Deep reinforcement learning multi-displacement processing method for facing flat annotation cover conflict

Abstract

The application belongs to the technical field of data processing, and provides a deep reinforcement learning multi-displacement processing method for planar annotation cap conflict, comprising: determining candidate planar annotations with cap conflict and the complexity of each cap conflict event in the target area, and determining the target displacement mode of the candidate planar annotations under the corresponding complexity; determining the cap number, cap area and semantic importance of each candidate planar annotation, and calculating the comprehensive score of the conflict importance of each candidate planar annotation; determining the agent selection strategy under the corresponding target displacement mode in each cap conflict event according to the comprehensive score of the conflict importance; and controlling the candidate planar annotations to perform displacement actions until the optimal position according to the target displacement mode and the agent selection strategy, in combination with the pre-established DQN model and displacement constraint reward strategy. The scheme improves the efficiency, accuracy, flexibility and reliability of the planar annotation cap conflict processing link.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a deep reinforcement learning multi-displacement processing method for surface annotation overlay conflicts. Background Technology

[0002] In the field of map annotation configuration, annotations, as a core element for conveying geographic information, directly affect map readability and usability through their automatic configuration. However, the automatic annotation configuration function of current mainstream GIS (Geographic Information System) software still has significant limitations, especially in handling conflicts caused by overlapping area annotations, which has become a key bottleneck in the automatic generation of geographic information.

[0003] Existing annotation shifting schemes are usually limited to a single shifting method, which is difficult to meet the comprehensive requirements of efficiency and flexibility in resolving overlapping conflicts in different overlapping conflict scenarios. With the development of deep learning technology, although some studies have applied neural networks and deep reinforcement learning to annotation recognition and point annotation configuration, there is still a gap in the handling of overlapping conflicts of area annotations. Existing deep reinforcement learning methods cannot dynamically match the shifting method according to the complexity of overlapping conflicts, nor have they established a sound agent selection mechanism. As a result, it is difficult to balance the conflict resolution efficiency and the stability of key annotation positions in complex scenarios, and it is impossible to effectively guide annotations to avoid new conflicts and maintain their correlation with the area elements they belong to during the shifting process.

[0004] It is evident that traditional annotation shifting methods suffer from technical problems such as insufficient efficiency, accuracy, flexibility, and reliability. Summary of the Invention

[0005] This invention provides a deep reinforcement learning multi-displacement processing method for surface annotation overlay conflicts, which addresses the shortcomings of traditional annotation displacement processing schemes that are not efficient and accurate enough, and lack scene adaptability and reliability.

[0006] This invention provides a deep reinforcement learning multi-displacement processing method for addressing overlapping conflicts of planar annotations, comprising: determining candidate planar annotations with overlapping conflicts within a target region and the complexity of each overlapping conflict event, and determining the target displacement method of the candidate planar annotations under the corresponding complexity; determining the overlapping quantity, overlapping area, and semantic importance of each candidate planar annotation, and calculating the comprehensive conflict importance score of each candidate planar annotation; determining an agent selection strategy under the corresponding target displacement method in each overlapping conflict event based on the comprehensive conflict importance score; and controlling the candidate planar annotations to perform displacement actions until the optimal position is reached based on the target displacement method and the agent selection strategy, combined with a pre-established DQN model and a displacement constraint reward strategy.

[0007] According to the method provided by the present invention, determining candidate planar annotations with overlapping conflicts within a target area includes: constructing a corresponding annotation rectangle for each planar annotation within the target area; determining whether there is spatial overlap between each annotation rectangle and other annotation rectangles to obtain a determination result; and taking the planar annotations with the determination result being yes as candidate planar annotations.

[0008] According to the method provided by the present invention, determining the complexity of each capping conflict event includes: determining the number of candidate surface annotations involved in capping in each capping conflict event; if the number of annotations is three or less, the complexity of the capping conflict event is determined to be annotation capping simple; if the number of annotations is more than three, the complexity of the capping conflict event is determined to be annotation capping complex.

[0009] According to the method provided by the present invention, determining the target displacement mode of the candidate area annotation under a corresponding complexity level includes: if the complexity of the current overlay conflict event is simple overlay, then the target displacement mode of the candidate area annotation is a fixed displacement mode, wherein, under the fixed displacement mode, the movement direction of the candidate area annotation is eight uniformly distributed directions with the center of the area feature as the origin; if the complexity of the current overlay conflict event is complex overlay, then the target displacement mode of the candidate area annotation is a free displacement mode, wherein, under the free displacement mode, the movement direction of the candidate area annotation is freely adjusted within a specified buffer zone.

[0010] According to the method provided by the present invention, based on the comprehensive score of conflict importance, a selection strategy for agents under the corresponding target displacement mode in each overlay conflict event is determined, including: under the fixed displacement mode, the agent selection strategy is: preferentially selecting one or two candidate surface annotations with low comprehensive conflict importance scores in the overlay conflict event as movable agents; under the free displacement mode, the agent selection strategy is: selecting one or two candidate surface annotations with high comprehensive conflict importance scores in the overlay conflict event as non-movable elements, and selecting all other candidate surface annotations other than non-movable elements as movable agents.

[0011] According to the method provided by the present invention, determining the number of overlaps, the area of ​​overlap, and the semantic importance of each candidate surface annotation includes: determining the number of times each candidate surface annotation overlaps with other candidate surface annotations in each overlap conflict event, and normalizing the number of overlaps to obtain the number of overlaps; determining the sum of the first overlap areas of each candidate surface annotation with other candidate surface annotations and the sum of the second overlap areas with surface elements not belonging to it in each overlap conflict event, adding the sum of the first overlap areas and the sum of the second overlap areas and normalizing them to obtain the area of ​​overlap; determining the target semantic level corresponding to the semantic information of each candidate surface annotation, and determining the semantic importance score corresponding to the target semantic level, and normalizing the semantic importance score to obtain the semantic importance.

[0012] According to the method provided by the present invention, determining the target semantic level corresponding to the semantic information of each candidate area annotation includes: determining the planning level, necessary information transmission content, industry standard information, and element spatial influence level corresponding to the semantic information of each candidate area annotation to obtain core information; and determining the target semantic level corresponding to the semantic information of the candidate area annotation from a pre-established core information and semantic level comparison table based on the core information.

[0013] According to the method provided by the present invention, the comprehensive conflict importance score of each candidate surface annotation is calculated, including: determining the factor weight values ​​corresponding to the number of overlays, the area of ​​overlays, and the semantic importance of each; and weighting and summing the number of overlays, the area of ​​overlays, and the semantic importance of each candidate surface annotation according to the factor weight values ​​to obtain the comprehensive conflict importance score.

