A method, device, equipment and storage medium for generating landscape plan

Abstract

The present invention relates to the field of landscape plan design technology, and specifically to a landscape plan generation method, device, equipment and storage medium. The present invention first initializes the design condition diagram according to the layout data of the area to be designed to obtain an initial landscape plan, then evaluates the initial landscape plan to obtain the design quality score of the initial landscape plan, and finally continuously adjusts the initial landscape plan according to the design quality score until a target landscape plan is obtained. From the above analysis, it can be seen that the present invention adjusts the landscape plan according to the design quality score, that is, guides the adjustment direction of the landscape plan through the design quality score, so that the quality of the target landscape plan obtained after the adjustment is higher to meet the actual use needs.

Description

Technical Field

[0001] The present invention relates to the technical field of landscape plan design, and in particular to a landscape plan generation method, device, equipment and storage medium. Background Art

[0002] Before the actual greening of parks, residential areas, streets, and industrial parks is carried out, a landscape plan will be designed first. The landscape plan will indicate the type of greening to be planted in the area and the recreational places to be built, and then the area will be constructed according to the landscape plan. Taking the residential area as an example, the existing technology inputs the basic information of the site boundary, entrance and exit locations, and internal building locations of the residential area into the adversarial network, and the adversarial network outputs the landscape plan of the residential area. However, since the adversarial network generates the landscape plan in a one-time generation method, even if the generated landscape plan cannot meet the actual needs well, it cannot improve the overall quality of the landscape plan by improving the local information of the landscape plan.

[0003] In summary, the landscape plan generated by the existing technology is difficult to meet the actual use requirements due to its low quality.

[0004] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method, device, equipment and storage medium for generating a landscape plan, which solves the problem that the landscape plan generated by the prior art is of low quality and difficult to meet actual usage requirements.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for generating a landscape plan, comprising:

[0008] Acquire layout data of the area to be designed, initialize the positions of design elements according to the layout data, and obtain an initial landscape plan;

[0009] determining a design quality score for the initial landscape plan;

[0010] According to the design quality score of the initial landscape plan, the initial landscape plan is adjusted through the trained deep reinforcement learning model to obtain the target landscape plan of the area to be designed.

[0011] In one implementation, initializing the positions of the design elements according to the layout data to obtain an initial landscape plan includes:

[0012] Identify pathways and scenic features within said design elements;

[0013] Determining the site boundaries and entrance and exit locations and internal building locations in the layout data;

[0014] Initializing the starting position and the ending position of the road according to the entrance and exit positions;

[0015] Randomly initializing the relative position of the viewing landscape between the interior of the site boundary and the interior building position;

[0016] An initial landscape plan is generated based on the site boundary, the entrance and exit locations, the internal building locations, the roads after the initialization location, and the viewing landscape.

[0017] In one implementation, determining the design quality score of the initial landscape plan includes:

[0018] Applying the trained quality scoring model to the initial landscape plan to obtain a first score for the initial landscape plan, wherein the first score is used to characterize the road traffic information and the viewing landscape information presented by the initial landscape plan;

[0019] Determine a second score of the initial landscape plan, the second score being used to characterize the road network data, green space data and open space area of ​​the initial landscape plan;

[0020] A design quality score of the initial landscape plan is obtained according to the first score and the second score.

[0021] In one implementation, adjusting the initial landscape plan according to the design quality score of the initial landscape plan through a trained deep reinforcement learning model to obtain a target landscape plan includes:

[0022] Applying the trained deep reinforcement learning model to the design quality score of the initial landscape plan and the initial landscape plan, so as to iteratively update the initial landscape plan through the trained deep reinforcement learning model to obtain an updated landscape plan;

[0023] Collecting the number of iterations corresponding to the landscape plan update map, and determining an incentive score according to the number of iterations and the design quality score of the landscape plan update map;

[0024] The landscape plan update map is continuously updated by using the incentive score and the trained deep reinforcement learning model until the updated landscape plan update map meets the set conditions, thereby obtaining a target landscape plan map.

[0025] In one implementation, the step of continuing to update the landscape plane update map by using the incentive score and the trained deep reinforcement learning model includes:

[0026] Obtaining the design area occupied by the design elements on the landscape plan update map;

[0027] Applying a trained deep reinforcement learning model to the incentive score and the landscape plane update map, so as to iteratively update the design area through the trained deep reinforcement learning model to obtain a design expansion area;

[0028] Random sampling is performed in the designed extension area through the trained deep reinforcement learning model to obtain a random sampling result, and the landscape plane update map is iteratively updated according to the relative position between the random sampling result and the designed area.

