattering density and environmental characteristics (R2)
[0099] Objective: Whether the density of the sampling points is consistent with environmental factors such as population density and key areas.
[0100] 2.1 Evaluation Method:
[0101] The Pearson correlation coefficient was used to measure the spatial matching degree between the point density map P(x,y) and the environmental weight map E(x,y).
[0102]
[0103] Where Pi is the density of the dots in the i-th grid;
[0104] Ei represents the environmental feature score of the i-th grid;
[0105] Pˉ, Eˉ: These are the mean values of the scattering density and the environmental score, respectively.
[0106] 2.2 Match score normalization:
[0107]
[0108] Compress the results to the range [0,1].
[0109] 3. Spatial correlation between the scattered application points and key protection areas
[0110] Objective: To determine whether the sampling points are close to high-risk areas or key targets (such as subway entrances, schools, government units, etc.).
[0111] 3.1 Evaluation Method:
[0112] Calculate the average distance from the scattered points to the nearest key region, and introduce a Gaussian penalty function:
[0113]
[0114] Where pi is the coordinate of the i-th point;
[0115] zj represents the coordinates of the j-th key protection area;
[0116] d(pi,zj) is the Euclidean distance;
[0117] Z represents the set of key regions;
[0118] N represents the total number of points to be scattered.
[0119] 3.2 Spatial Relationship Score:
[0120]
[0121] Where σd represents the radius of influence of the high-risk area (e.g., 200 meters).
[0122] 4. Calculation method for the probability of authenticity
[0123] The final probability of authenticity (Preal) can be achieved by integrating the scores of each item using a weighted average:
[0124] P real =w1S1+w2S2+w3S3
[0125] S1, S2, S3: Corresponding to the scores of the three evaluation items mentioned above;
[0126] w1, w2, w3: Weights of each item, satisfying w1+w2+w3=1, which can be automatically optimized through experimental settings or cross-validation.
[0127] Furthermore, the generator and discriminator continuously optimize the performance of the conditional generator model through adversarial training, specifically as follows:
[0128] The generator produces point data based on the current parameters;
[0129] The discriminator distinguishes between real and generated data points:
[0130] If the probability of authenticity is less than a preset threshold, the generated data is considered fake. The conditional generator then adjusts its parameters according to a loss function (such as Wasserstein Loss or standard GAN Loss). Based on the adjusted parameters, the generator trains for the next round of data generation, forming an iterative adversarial training process between the generator and the discriminator. This creates a closed-loop training cycle that continuously improves the realism and reasonableness of the generated data, gradually approximating the distribution of real data. The training process iterates until the discriminator can no longer distinguish between generated and real data, meaning the generator can generate sufficiently natural and reasonable data distributions. Ultimately, this results in a conditional generator model that can dynamically generate highly reliable data based on the characteristics of the input environment.
[0131] The loss function can be the standard GAN Loss or WGAN (Wasserstein GAN) to improve training stability, or a diversity loss (such as feature divergence) can be added to encourage a diverse distribution of the points.
[0132] Once the condition generator model is trained, it can be deployed as an inference engine to dynamically generate data points online based on real-time input features (such as current pedestrian flow and changes in police activity).
[0133] In a preferred embodiment of the present invention, a feedback interface is provided for receiving security event data and fine-tuning the condition generator model online based on the security event data.
[0134] Specifically, in this embodiment, the aforementioned security event data includes, but is not limited to, the coverage rate and response time of alarms occurring in the monitored areas. By performing online fine-tuning on the condition generator model, the accuracy and adaptability of the monitoring points are further improved.
[0135] In a preferred embodiment of the present invention, dynamically generating real-time data distribution data includes:
[0136] Deploy dynamic triggering conditions in the security map system. These dynamic triggering conditions can be timed triggers or event-driven triggers.
[0137] Specifically, in this embodiment, when the dynamic trigger condition is timed triggering, the timer period can be set according to requirements, such as refreshing once per hour, in which case the point layout is refreshed hourly. When the dynamic trigger condition is event-driven triggering, the point layout is refreshed when a driving event is detected, wherein driving events include, but are not limited to, major event warnings and emergency alarms. The point data is preferably pushed to the map display system through a standard interface to achieve visualized deployment.
[0138] After the model training is complete, the condition generator model is deployed in the security map system to perform the following operations:
[0139] 1. Real-time collection of current environmental characteristic data;
[0140] 2. Input features into the trained conditional generator model;
[0141] 3. Quickly generate scattered data layout (average inference latency less than 200ms);
[0142] 4. Map the data points to the actual security map interface for maintenance personnel to refer to or automatically generate deployment plans.