[0014] According to the method provided by the present invention, a displacement constraint reward strategy is established, comprising: during the execution of a displacement action, if there is no overlap conflict between the movable intelligent body and other planar annotations, a positive value is set as a reward value; otherwise, the negative value of the absolute value of the overlap area between the movable intelligent body and other planar annotations is multiplied by a first set factor as a penalty value, thus obtaining a first overlap constraint reward strategy; if there is no overlap conflict between the movable intelligent body and planar elements not belonging to itself, a positive value is set as a reward value; otherwise, the negative value of the absolute value of the overlap area between the movable intelligent body and planar elements not belonging to itself is multiplied by a second set factor as a penalty value, thus obtaining a second overlap constraint reward strategy; if the target displacement method is a free displacement method, if the movable intelligent body does not exceed the set displacement range, a non-negative value is set as a reward value; if the movable intelligent body exceeds the set displacement range, a negative value is set as a penalty value, thus obtaining a displacement range constraint reward strategy; the first overlap constraint reward strategy, the second overlap constraint reward strategy, and the displacement range constraint reward strategy are used as the displacement constraint reward strategy.

[0015] According to the method provided by the present invention, the DQN model includes: an online network for predicting the expected reward value corresponding to each candidate shift action using existing fixed parameters, and outputting action prediction information based on the expected reward value; and a target network for generating a target reward value using current fixed parameters, wherein the parameters of the target network are copied from the online network at a set period.

[0016] This invention provides a deep reinforcement learning-based multi-displacement processing method for isometric annotation overlay conflicts. Through scene-adaptive target displacement methods, multi-dimensional conflict importance quantification, precise agent selection, and the synergistic application of the DQN model and displacement constraint reward strategy, it achieves efficient adaptive resolution of isometric annotation overlay conflicts. This method balances the efficiency and flexibility of conflict handling while ensuring the positional stability of high-semantic-level isometric annotations through the comprehensive conflict importance score, avoiding deviations in the transmission of core geographic information. Simultaneously, the combination of the DQN model's reinforcement learning mechanism and the displacement constraint reward strategy effectively guides annotations to avoid new overlay conflicts and maintain their relevance to their respective isometric features during displacement. This significantly improves the rationality of annotation configuration and map readability, addressing the pain points of traditional annotation displacement processing schemes, such as insufficient scene adaptability, low processing efficiency, and insufficient reliability. It enhances the efficiency, accuracy, flexibility, and reliability of isometric annotation overlay conflict handling, providing reliable technical support for the automatic configuration of high-volume, high-quality map annotations in the era of big data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the deep reinforcement learning multi-displacement processing method for surface annotation overlay conflicts provided in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram illustrating the principle of constructing annotation rectangles for planar annotations;

[0020] Figure 3 This is a schematic diagram of the shifting principle under the fixed shifting method;

[0021] Figure 4 This is a schematic diagram of the shifting principle under the free shifting mode;

[0022] Figure 5 This is a schematic diagram illustrating the principle of an adaptive solution for conflicts in all types of planar annotations;

[0023] Figure 6 This is a schematic diagram illustrating the implementation principle of the agent selection strategy;

[0024] Figure 7 This is a schematic diagram of the agent selection strategy under the fixed shift method;

[0025] Figure 8 This is a schematic diagram illustrating the principle of movement range constraint in the free displacement method;

[0026] Figure 9 This is a schematic diagram illustrating the change in the loss function of the DQN model;

[0027] Figure 10 This is a schematic diagram illustrating the change in the reward function of the DQN model;

[0028] Figure 11 This is a schematic diagram of the fitness curve of the ant colony algorithm. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] The following is combined Figures 1 to 11 This invention describes the detailed scheme of a deep reinforcement learning multi-displacement processing method for surface annotation overlay conflicts provided by embodiments of the present invention.

[0031] like Figure 1 As shown, the deep reinforcement learning multi-displacement processing method for planar annotation overlay conflicts provided in this embodiment of the invention mainly includes the following steps:

[0032] Step 110: Determine the candidate surface annotations with overlay conflicts within the target area and the complexity of each overlay conflict event, and determine the target displacement method of the candidate surface annotations under the corresponding complexity.

[0033] Understandably, the relocation method for planar annotations can be categorized based on the distribution quantity among them. Specifically, if there are many candidate planar annotations involved in a conflict, a free relocation method can be used; if there are few candidate planar annotations involved in a conflict, a fixed relocation method can be used. In other words, the specific relocation method can be determined based on the density (i.e., complexity) of the overlapping conflict distribution among planar annotations.

[0034] This embodiment focuses on the spatial overlap conflict between planar annotations corresponding to small planar features. The solution to this problem is closely related to the complexity of the actual overlap scenario. Therefore, different displacement methods can be configured according to different levels of complexity to improve the scenario adaptability in the conflict resolution process.

[0035] Step 120: Determine the number of overlays, the overlay area, and the semantic importance of each candidate facet annotation, and calculate the comprehensive conflict importance score of each candidate facet annotation.

[0036] In practical applications, the difficulty of resolving overlapping conflicts of planar annotations varies with various factors. For example, the more planar annotations involved in the conflict, the greater the coordination difficulty. The rationality of critical planar annotations needs to be prioritized during conflict resolution. Differences in different factors lead to varying degrees of importance for different conflicts. Therefore, by solving the comprehensive score of conflict importance, the influence of each factor can be quantified to predict the importance level of each type of conflict, providing a crucial reference for quickly identifying conflicts between annotations and determining the appropriate relocation method, thus making the conflict resolution process more targeted. This embodiment analyzes three factors affecting the degree of planar annotation overlapping: the number of overlapping annotations, the overlapping area, and semantic importance, and these factors are used as influencing factors in solving the comprehensive score of conflict importance.

[0037] Step 130: Based on the comprehensive score of conflict importance, determine the agent selection strategy under the corresponding target shifting mode in each overlay conflict event.

[0038] It is understandable that the core operating logic of the DQN (Deep Q-Network) model relies on the dynamic interaction between the agent and the environment. As the core carrier of decision execution and learning feedback, the agent plays a key role in perceiving the state from the environment, outputting actions, receiving rewards, and iteratively optimizing strategies. This embodiment improves the scenario adaptability of the conflict resolution process by configuring the agent selection strategy in a targeted manner under different conflict scenarios and different shifting methods, thereby improving the efficiency and accuracy of conflict handling.