[0029] In one implementation, iteratively updating the landscape plane update map according to the relative position between the random sampling result and the design area includes:

[0030] Determine a number of design grids included in the design area, and determine a random sampling grid in the random sampling result;

[0031] When the random sampling grid is located in a plurality of the design grids, updating the landscape plane update map by deleting the random sampling grid;

[0032] When the random sampling grid is located outside the plurality of design grids, the landscape plane update map is updated by adding the random sampling grid.

[0033] In one implementation, the training method of the trained quality scoring model and the trained deep reinforcement learning model includes:

[0034] Obtaining a landscape plan sample image and a benchmark score corresponding to the landscape plan sample image;

[0035] Applying a quality scoring model to the landscape plan sample image to obtain a training score output by the quality scoring model;

[0036] Determining a difference between the training score and the benchmark score, and adjusting parameters of the quality scoring model according to the difference to obtain a trained quality scoring model;

[0037] Based on the trained quality scoring model, a deep reinforcement learning model is trained to obtain a trained deep reinforcement learning model.

[0038] In a second aspect, an embodiment of the present invention further provides a landscape plan generating device, wherein the device comprises the following components:

[0039] An initial landscape plan generation module is used to obtain layout data of the area to be designed, initialize the positions of design elements according to the layout data, and obtain an initial landscape plan;

[0040] A scoring module, used to determine a design quality score of the initial landscape plan;

[0041] An updating module is used to adjust the initial landscape plan according to the design quality score of the initial landscape plan through a trained deep reinforcement learning model to obtain a target landscape plan of the area to be designed.

[0042] In a third aspect, an embodiment of the present invention further provides a terminal device, wherein the terminal device includes a memory, a processor, and a landscape plan generation program stored in the memory and executable on the processor, and when the processor executes the landscape plan generation program, the steps of the landscape plan generation method described above are implemented.

[0043] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a landscape plan generation program is stored. When the landscape plan generation program is executed by a processor, the steps of the landscape plan generation method described above are implemented.

[0044] Beneficial effect: The present invention first initializes the design condition diagram according to the layout data of the area to be designed to obtain an initial landscape plan, then evaluates the initial landscape plan to obtain the design quality score of the initial landscape plan, and finally continuously adjusts the initial landscape plan according to the design quality score until the target landscape plan is obtained. From the above analysis, it can be seen that the present invention adjusts the landscape plan according to the design quality score, that is, guides the adjustment direction of the landscape plan through the design quality score, so that the quality of the target landscape plan obtained after the adjustment is higher to meet the actual use needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the overall flow chart of the present invention;

[0046] Figure 2 is a training flow chart of a deep reinforcement learning model in an embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of layout data in an embodiment of the present invention;

[0048] Figure 4 is a schematic diagram of the positional relationship between the design elements and layout data in an embodiment of the present invention;

[0049] Figure 5A schematic diagram of a design area at time t in an embodiment of the present invention;

[0050] Figure 6 For Figure 5 The schematic diagram of the design expansion area at time t is formed on the basis of

[0051] Figure 7 For Figure 6 Schematic diagram of random sampling based on ;

[0052] Figure 8 is a schematic diagram of a landscape plan at time t+1 in an embodiment of the present invention;

[0053] Fig. 9 A structural diagram of a landscape plan generating device provided by the present invention;

[0054] Fig.10 This is a block diagram of the internal structure principle of the terminal device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following is a clear and complete description of the technical solution of the present invention in combination with the embodiments and the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] According to research, before the actual greening planting is carried out in parks, residential areas, streets and industrial parks, a landscape plan is designed first, which indicates the type of greening to be planted in the area and the leisure places to be built, and then the area is built according to the landscape plan. Taking the residential area as an example, the existing technology inputs the basic information of the site boundary, entrance and exit locations and internal building locations of the residential area into the adversarial network, and the adversarial network outputs the landscape plan of the residential area. However, since the adversarial network generates the landscape plan in a one-time generation method, even if the generated landscape plan cannot meet the actual needs well, it cannot improve the overall quality of the landscape plan by improving the local information of the landscape plan.