[0143] It is understood that the map point data dynamic generation system of the present invention can be applied to various security scenarios such as static point planning, dynamic security deployment, and emergency response deployment. Moreover, the point data can be used not only for physical deployment (such as cameras and police force allocation) but also for virtual simulation and drill systems.
[0144] This invention also provides a dynamic map point data generation system based on generative adversarial networks, applying the aforementioned dynamic map point data generation method, such as... Figure 4 As shown, the map point data dynamic generation system includes:
[0145] The data acquisition and processing module 1 is used to collect environmental feature data and historical security case data under different application scenarios from security-related maps, and to perform standardized encoding processing on the environmental feature data and historical security case data to obtain input feature vectors;
[0146] The point data generation module 2 is connected to the data acquisition and processing module 1 and is used to input the input feature vector into the condition generator model to generate point data.
[0147] The adversarial training optimization module 3 is connected to the point data generation module 2. It is used to use a discriminator to judge the authenticity of the point data and obtain the authenticity probability. When the authenticity probability is less than a preset threshold, the parameters of the generator model are adjusted and the condition generator model is iteratively trained until the authenticity probability is not less than the preset threshold. Then, the generator model is deployed to the security map system, and then real-time point data is dynamically generated based on the current environmental feature data associated with the security map system.
[0148] In summary, to address the problems of long update cycles and inability to dynamically adjust existing data distribution data according to environmental changes, this invention proposes a dynamic data distribution generation method based on Generative Adversarial Networks (GANs). This method can intelligently generate new data distribution data based on real-time changing environmental characteristics, achieving real-time performance and dynamic adaptability of the data distribution.
[0149] To address the issues of existing point distribution patterns being too simplistic and lacking naturalness and diversity, this invention employs an adversarial training mechanism to generate point data with richer spatial distribution characteristics and more natural point distribution layouts that better reflect actual threat distribution patterns, thereby enhancing the realism and diversity of point distribution strategies.
[0150] To address the problem that existing spot-marking systems cannot effectively identify potential high-risk areas, this invention introduces generative adversarial learning. By utilizing historical case data and real-time environmental data, a generator is trained to automatically learn and predict potential risk areas, thereby effectively covering hidden high-risk areas during spot-marking and improving security and prevention capabilities.
[0151] To address the issues of poor flexibility and difficulty in quickly switching the distribution layout of existing distribution strategies according to different application scenarios, this invention uses an input feature control mechanism to enable the generated model to automatically adjust the distribution density and layout style according to different scenario inputs (such as patrols, major events, and emergency responses), thereby achieving dynamic adaptation and self-optimization of the distribution scheme.
[0152] The main beneficial effects of this invention include:
[0153] 1. Technical effects
[0154] • Achieve dynamic real-time generation of data points.
[0155] This invention allows the generator to update the point layout within hundreds of milliseconds (average inference time of about 180ms) by inputting environmental features in real time (such as changes in pedestrian flow and police situation). Compared with the traditional static database point update cycle (which usually takes 24 to 72 hours), the response speed is improved by more than 480 times.
[0156] • Enhance the naturalness and rationality of the sprinkler layout.
[0157] By generating adversarial training mechanisms, the scattered points generated by this invention are about 27% better than traditional rule-based scattered point methods in terms of spatial distribution patterns and diversity indicators (based on spatial entropy assessment). This makes the scattered points more consistent with the actual threat distribution patterns and significantly improves the naturalness and spatial rationality of the scattered points.
[0158] • Enhance the ability to identify and cover high-risk areas
[0159] This invention utilizes historical case data to train a discriminator and a generator, enabling the point-scattering layout to proactively cover potentially high-risk areas. Test results show that, under the same number of points, the potential risk area coverage rate of this invention is approximately 32.6% higher than that of traditional heatmap point-scattering methods.
[0160] Supports flexible adaptation across multiple scenarios.
[0161] This invention introduces a Conditional GAN mechanism, which can dynamically adjust the spotting strategy according to different application scenarios (such as daily patrols, major event security, and emergency response). The time to switch spotting styles is less than 1 second, meeting diverse security needs.
[0162] 2. Economic benefits
[0163] • Reduce the manual planning costs for security deployment
[0164] The automatic generation of the distribution layout reduces reliance on professional planners and is expected to save approximately 40%-60% in deployment planning costs.