[0039] Step 140: Based on the target shifting method and agent selection strategy, and combined with the pre-established DQN model and shifting constraint reward strategy, control the candidate surface annotations to perform shifting actions until the optimal position is reached.

[0040] In this embodiment, by adapting the shift constraint reward strategy to the surface annotation conflict handling scenario, and combining the DON model with the pre-determined target shift method and agent selection strategy, the effective handling of surface annotation overlay conflicts can be achieved more efficiently and accurately.

[0041] In one embodiment, determining candidate surface annotations with overlay conflicts within the target area specifically includes:

[0042] First, construct the corresponding annotation rectangle for each faceted annotation within the target area.

[0043] In this embodiment, rectangular annotation models can be used for planar features. By constructing standardized geometric rectangles as spatial proxy models for planar annotations, the original planar annotations are as follows: Figure 2 As shown in sub-figure A1, the resulting annotation rectangle is as follows: Figure 2 As shown in sub-figure B1, see [link / reference]. Figure 2 The C1 subgraph can be replaced by annotated rectangles to participate in the displacement coordination and conflict resolution process.

[0044] Then, determine whether there is spatial overlap between each annotation rectangle and other annotation rectangles, and obtain the judgment result.

[0045] Finally, the face annotations that are judged as yes are selected as candidate face annotations.

[0046] In this embodiment, during the selection process of candidate surface annotations with overlapping conflicts, it is necessary to consider the overlap between surface annotations in order to accurately screen candidate surface annotations with overlapping anomalies.

[0047] In one embodiment, determining the complexity of each capping conflict event specifically includes:

[0048] First, determine the number of candidate facet annotations involved in each capping conflict event.

[0049] In one scenario, if the number of annotations is three or less, the complexity of the overlay conflict event is determined to be simple annotation overlay.

[0050] In this case, the capping conflict event contains only three or fewer candidate facet annotation capping conflicts. The conflict relationship is clear and the scope of influence is limited. Therefore, it can be classified as simple annotation capping.

[0051] In another scenario, if the number of annotations is greater than three, the complexity of the overprinting conflict event is determined to be annotation overprinting complexity.

[0052] In this scenario, the overlap conflict involves complex overlap conflicts between more than three candidate areal annotations. The conflict relationships are intricate and the impact is significant, thus it can be categorized as an annotation overlap complexity issue. By dynamically optimizing the positional relationships of all conflicting annotations, we can ensure the effective resolution of spatial overlap conflicts between areal annotations while preserving the correspondence between areal annotations and geographic targets.

[0053] In this embodiment, the complexity can be divided by the number of annotations involved in capping, specifically into two cases: simple annotation capping and complex annotation capping.

[0054] In one embodiment, determining the target shifting method for candidate surface annotations at a corresponding complexity level specifically includes:

[0055] If the complexity of the current overlay conflict event is simple (address overlay is simple), then the target displacement method for the candidate area annotation is a fixed displacement method. Under the fixed displacement method, the movement direction of the candidate area annotation is one of eight evenly distributed directions with the center of the area feature as the origin.

[0056] In scenarios with simple annotation overlay, where only a small number of annotations overlap, a fixed displacement method can effectively resolve conflicts. The advantage of the fixed displacement method is that it can limit the movement range of area annotations to the greatest extent, thereby ensuring a high degree of correlation between area annotations and their corresponding area features.

[0057] In the fixed displacement mode, an 8-directional movement strategy is specifically adopted, such as... Figure 3 As shown in sub-figure A2, the movement range of the planar annotations is limited to eight evenly distributed directions with the center of the planar element as the origin. That is, the planar annotations can only move in the up, down, left, right, and four diagonal directions, and stop at these eight preset positions. By selectively moving the planar annotations to these eight preset positions, the overlapping conflict between planar annotations can be effectively resolved. The specific effect is as follows: Figure 3 As shown in subgraph B2.

[0058] If the complexity of the current overlay conflict event is annotation overlay complexity, then the target shifting method for the candidate surface annotation is the free shifting method. In the free shifting method, the movement direction of the candidate surface annotation is freely adjusted within the specified buffer.

[0059] In complex scenarios where multiple planar annotations overlap each other, the limited movement range allowed by fixed displacement methods makes it difficult to completely eliminate overlap conflicts. To address this, this embodiment introduces a free displacement method. Specifically, using the center point of the planar element as a reference point and adding guide lines, the movement range of the planar annotations is expanded, allowing them to explore suitable placement positions within a wider space, thus effectively solving the problem of multiple annotations overlapping in complex scenarios.

[0060] In the free displacement mode, the overlap conflict between multiple planar annotations is quite complex, for example... Figure 4 As shown in sub-figure A3, a better solution needs to be found over a wider range using free displacement. Therefore, area annotations allow for free adjustment within a specific spatial range. To clarify the relationship between area annotations and their corresponding area features, guide lines are configured for the area annotations to indicate their target. Ultimately, by optimizing the spatial position of the area annotations, the overlap conflict is resolved, as detailed below. Figure 4 As shown in sub-figure B3.

[0061] The free-shifting method grants planar annotations the ability to flexibly adjust within the permitted space, freeing them from the constraints of fixed candidate positions. It also allows agents to perform more refined searches and explorations of potential conflict-free areas. Therefore, the free-shifting method can effectively adapt to challenging scenarios such as multiple annotation overlays and complex building distributions, significantly improving the efficiency of resolving spatial overlap conflicts. Specifically, for example… Figure 4 The C3 and D3 subgraphs are shown in the diagram.

[0062] By flexibly configuring appropriate displacement methods based on the complexity of different conflict scenarios, the spatial overlap conflict of area annotations can be resolved in a case-by-case and targeted manner. For simple annotation overlap scenarios, fixed displacement is preferred; for complex overlap scenarios, free displacement is selected. Through the organic integration and synergistic application of the two displacement methods, the method's adaptability to diverse spatial distribution scenarios of annotations is significantly improved, ensuring that overlap conflicts of various area annotations in a single map can be comprehensively and efficiently resolved, thereby guaranteeing the clarity of map visualization and the accuracy of information transmission.

[0063] In one embodiment, determining the number of overlays, the overlay area, and the semantic importance of each candidate surface annotation specifically includes:

[0064] On the one hand, determine the number of times each candidate face annotation overlaps with other candidate face annotations in each overlap conflict event, and normalize the number of overlaps to obtain the number of overlaps.