[0057] In order to solve the above technical problems, the present invention provides a method, device, equipment and storage medium for generating a landscape plan, which solves the problem that the landscape plan generated by the prior art is difficult to meet the actual use requirements due to its low quality. In specific implementation, the layout data of the area to be designed (the layout data is also the site design conditions) is first obtained, and the trained deep reinforcement learning model is applied to the layout data to obtain the initial landscape plan of the area to be designed; then the design quality score of the initial landscape plan is determined; finally, according to the design quality score, the initial landscape plan is adjusted to obtain the target landscape plan.

[0058] For example, if the area to be designed is a residential area (a residential area is also a community), the layout data includes the entrance and exit locations of the residential area, the site boundary (that is, the site red line of the residential area), and the location of each building within the site boundary. First, based on the above three layout data, the initial landscape plan is obtained by initializing the position of the design elements. Design elements may include roads and ornamental landscapes, which may include trees, grass and shrubs, structures (such as pavilions) and water bodies. Initialization is to initialize the starting position of the road inside the residential area to the entrance of the residential area and the entrance of the unit building, and randomly initialize the starting point of the ornamental landscape.

[0059] Evaluate initial landscaping plans The design quality score is used to indicate the extent to which the internal landscape of the residential area built according to the landscape plan can meet the use requirements, and then the design quality score and the initial landscape plan are compared. Input to the deep reinforcement learning model DQN, which adjusts the initial landscape plan according to the design quality score , and obtain the target landscape plan that can meet the usage requirements.

[0060] The landscape plan generation method of this embodiment can be applied to a terminal device, which can be a terminal product with an image generation function, such as a computer. Figure 1 As shown in , the landscape plan generation method specifically includes the following steps:

[0061] S100, obtaining layout data of the area to be designed, initializing the positions of design elements according to the layout data, and obtaining an initial landscape plan;

[0062] S200, determining a design quality score of the initial landscape plan;

[0063] S300: According to the design quality score of the initial landscape plan, the initial landscape plan is adjusted through a trained deep reinforcement learning model to obtain a target landscape plan of the area to be designed.

[0064] The design quality score of step S200 includes a first score and a second score, wherein the first score is used to characterize the degree of traffic convenience and the viewing landscape information (that is, the attribute information of the viewing landscape), and the second score is used to characterize the road network data, green space data, and the area of ​​the open space area (the open space area is the area with neither buildings nor greenery). The first score can be obtained by applying the trained quality scoring model to the initial landscape plan.

[0065] Embodiment 1: In this embodiment, a deep reinforcement learning model and a quality scoring model are trained, wherein the training of the quality scoring model is completed first, and then the deep reinforcement learning model is trained based on the trained quality scoring model.

[0066] Among them, the quality scoring model is the residual neural network ResNet. Training the quality scoring model requires building a training data set, and using the training data set to train the quality scoring model. Building a training data set includes: using a search engine to crawl data with the keywords "residential area plan", "community plan", and "residential community plan", cleaning the crawled data, removing duplicate and incomplete landscape plans, and obtaining the original data set. The image segmentation tool SegNet is used for each landscape plan of the original data set to obtain the site boundaries, entrance and exit locations, building locations, roads, trees, grass and shrubs, structures (structures are leisure landscapes, such as pavilions), and water bodies contained in each landscape plan. The site boundaries, entrance and exit locations, building locations, roads, trees, grass and shrubs, structures, and water bodies are encoded with data to obtain the feature data of each picture. And based on the road layout (that is, traffic flow lines), landscape sight, and landscape design diversity, a quality score label (the quality score label is also the benchmark score) is set for each landscape plan. Among them, the landscape sight is the degree of obstruction of the sight by trees, grass and shrubs, structures, and water bodies. The diversity of landscape design includes the diversity of trees, the diversity of grasses and shrubs, the diversity of structures, and the diversity of water bodies.

[0067] The feature data of each landscape plan view is input into the residual neural network ResNet to obtain the training score output by the residual neural network ResNet, the difference between the training score and the benchmark score is calculated, and the parameters of the residual neural network ResNet are adjusted according to the difference to obtain the trained residual neural network ResNet.