[0165] • Improve the efficiency of security resource allocation
[0166] By deploying security points more precisely, more efficient prevention and control coverage can be achieved with the same amount of security resources, reducing overall security investment costs.
[0167] 3. Social effects
[0168] • Enhance public safety capabilities
[0169] This invention can respond quickly and dynamically to environmental changes, improve the speed of perception and response to sudden safety incidents, and enhance the level of public safety protection.
[0170] Promote the intelligent upgrading of security systems
[0171] By introducing generative AI models (GANs), we can promote the transformation of the security industry from rule-driven to data-driven and intelligence-driven, and help build smart cities and smart security systems.
[0172] The above description is merely a preferred embodiment of the present invention and does not limit the implementation and protection scope of the present invention. Those skilled in the art should realize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.
Claims
1. A method for dynamically generating map point data based on generative adversarial networks, characterized in that, include: Step S1: Collect environmental feature data and historical security case data under different application scenarios from security-related maps, and perform standardized encoding processing on the environmental feature data and the historical security case data to obtain input feature vectors; Step S2: Input the input feature vector into the condition generator model to generate scattered data; Step S3: Use a discriminator to determine the authenticity of the scattered data to obtain the authenticity probability, and determine whether the authenticity probability is less than a preset threshold. If so, the parameters of the generator model are adjusted, and then the process returns to step S2 to iteratively train the condition generator model. If not, the generator model is deployed to the security map system, and then real-time point data is dynamically generated based on the current environmental feature data associated with the security map system.
2. The method for dynamically generating map point data according to claim 1, characterized in that, In step S1, the security-related map is divided into multiple grids, and the environmental feature data and historical security case data are associated with the corresponding grids.
3. The method for dynamically generating map point data according to claim 1, characterized in that, The environmental feature data includes at least one of the following: geographical environment data, population density data, and special protection target data.
4. The method for dynamically generating map point data according to claim 1, characterized in that, In step S1, the standardization encoding process includes: The continuous values in the environmental feature data and the historical security case data are normalized and the discrete categories are one-hot encoded. Then, the normalized and one-hot encoded feature values are concatenated to form the input feature vector.
5. The method for dynamically generating map point data according to claim 2, characterized in that, In step S2, the conditional generation model predicts the probability of scattering points within each grid based on the input feature vector, and outputs the point coordinates, point density, and point importance of the grid whose scattering probability is greater than a preset probability value as the scattering data.
6. The method for dynamically generating map point data according to claim 1, characterized in that, The application scenario is the input of external condition information to the condition generator model.
7. The method for dynamically generating map point data according to claim 5, characterized in that, In step S3, determining the authenticity probability of the scattered data includes: The rationality of the distribution of the scattered points is assessed based on the coordinates of the points, the degree of matching between the scattered point density and the environmental characteristics is assessed based on the density of the points, and the spatial correlation between the scattered points and the key protection areas is assessed based on the importance of the points. The probability of authenticity is calculated based on the rationality of the distribution of the scattered points, the degree of matching, and the spatial correlation.
8. The method for dynamically generating map point data according to claim 1, characterized in that, A feedback interface is provided to receive security event data and to fine-tune the condition generator model online based on the security event data.
9. The method for dynamically generating map point data according to claim 1, characterized in that, The dynamically generated real-time data distribution data includes: Dynamic triggering conditions are deployed in the security map system, wherein the dynamic triggering conditions are timed triggering and / or event-driven triggering.
10. A dynamic map point generation system based on generative adversarial networks, characterized in that, The map point data dynamic generation system, using the map point data dynamic generation method as described in any one of claims 1-9, comprises: The data acquisition and processing module is used to collect environmental feature data and historical security case data under different application scenarios from security-related maps, and to perform standardized encoding processing on the environmental feature data and the historical security case data to obtain input feature vectors; The data generation module is connected to the data acquisition and processing module and is used to input the input feature vector into the condition generator model to generate data. The adversarial training optimization module, connected to the data generation module, is used to use a discriminator to determine the authenticity of the data and obtain the authenticity probability. When the authenticity probability is less than a preset threshold, the generator model is adjusted in terms of parameters and the conditional generator model is iteratively trained until the authenticity probability is not less than the preset threshold. Then, the generator model is deployed to the security map system, and real-time data is dynamically generated based on the current environmental feature data associated with the security map system.