[0065] The number of overlapping conflicts caused by areal annotations determines the difficulty of resolving these conflicts. Spatial overlapping conflicts between multiple areal annotations are often more difficult to resolve than those between a few. For example, cluster conflicts formed by multiple overlapping areal annotations are far more difficult to resolve than scattered conflicts between two or three areal annotations. Therefore, we use the number of overlapping candidate areal annotations as an influencing factor to identify the importance of their conflicts and construct a quantitative function to achieve objective evaluation. First, we construct a statistical function for the number of overlapping areal annotations. m Any overlay conflict event of a candidate facet annotation S , can be represented as follows:

[0066] (1)

[0067] annotation of any candidate surface The number of times it overlaps with other candidate face annotations in this conflict is counted, i.e.:

[0068] (2)

[0069] in, For the first i Candidate facet annotations Number of times the cap is pressed, For indicator functions, if the first... i Candidate facet annotations With the j Candidate facet annotations If a space cap exists, then ,on the contrary .

[0070] Then, the number of capping cycles is normalized to obtain a standardized number of capping cycles, as shown in the formula:

[0071] (3)

[0072] (4)

[0073] in, For the first i Candidate facet annotations The number of caps, The maximum original number of caps under the current conflict event, if This indicates no conflict. .

[0074] On the other hand, the sum of the first overlapping area of ​​each candidate face annotation with other candidate face annotations and the sum of the second overlapping area with face elements that are not to which it belongs are determined in each overlapping conflict event. The sum of the first overlapping area and the sum of the second overlapping area are added together and normalized to obtain the overlapping area.

[0075] When spatial overlap conflicts occur between planar annotations, the difficulty of resolving the conflict increases with the size of the overlap area. It's important to note that the overlap area must encompass two core scenarios: first, the area where conflicting annotations overlap with each other; and second, the area where a conflicting annotation overlaps with a planar feature that it doesn't belong to. The sum of these two areas directly determines the severity of the conflict in the overlap dimension. The larger the sum, the more significant the damage to annotation readability and feature relevance, and the higher the importance of the conflict. Therefore, a quantification function needs to be constructed to calculate the combined impact of these two areas. First, the sum of the first overlap areas of the candidate planar annotation and other candidate planar annotations is calculated using the following formula:

[0076] (5)

[0077] in, Indicates the first i Candidate facet annotations Sum the areas to get the total value. Indicates the first iCandidate facet annotations The sum of the first overlay areas between the annotations and other candidate areal annotations. Indicates the first i Candidate facet annotations The sum of the second overlapping areas between itself and other non-self-owned areal features.

[0078] The sum of the areas is normalized as follows:

[0079] (6)

[0080] (7)

[0081] in, For the first i Candidate facet annotations The area of ​​the cover, This represents the maximum total overlay area of ​​all candidate areal annotations in the current conflict event. If... If there is no cap, then .

[0082] On the other hand, the target semantic level corresponding to the semantic information of each candidate face annotation is determined, and the semantic importance score corresponding to the target semantic level is determined. The semantic importance score is then normalized to obtain the semantic importance.

[0083] In a specific implementation, the target semantic level corresponding to the semantic information of each candidate facet annotation is determined, specifically including:

[0084] First, determine the planning level, necessary information transmission content, industry standard information, and spatial influence level of each candidate surface annotation to obtain the core information.

[0085] Then, based on the core information, the target semantic level corresponding to the semantic information of the candidate surface annotation is determined from the pre-established core information and semantic level comparison table.

[0086] The semantic information of isometric feature annotations is a key indicator for defining the core attributes of isometric features and directly determines the information value of the annotations in spatial labeling. For example, isometric annotations with high semantic levels, such as government buildings and nature reserves, have a much higher priority in information transmission than isometric annotations with low semantic levels, such as residential areas and temporary construction zones. When isometric annotations of different semantic levels overlap and conflict, the importance of the conflict also varies with the difference in semantic value of the annotations: overlap conflicts of high semantic annotations directly affect the transmission of core spatial information and may even lead to users' misjudgment of feature attributes; therefore, their importance is much higher than that of overlap conflicts of low semantic annotations.

[0087] Therefore, when dealing with annotation overlay issues, it is necessary to first classify and judge the semantic information of conflicting annotations. In this embodiment, the semantic information of area annotations is divided into three semantic levels: core level, important level, and ordinary level according to the importance of the elements. In practical applications, the existing annotation semantic classification standards are mainly used to achieve annotation semantic classification based on four categories: planning level of area elements, necessary content for information transmission, industry standard information, and spatial influence level of elements.

[0088] Specifically, if the planning level is a national or provincial administrative unit or a core planning area, and the necessary information transmission content involves unique identifiers or irreplaceable information related to security control, and the industry standard information involves compliance with the highest priority labeling requirements of the state or industry, and the spatial impact level of the elements is wide-area radiation, then it can be classified as core level; if the planning level is a district or county administrative unit or a regional planning area, and the necessary information transmission content involves supplementary key information such as function or ownership, and the industry standard information involves compliance with the medium priority labeling requirements of the industry, and the spatial impact level of the elements is regional radiation, then it can be classified as important level; if the planning level is a community or street-level unit or ordinary land use, and the necessary information transmission content involves auxiliary information such as characteristics or temporary attributes, and the industry standard information involves compliance with the basic labeling requirements of the industry, and the spatial impact level of the elements is local radiation, then it can be classified as ordinary level.

[0089] After determining the target semantic level, the semantic importance score can be determined using the following formula:

[0090] (8)

[0091] in, For the first i Candidate facet annotations The semantic importance score.

[0092] The semantic importance scores are further normalized as follows:

[0093] (9)

[0094] in, For the first i Candidate facet annotations semantic importance The maximum original semantic score of all candidate face annotations in the current conflict event. The minimum original semantic score among all candidate facet annotations in the current conflict event. If a single-level annotation exists in the conflict event, then... .

[0095] In one embodiment, a comprehensive conflict importance score is calculated for each candidate facet annotation, specifically including:

[0096] First, determine the factor weight values ​​corresponding to the number of caps, the cap area, and the semantic importance.

[0097] Then, based on the factor weight values, the number of overlays, the overlay area, and the semantic importance of each candidate surface annotation are weighted and summed to obtain the comprehensive score of conflict importance.