[0068] After completing the training of the residual neural network ResNet, start training the deep reinforcement learning model DQN. Training DQN includes the following steps:

[0069] like Figure 2 As shown, the sample layout data of the residential area includes the sample site boundary, the sample entrance and exit locations, and the sample internal building locations. The landscape plan of the residential area is initialized to obtain the initialized sample landscape plan (that is, Figure 2 The initial environment in the example), the deep reinforcement learning model DQN is applied to the initialized sample landscape plan, and DQN adjusts its actions by observing the sample landscape plan, and DQN adjusts the sample landscape plan through the action to obtain the updated map of the sample map (that is, Figure 2The purpose of training is to continuously adjust the Q value distribution of DQN according to the update graph and reward until the cumulative reward generated by each iterative training reaches 1000, then stop training.

[0070] Embodiment 2, based on embodiment 1, in this embodiment, step S100 includes the following specific steps S101 to S105:

[0071] S101, determining roads and scenic spots in the design elements.

[0072] The ornamental landscape includes trees, grasses and shrubs, structures and water bodies.

[0073] S102: Determine the site boundary and entrance and exit locations and internal building locations in the layout data.

[0074] Site boundaries, entrance and exit locations, and internal building locations Figure 3 As shown, if the area to be designed is a residential area, then the site boundary is the boundary line of the residential area, the entrance and exit locations are the entrance and exit locations of the residential area and internal unit buildings, and the internal building locations are the locations of each building inside the residential area. Figure 3 The triangle inside the site boundary is the entrance and exit of each building. Figure 3 The triangle outside the site boundary represents the entrance and exit of the residential area.

[0075] S103: Initialize the starting position and the ending position of the road according to the entrance and exit positions.

[0076] S104: Randomly initialize the relative position of the viewing landscape between the interior of the site boundary and the internal building position.

[0077] S105, generating an initial landscape plan according to the site boundary, the entrance and exit locations, the internal building locations, the roads after the initialization location, and the viewing landscape .

[0078] That is, first generate an original plan including several grids, then mark the site boundaries, entrance and exit locations, and internal building locations on the original plan, and mark the locations of various design elements on the original plan to convert the original plan into an initial landscape plan. , where the initial landscape plan The dimensions are 1000x1000. Figure 4 This is one of the initial landscape plans. .

[0079] Embodiment 3, this embodiment is based on embodiment 1 or embodiment 2. In this embodiment, step S200 includes the following specific steps S201, S202, and S203:

[0080] S201, applying the trained quality scoring model to the initial landscape plan to obtain a first score of the initial landscape plan The first score is used to characterize the road traffic information and the viewing landscape information presented by the initial landscape plan.

[0081] That is, the quality scoring model obtains the first score based on the quality of the initial landscape plan. .

[0082] The quality scoring model can also evaluate the difference between the roads on the initial landscape plan and the standard roads, the difference between the landscape height and the standard height, and the number of landscapes to obtain a first score , where the landscape height refers to the height of the trees, grasses and shrubs, structures, and water bodies (the water bodies are the water play facilities in the community) on the initial landscape plan, and the landscape quantity refers to the total number or total number of types of trees, grasses and shrubs, structures, and water bodies.

[0083] If the quality score model outputs a result greater than or equal to 8, then let is equal to 20; if the result output by the quality scoring model is less than 8, then let Equals -10.

[0084] S202, determining a second score of the initial landscape plan, where the second score is used to characterize the road network data, green space data and open space area of ​​the initial landscape plan.

[0085] The second fraction is equal to add add ,in It is the score corresponding to the road network density of the roads on the initial landscape plan, when the road network density is less than 80m / hm²; is -10, when the road network density is greater than or equal to 80m / hm², is -20. is the score corresponding to the green coverage rate on the initial landscape plan, that is, when the green coverage rate is less than 35%, = -10; when the green coverage rate is greater than or equal to 35%, Equals 20. is the area of ​​the open space, that is, the blank area. When the area of ​​the open space on the initial landscape plan is zero, is equal to 1000, otherwise, is equal to 0.

[0086] S203: Obtaining a design quality score of the initial landscape plan according to the first score and the second score .

[0087] ;

[0088] In the formula, , , are constants, and , where M represents the quality scoring model, L represents the road network, and G represents the green space.

[0089] Embodiment 4, based on embodiment 1 or embodiment 2 or embodiment 3, in this embodiment, step S300 includes the following specific steps S301 to S307:

[0090] S301, applying the trained deep reinforcement learning model to the design quality score of the initial landscape plan and the initial landscape plan, so as to iteratively update the initial landscape plan through the trained deep reinforcement learning model to obtain an updated landscape plan. .