[0098] To comprehensively evaluate the importance of candidate area annotation overlay conflicts, it is necessary to integrate three core influencing factors: overlay quantity, overlay area, and semantic importance. A weighted summation is used to achieve quantitative fusion of multi-dimensional information. Considering the influence of each factor on the priority of conflict resolution in practical applications, weights are assigned to the three factors: the weight of overlay quantity is set to 0.4, the weight of overlay area is set to 0.3, and the weight of semantic importance is set to 0.3. Based on this, the formula for calculating the comprehensive conflict importance score is as follows:

[0099] (10)

[0100] in, For the first i Candidate facet annotations s i The overall score of the importance of the conflict, The factor weight value is the number of caps. The factor weight value is the coverage area. The factor weights represent semantic importance.

[0101] This embodiment designs a hierarchical and progressive agent selection mechanism, combining annotation priority and spatial topology to accurately determine movable agents. To address the diversity of conflict resolution scenarios, a fixed-free dual-mode displacement framework is proposed. Strategy optimization is achieved through a DQN model, and an action decision-making system is constructed, incorporating constraints such as annotation overlap, annotation overlap with other planar elements, and annotation movement range. Ultimately, this forms a comprehensive adaptive solution for planar annotation conflicts covering all types of scenarios, from simple to complex. The specific process is as follows: Figure 5 As shown.

[0102] See Figure 5 In the process of extracting and ranking conflicting elements, the conflict of the areal annotations in the target area can be quantified first to count the number of overlapping elements. N norm Coverage area A norm and semantic importance W normThe conflict importance score of each candidate facet annotation is obtained through weighted averaging, thus prioritizing the conflict annotations. Simultaneously, a fixed shift method (i.e., ...) is matched based on the complexity of the conflict events. Figure 5 Fixed model) or free displacement method (i.e. Figure 5 (The free model in the middle) and construct the corresponding mobile intelligent agent.

[0103] In the training and experience pool interaction phase of the DQN model, the experience pool serves as the core for storing the interaction data between the agent and the environment, involving states. s t ,action a t ,award r t These parameters support dual-network training of the current network and the target network. The current network outputs the Q-value of the action, guiding the agent to select shift actions; the target network calculates the target Q-value and optimizes the parameters of the current network through a loss function; the parameters of the target network are kept stable through soft updates to avoid training fluctuations.

[0104] In the constraint reward condition judgment stage, dynamic rewards and static rewards can be used to guide the agent to move reasonably. Dynamic rewards restrict the overlap between annotations and the overlap between annotations and non-belonging surface elements, while static rewards restrict the movement range. The combination of the two types of rewards can ensure that the agent's movement not only eliminates overlap but also maintains its relevance to the surface elements it belongs to.

[0105] In the reasonable solution output stage, after the DNQ model is trained, the final configuration scheme with no overlay and stable annotation position is output according to the scheme layout and reward function changes.

[0106] In practical applications, the appropriateness of agent selection is crucial to the accuracy of map annotation locations and significantly impacts conflict resolution efficiency when dealing with overlapping situations between different candidate isometric annotations. For example, in simple scenarios where annotation overlap can be resolved using a fixed displacement method, selecting one of the candidate isometric annotations causing spatial overlap conflict as a movable agent can resolve the conflict. However, if all candidate isometric annotations are included as movable agents in the model computation, introducing too many agents will lead to unnecessary redundant computation, thus affecting the model's training convergence speed, final policy performance, and overall operating efficiency, as well as the accuracy of annotation locations.

[0107] Therefore, the conflict can be resolved simply by selecting the party involved in the conflict as a mobile agent to participate in the relocation, thereby improving computational efficiency. Similarly, for the free relocation method, it is also necessary to select appropriate candidate surface annotations that cause overlays and use them as mobile agents to participate in the relocation, in order to ensure the accuracy of map annotations and improve algorithm efficiency as much as possible. For details on the agent selection strategy in this embodiment, please refer to [link to relevant documentation]. Figure 6 .

[0108] In one embodiment, based on the comprehensive score of conflict importance, the agent selection strategy for the corresponding target shifting method in each overlay conflict event is determined, specifically including:

[0109] In the fixed displacement mode, the agent selection strategy is to prioritize one or two candidate face annotations with low overall conflict importance scores in the overlay conflict event as movable agents.

[0110] Combination Figure 6 As shown, in the agent selection strategy, for binary annotation conflict events containing only two candidate facet annotations, the lower-scoring, lower-level annotation can be prioritized as a movable agent for displacement adjustment based on the comprehensive score of conflict importance. Specifically, as follows... Figure 7 As shown in subgraph A4, if the conflict remains unresolved, the higher-scoring, higher-level annotation is used to replace the displacement, as detailed below. Figure 7 As shown in subgraph B4; for a ternary annotation conflict event containing three candidate face annotations, priority is given to collaboratively moving the two candidate face annotations with the lower scores, as detailed below. Figure 7 As shown in the C4 subgraph, if this scheme fails, the agent selection scheme is dynamically optimized based on the annotation space distribution characteristics, as detailed below. Figure 7 As shown in D4; if the above selection of binary and ternary annotations cannot resolve the conflict, then a free shift method is required for agent selection.

[0111] In the free-movement mode, the agent selection strategy is as follows: one or two candidate surface annotations with high comprehensive scores of conflict importance in the overlapping conflict event are selected as non-movable elements, and all other candidate surface annotations other than non-movable elements are selected as movable agents.

[0112] Combination Figure 6 As shown, for complex overlay scenarios consisting of three or more annotations, a free-shifting mode needs to be activated. In this case, the one or two candidate area annotations with the highest scores should be fixed as non-moving elements, while the remaining candidate area annotations are incorporated into a set of movable agents for global optimization. This hierarchical strategy, by combining annotation importance level assessment with spatial relationship analysis, achieves systematic conflict resolution from simple to complex scenarios.

[0113] In one embodiment, the DQN model specifically includes:

[0114] An online network is used to predict the expected reward value corresponding to each alternative shift action using existing fixed parameters, and output action prediction information based on the expected reward value.

[0115] The target network is used to generate a target return value using current fixed parameters. The parameters of the target network are copied from the online network at set intervals.

[0116] In this embodiment, combined with Figure 5 As shown, the DQN model employs a dual-network structure: a current network and a target network. The online network is the actual neural network used to generate action choices during training. Its main function is to calculate the Q-value of each action based on the current state, and the agent selects the corresponding action based on the magnitude of these Q-values. The formula is:

[0117] (11)

[0118] in, express The environmental state at any given time, express The action chosen at any moment Indicates online network parameters, This represents the forward propagation process of the network, where the output dimension equals the number of all possible actions, and the actions are taken from the network output vector. The corresponding dimension value is the action's... Value, that is .

[0119] The target network is a copy introduced into the DQN model, used to calculate the target Q-value. The target network's role is to maintain stability during training, and its calculation formula is as follows:

[0120] (12)

[0121] in, express The goal of the moment value, express Perform actions at all times The reward environment obtained later This is a discount factor, representing the weight of future rewards; a larger value indicates a greater emphasis on long-term rewards. express t The next state at time +1, Indicates the next state output by the target network. Next action value, This represents the maximum possible action to be taken in the next state. value, Indicates the first t +1 step available action This represents the parameters of the target network.

[0122] The DQN model internally incorporates an experience replay mechanism. Each experience (state, action, reward, and next state) acquired by the agent during interaction with the environment is stored in an experience pool. When the pool is full, newer experiences overwrite the oldest, ensuring the replay pool contains the most up-to-date interaction data. During training, the model continuously draws a batch of experiences randomly from the pool to calculate the target Q-value. Then, it calculates the difference between the current Q-value and the target Q-value (the loss function). Finally, it minimizes the loss function using gradient descent, thereby updating the network parameters. The specific formula is:

[0123] (13)

[0124] in, The value of the loss function. This serves as an experience pool, storing the agent's action experience and participating in random sampling during experience replay to reduce sample correlation and improve training stability. This means calculating the expected value of the state-action-reward-next state sample obtained from the experience pool, that is, using the average loss of the batch samples to represent the overall expected loss.

[0125] In one embodiment, a shift-constrained reward strategy is established, specifically including:

[0126] During the execution of the displacement action, if there is no overlap conflict between the movable intelligent agent and other surface annotations, a positive value will be set as the reward value; otherwise, the negative value of the absolute value of the overlap area between the movable intelligent agent and other surface annotations will be multiplied by a first set multiple as the penalty value, thus obtaining the first overlap constraint reward strategy.

[0127] Understandably, although the two shifting methods differ, both aim to ensure that the agent does not overlap with other annotations during the shift. This is a core objective for resolving conflicts. Therefore, it is necessary to construct annotation overlap constraints for both shifting methods. By constructing these constraints, annotations acting as agents can be guided to avoid overlapping with other annotations during the shift. If overlapping occurs, the overlapping area between the annotation and other annotations is calculated. and the negative value of the absolute value of the covering area. As a basis for punishment, and multiplied by The amplification coefficient increases the significance of the penalty. If no cover is generated, a reward of 1 is assigned to indicate that no penalty is imposed on the action. The reward value in the first cover constraint reward policy is... The specific calculation formula is as follows:

[0128] (14)

[0129] If there is no overlap conflict between the movable agent and the non-dominant surface feature, a positive value will be set as the reward value; otherwise, the negative value of the absolute value of the overlap area between the movable agent and the non-dominant surface feature will be multiplied by a second set factor as the penalty value, thus obtaining the second overlap constraint reward strategy.

[0130] When candidate area annotations, acting as mobile intelligent agents, participate in movement, this embodiment considers the overlap between the candidate area annotation and its own area feature as a reasonable overlap. This ensures that the final position of the annotation does not overlap with other annotations, and also prevents the annotation from overlapping with other area features not belonging to it. If a candidate area annotation overlaps with a non-self area feature during its movement, the size of the overlap area is calculated, and the absolute value of this overlap area is determined. The negative value is used as the basis for the penalty, and multiplied by The amplification coefficient increases the significance of the penalty. Compared to the annotation-overlay constraint, its amplification coefficient is an order of magnitude smaller, distinguishing the levels of constraints between two different constraints. This indicates that resolving the annotation-overlay constraint is the primary focus when resolving spatial overlay conflicts between annotations. If no overlay occurs, a reward value of 1 is assigned, indicating no penalty constraint on its action. Therefore, the reward value in the second overlay constraint reward strategy is... The specific calculation formula is as follows:

[0131] (15)

[0132] If the target displacement method is free displacement, if the movable agent does not exceed the set displacement range, a non-negative value will be set as the reward value; if the movable agent exceeds the set displacement range, a negative value will be set as the penalty value, resulting in a displacement range-constrained reward policy.

[0133] The core difference between free-movement and fixed-movement methods lies in their approaches. Fixed-movement methods iterate through preset discrete positions to find a suitable target location, while free-movement methods achieve this through continuous movement. While this continuous movement enhances exploration flexibility, it also carries the risk of excessive deviation. Without constraints on its movement range, the agent might explore areas unrelated to the isometric features to which the annotation belongs, significantly weakening the correlation between the annotation and the feature, thus violating the core principles of annotation.

[0134] Therefore, in the free-movement method, it is necessary to construct clear movement range constraints. By defining reasonable exploration boundaries, the advantages of continuous movement in complex conflict scenarios can be retained, while ensuring that the annotation always explores within a space strongly related to the elements. Ultimately, a balance is found between eliminating overlay and maintaining relevance. In this embodiment, the movement range is defined by extending a certain range outward from the conflict annotation itself, such as... Figure 8As shown in subgraph A5, after processing by the DQN model, conflicting annotations find suitable solutions within their respective movement ranges, as follows: Figure 8 As shown in subgraphs B5 and C5, the conflict between annotations is ultimately resolved, as follows: Figure 8 As shown in the D5 sub-figure.

[0135] By constructing a conflict annotation movement range constraint for the free-shifting method, if a candidate face annotation shifted using the free-shifting method exceeds the constraint range, a penalty value of -10 is assigned; conversely, a value of 0 is assigned as a reward value. The reward value in the movement range constraint reward strategy... The specific formula is:

[0136] (16)

[0137] The first capping constraint reward strategy, the second capping constraint reward strategy, and the movement range constraint reward strategy are used as the displacement constraint reward strategy.

[0138] After the agent performs an action, the aforementioned constraint reward is evaluated. By summing the reward values ​​obtained from evaluating each constraint condition, the total reward value can be obtained. The specific formula is as follows:

[0139] (17)

[0140] By calculating its total reward value The system checks if the value is greater than or equal to 0 to determine if all the above constraints are met. If the value is 0, the current round of movement ends and training continues in the next round. If the value is less than 0, the system continues to move until the constraints are met or the agent's movement limit is reached, at which point the training for that round ends.

[0141] To verify the effectiveness of the method provided in this embodiment, a study area containing 201 areal features was selected to verify the effectiveness of the method. The study area contains 201 areal annotations. The comprehensive score of conflict importance was calculated for the annotation data of the study area to extract the candidate areal annotations involved in the overlay conflict and the ranking of the importance among the candidate areal annotations. After analysis, a total of 53 candidate areal annotations with spatial overlay conflict were found in the study area. After sorting according to the comprehensive score of conflict importance, the specific information of the top 10 candidate areal annotations can be found in Table 1.