[0091] For the The landscape plan obtained by iterative update is .

[0092] S302, collecting the number of iterations corresponding to the landscape plan update map, and calculating the design quality score of the landscape plan update map according to the number of iterations and the design quality score of the landscape plan update map. , determine the incentive score .

[0093] ;

[0094] ;

[0095] In the formula, is the basic score, once per iteration, Increase by 1, that is Equal to landscape level

[0096] The number of iterations corresponding to the face update graph, B represents the basis term. , , are constants, and . For the The first score of the landscape plane update map generated by the iteration update, Update the landscape plan (this is the The score corresponding to the road network density on the generated iterative update) is is the score corresponding to the green coverage rate on the landscape plan update map, Update the fraction of open space areas on the landscape plan.

[0097] S303: Obtaining the landscape plan update map The design area occupied by the design elements above.

[0098] Will Figure 5 The grid where the mark "1" is located is defined as the first row and the first column. Figure 5 As shown, the landscape plan update The design area occupied by the design elements includes a first design area, a second design area and a third design area.

[0099] The first design area includes the grid in the 5th row and 2nd column, the grid in the 5th row and 3rd column, and the grid in the 6th row and 2nd column.

[0100] The second design area includes three grids of the 6th, 7th and 8th columns of the 8th row, three grids of the 6th, 7th and 8th columns of the 9th row, and three grids of the 6th, 7th and 8th columns of the 10th row.

[0101] The third design area includes two grids at the 10th and 11th columns of the 3rd row, and also includes a grid at the 11th column of the 4th row and a grid at the 11th column of the 5th row.

[0102] S304: Apply the trained deep reinforcement learning model to the incentive score and the landscape plane update map, so as to iteratively update the design area through the trained deep reinforcement learning model to obtain a design expansion area.

[0103] The deep reinforcement learning model updates the landscape plan update map by updating the design area range. The grids on the design site boundary and the grids adjacent to the design area (adjacent grids include grids forming diagonals and grids with shared edges) are also used as new design areas to obtain the design expansion area.

[0104] like Figure 6 As shown, Figure 6 The grid where the mark "1" is located is defined as the first row and the first column. The design extension area includes a first design extension area, a second design extension area and a third design extension area.

[0105] like Figure 6As shown, the first design expansion area includes the grid of the 5th row and 2nd column, the grid of the 5th row and 3rd column, the grid of the 6th row and 2nd column, and also includes the three grids of the 4th row, 2nd column, 3rd column, and 4th column, and also includes the grid of the 5th row and 4th column, and also includes the two grids of the 6th row, 3rd column and 4th column, and also includes the two grids of the 7th row, 2nd column and 3rd column.

[0106] like Figure 6 As shown, the second design extension area includes the three grids of the 6th, 7th and 8th columns of the 8th row, the two grids of the 6th and 8th columns of the 9th row, the three grids of the 6th, 7th and 8th columns of the 10th row, the two grids of the 6th and 7th columns of the 7th row, the two grids of the 5th and 9th columns of the 8th row, the two grids of the 5th and 9th columns of the 9th row, the two grids of the 5th and 9th columns of the 10th row, and the two grids of the 6th and 7th columns of the 11th row.

[0107] like Figure 6 As shown, the third design expansion area includes the three grids of the 9th, 10th and 11th columns of the 2nd row, the two grids of the 10th and 11th columns of the 3rd row, the grid of the 11th column of the 4th row and the grid of the 11th column of the 5th row, the two grids of the 9th and 10th columns of the 4th row, the grid of the 10th column of the 5th row, and the two grids of the 10th and 11th columns of the 6th row.

[0108] S305, performing random sampling in the designed extension area through the trained deep reinforcement learning model to obtain a random sampling result.

[0109] S306, determining a number of design grids included in the design area, and determining a random sampling grid in the random sampling result.

[0110] S307, when the random sampling grid is located in the plurality of the design grids, the landscape plan update map is updated by deleting the random sampling grid, until the landscape plan update map after the update meets the set conditions, thereby obtaining a target landscape plan; when the random sampling grid is located outside the plurality of the design grids, the landscape plan update map is updated by adding the random sampling grid, until the landscape plan update map after the update meets the set conditions, thereby obtaining a target landscape plan.