[0142] Table 1. Detailed information on some candidate areal annotations in the study area.

[0143]

[0144] After calculating the overall score of conflict importance, a series of refined processing procedures were performed on the annotations within the study area. First, the annotation rectangle corresponding to each planar annotation was extracted, and these highlighted annotation rectangles were used as annotation replacements. Then, the displacement method was determined based on the difference in the number of overlays, and movable agents were selected according to the agent selection strategy.

[0145] When handling annotation space overlay conflicts, the DQN model parameters are set as follows: the experience pool is set to 10000 (sufficiently storing experience to avoid overfitting), and the training batch size is 64; the initial exploration rate is 0.8, the decay rate is 0.995 (sufficient exploration in the early stages, favoring the optimal strategy in the later stages), the total training batches are 600, and the target network is updated every 10 rounds (balancing learning effect and computational efficiency); the learning rate is 0.001 (ensuring learning efficiency), and the discount factor is 0.99 (emphasizing long-term rewards); the action step size for each agent is 0.00001, and the action limits for fixed shifts and free shifts are set to 15 and 20 respectively (the process ends when the limit or constraint condition is met, and all agents are iterated sequentially). For overlay conflict scenarios, each agent iterates for 600 rounds.

[0146] In simple overlay conflict scenarios, due to the relatively simple conflict relationship and the relatively simple interaction logic between annotations, the model took a total of 2508.152 seconds after a series of calculations and adjustments. However, in complex overlay conflict scenarios, due to the existence of multi-level overlay relationships and the intricate mutual influence between annotations, the model needs to consume more resources for conflict analysis and position optimization, and the total time reached 5849.29 seconds.

[0147] Further calculations show that the average time cost for a single agent to complete a single round of calculation in a simple overlay conflict scenario is 0.21 seconds. This means that the agent can quickly adjust its position to resolve the conflict in this scenario. However, in a complex overlay conflict scenario, the average time cost for a single agent to complete a single round of calculation is 0.98 seconds, reflecting that the process of position optimization by the agent is more time-consuming in complex scenarios. From the overall time consumption perspective, the model still maintains high processing efficiency for handling overlay problems between annotations. Table 2 shows the statistical data of the processing results.

[0148] Table 2 Statistical data of processing results

[0149]

[0150] As shown in Table 2, the agents using fixed shift and free shift methods achieved success rates of 100% and 78.6% respectively, demonstrating highly efficient processing capabilities. The change in the loss function of the DQN model during conflict resolution is as follows: Figure 9 As shown.

[0151] The change in the loss function under a simple capping environment shows that... Figure 9 The loss function curves of the first twenty agents show a high degree of similarity during the training of the DQN model, all exhibiting a significant decrease and a tendency to plateau. This indicates that the model performance is stabilizing. This phenomenon verifies the effectiveness of the DQN algorithm in adaptive parameter adjustment; that is, by continuously optimizing the neural network weights, the model can gradually approach the optimal solution. Overall, the changes in its loss function demonstrate that the DQN model exhibits strong robustness in spatial occlusion conflict resolution tasks, achieving efficient conflict resolution even when network parameters are not completely stable through dynamic adjustment.

[0152] Analyzing the changes in the loss function under complex capping environments, and observing... Figure 9 The loss function curves for the last ten agents reveal that, due to the complex overlapping conflicts between multiple annotations and the intricate distribution of areal elements in the complex overlapping environment, conflict resolution is far more difficult than in the simple overlapping environment, which only involves overlapping conflicts between annotations. Specifically, the fluctuation range of the loss function in the complex overlapping environment is significantly greater than that in the simple overlapping environment, and its downward trend is also much weaker. This phenomenon indirectly reflects that the DQN model needs to continuously adjust its parameters and strategies to find the optimal solution for resolving spatial overlapping conflicts between annotations when dealing with such complex conflicts.

[0153] On the other hand, the reward function of the DQN model is as follows: Figure 10 As shown, the change in the reward function under a simple overlay environment reveals that... Figure 10 The reward values ​​of the first twenty agents all showed a continuous upward trend, intuitively reflecting the gradual adaptation and performance optimization of the agents to the environment during training. Furthermore, guided by their respective reward functions, different agents successfully eliminated spatial overlay conflicts between annotations, indicating that the current training state of the agents meets the core requirements of the task. In a complex overlay environment, a deep analysis of the dynamic evolution of the last ten reward functions reveals that the overall reward functions of the agents all show a certain upward trend. Further combined with... Figure 10 The intuitive data and trends presented are comprehensively analyzed, and the current trajectory of the reward function clearly shows that the system has successfully reached the optimal conflict solution, which strongly verifies the effectiveness and reliability of the DQN model with the free shift mechanism in solving the complex spatial overlay conflict problem between annotations.

[0154] The following comparative experiments further verify the advantages of the method provided in this embodiment. In the first set of comparative experiments, on the one hand, the traditional ant colony algorithm is used to solve the spatial overlay conflict between annotations; on the other hand, the optimization effect of different single displacement methods is tested simultaneously. Through multi-dimensional comparison, the advantages and disadvantages of each method are clarified. The core design of the ant colony algorithm is as follows:

[0155] Pheromon concentration indicates the quality of an agent's actions; the heuristic function evaluates positional suitability based on the coverage area. Action selection combines greedy selection and roulette wheel selection to balance convergence efficiency and search breadth; pheromone updates dynamically guide the search through the evaporation of old pheromones and the release of new pheromones, approximating the global optimum. Considering the characteristics of annotation conflict scenarios, key parameters are set as follows: pheromone evaporation coefficient 0.15, pheromone importance factor 1.2, heuristic function importance factor 1.7, greedy selection probability 0.7, and pheromone increment constant 150.

[0156] For the ant colony algorithm utilizing both fixed and free movement methods, 600 training rounds are set, with the maximum number of actions per round consistent with the method provided in this embodiment, set to 15. The global fitness is calculated by summing the area covered by each ant with other annotations and structures after each training round, using the following formula:

[0157] (18)

[0158] in, For fitness value, This refers to the total coverage area noted.

[0159] In this comparative experiment, conflicting annotation elements were analogized to ants in an ant colony algorithm, and the goal of moving annotations to a location where they do not overlap with any other elements was analogized to a food source in the ant colony algorithm. Through this mapping, a logical transformation was achieved from ants optimizing their paths to find food to conflicting annotations continuously moving to find conflict-free locations, ultimately achieving the goal of resolving spatial overlap conflicts between annotations. The statistical data of the processing results of the ant colony algorithm can be seen in Table 3.

[0160] Table 3 Statistical data of ant colony algorithm processing results

[0161]

[0162] As shown in Table 3, the success rates for handling overlapping conflicts between annotations were 25.6% and 0%, respectively. This is significantly lower than the 94.3% success rate of the method proposed in this embodiment. This stark contrast powerfully demonstrates the superiority of the method provided in this embodiment over the ant colony algorithm in handling such problems.

[0163] The fitness curve of the ant colony algorithm is as follows: Figure 11 As shown, based on the analysis of the dynamic changes in the curve, it exhibits extremely significant oscillation characteristics. Specifically, during the algorithm's operation, the fitness value fluctuates frequently within a large range, lacking a stable downward or upward trend. In stark contrast, the reward function used in the method provided in this embodiment exhibits a highly convergent state. Under the same number of iterations or computational load, the reward function value can quickly and stably approach a relatively fixed optimal value. This stark difference indicates that, under the same computational load setting, the ant colony algorithm struggles to achieve efficient convergence. Further observation of the data in the curve reveals that when different shift models are used, the fitness curve of the ant colony algorithm shows similar trends. Specifically, regardless of how the shift method is adjusted, the fitness curve never shows a clear convergence trend, but rather fluctuates continuously within a certain range. This phenomenon confirms that when dealing with this type of map annotation overlay problem, even if the key parameter of the shift method is changed, the ant colony algorithm struggles to overcome its learning bottleneck and cannot effectively optimize the search process to find an ideal solution.

[0164] The results of the above comparative experiments fully verify that the method provided in this embodiment has significant advantages in terms of conflict handling efficiency, algorithm robustness, and engineering applicability.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep reinforcement learning multi-displacement processing method for facing the conflict of surface inscription grommet, characterized in that, The application comprises the following steps: determining candidate planar annotations in the target area that exist in the cap conflict, determining the number of annotations of the candidate planar annotations involved in the cap conflict in each cap conflict event; if the number of annotations is less than three, determining that the complexity of the cap conflict event is simple annotation cap; if the number of annotations is greater than three, determining that the complexity of the cap conflict event is complex annotation cap; if the complexity of the current cap conflict event is simple annotation cap, the target displacement mode of the candidate planar annotation is a fixed displacement mode, wherein in the fixed displacement mode, the moving direction of the candidate planar annotation is eight evenly distributed directions with the center of the planar element as the origin; if the complexity of the current cap conflict event is complex annotation cap, the target displacement mode of the candidate planar annotation is a free displacement mode, wherein in the free displacement mode, the moving direction of the candidate planar annotation is freely adjusted in the specified buffer area; determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, and calculating the comprehensive conflict importance score of each candidate planar annotation; determining the agent selection strategy in the corresponding target displacement mode in each cap conflict event according to the comprehensive conflict importance score, including: in the fixed displacement mode, the agent selection strategy is: preferentially selecting one or two candidate planar annotations with low comprehensive conflict importance score in the cap conflict event as movable agents; in the free displacement mode, the agent selection strategy is: selecting one or two candidate planar annotations with high comprehensive conflict importance score in the cap conflict event as non-moving elements, and selecting all other candidate planar annotations except the non-moving elements as movable agents; controlling the candidate planar annotations to perform displacement actions until the optimal position according to the target displacement mode and the agent selection strategy, and combining the pre-established DQN model and the displacement constraint reward strategy.

2. The deep reinforcement learning multi-displacement processing method for face-shaped annotation press cover conflict according to claim 1, wherein, The method comprises the following steps: constructing a corresponding annotation rectangle for each planar annotation in the target area; judging whether there is spatial overlap between each annotation rectangle and other annotation rectangles to obtain a judgment result; regarding the planar annotations with the judgment result as candidate planar annotations.

3. The deep reinforcement learning multi-displacement processing method for face-shaped inscription press cover conflict according to claim 1, characterized in that, The method comprises the following steps: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and the semantic importance of each candidate planar annotation, including: determining the number of caps, the cap area and 4. The deep reinforcement learning multi-displacement processing method for face-shaped inscription press cover conflict according to claim 3, characterized in that, The semantic information corresponding to the target semantic level of each candidate planar annotation is determined, including: The semantic information corresponding to the target semantic level of each candidate planar annotation is determined, including: The core information is obtained by determining the planning level, information transmission necessary content, industry standard information and element space influence level of the semantic information of each candidate planar annotation.

5. The deep Q-network method for handling conflicts between text labels and depth according to claim 1, wherein, According to the core information, the target semantic level corresponding to the semantic information of the candidate planar annotation is determined from the pre-established core information and semantic level correspondence table. The conflict importance comprehensive score of each candidate planar annotation is calculated, including: The factor weight values corresponding to the number of caps, cap area and semantic importance are determined.

6. The deep Q-network method for handling conflicts between text labels and depth according to claim 1, wherein, According to the factor weight values, the number of caps, cap area and semantic importance of each candidate planar annotation are weighted and summed to obtain the conflict importance comprehensive score. A displacement constraint reward strategy is established, including: During the execution of the displacement action, if there is no cap conflict between the movable agent and other planar annotations, a positive value is set as the reward value, otherwise, the negative value of the cap area absolute value between the movable agent and other planar annotations is enlarged by a first set multiple to obtain a first cap constraint reward strategy. If there is no cap conflict between the movable agent and the planar element not belonging to itself, a positive value is set as the reward value, otherwise, the negative value of the cap area absolute value between the movable agent and the planar element not belonging to itself is enlarged by a second set multiple to obtain a second cap constraint reward strategy. If the target displacement mode is a free displacement mode, if the movable agent does not exceed the set movement range, a non-negative value is set as the reward value, if the movable agent exceeds the set movement range, a negative value is set as the punishment value, to obtain a movement range constraint reward strategy.

7. The deep Q-network method for handling conflicts between text labels and depth according to claim 1, wherein, The first cap constraint reward strategy, the second cap constraint reward strategy and the movement range constraint reward strategy are used as the displacement constraint reward strategy. The DQN model includes: An online network for predicting the expected return value corresponding to each candidate displacement action through original fixed parameters, and outputting action prediction information according to the expected return value; A target network for generating a target return value through current fixed parameters, the parameters of the target network being copied from the online network at a set period.

Images