[0111] That is, randomly select a grid from all grids included in the first design extension area, the second design extension area, and the third design extension area. Figure 7 The grid where the mark "1" is located is defined as the first row and the first column, such as Figure 7As shown in the figure, the grid in row 5, column 3 and the grid in row 7, column 7 are the randomly sampled grids. Since the grid in row 5, column 3 is within the first design region, we need to delete it. Figure 5 The grid in row 7 and column 7 is located outside the first, second, and third design regions. Figure 5 By deleting and adding grids, the updated landscape plan is as follows: Figure 8 shown.

[0112] Repeat steps S301 to S307 until the updated landscape plan meets the set conditions, and the landscape plan at this time is the target landscape plan. The set conditions are that the first score of the landscape plan at this time (the first score is used to represent the degree to which the landscape plan at this time meets the actual viewing needs) is greater than 8, the green coverage rate is greater than 35%, and the road network density is greater than 80m / hm².

[0113] In summary, the present invention uses a reinforcement learning method to generate a residential landscape plan, combining a data-based method with a rule-based method. The present invention trains a scoring model based on existing residential landscape plans, learns the layout of various landscape design elements in existing plans, and guides the layout design of each element in the reinforcement learning process together with indicators such as green space rate. It can automatically learn adaptation strategies through interaction with the environment to generate a residential landscape plan without detailed layout rules.

[0114] This embodiment also provides a landscape plan generating device, such as Fig. 9 As shown, the device comprises the following components:

[0115] The initial landscape plan generation module 01 is used to obtain layout data of the area to be designed, and apply the trained deep reinforcement learning model to the layout data to obtain the initial landscape plan of the area to be designed;

[0116] Scoring module 02, used to determine the design quality score of the initial landscape plan;

[0117] The updating module 03 is used to adjust the initial landscape plan according to the design quality score of the initial landscape plan through the trained deep reinforcement learning model to obtain a target landscape plan.

[0118] Based on the above embodiments, the present invention further provides a terminal device, whose principle block diagram can be shown as follows: Fig.10As shown. The terminal device includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal device is used to provide computing and control capabilities. The memory of the terminal device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating a landscape plan is implemented. The display screen of the terminal device can be a liquid crystal display screen or an electronic ink display screen.

[0119] Those skilled in the art will understand that Fig.10 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal device to which the scheme of the present invention is applied. The specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0120] In one embodiment, a terminal device is provided. The terminal device includes a memory, a processor, and a landscape plan map generation program stored in the memory and executable on the processor. When the processor executes the landscape plan map generation program, the following operation instructions are implemented:

[0121] Acquire layout data of the area to be designed, initialize the positions of design elements according to the layout data, and obtain an initial landscape plan;

[0122] determining a design quality score for the initial landscape plan;

[0123] According to the design quality score of the initial landscape plan, the initial landscape plan is adjusted through the trained deep reinforcement learning model to obtain the target landscape plan of the area to be designed.

[0124] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a landscape plan, characterized in that: include: Acquire layout data of the area to be designed, initialize the positions of design elements according to the layout data, and obtain an initial landscape plan; determining a design quality score for the initial landscape plan; According to the design quality score of the initial landscape plan, the initial landscape plan is adjusted through the trained deep reinforcement learning model to obtain the target landscape plan of the area to be designed, including: applying the trained deep reinforcement learning model to the design quality score of the initial landscape plan and the initial landscape plan, so as to iteratively update the initial landscape plan through the trained deep reinforcement learning model to obtain an updated landscape plan, and obtaining the design area occupied by the design elements on the updated landscape plan, so as to iteratively update the design area through the trained deep reinforcement learning model; Initializing the positions of the design elements according to the layout data to obtain an initial landscape plan includes: Identify pathways and scenic views as design elements; Determining the site boundaries and entrance and exit locations and internal building locations in the layout data; Initializing the starting position and the ending position of the road according to the entrance and exit positions; Randomly initializing the relative position of the viewing landscape between the interior of the site boundary and the interior building position; Generate an initial landscape plan according to the site boundary, the entrance and exit locations, the internal building locations, the roads after the initialization location, and the viewing landscape; Determining the design quality score of the initial landscape plan includes: Applying the trained quality scoring model to the initial landscape plan view to obtain a first score for the initial landscape plan view, wherein the first score is used to characterize the road traffic information and the viewing landscape information presented by the initial landscape plan view; Determine a second score of the initial landscape plan, the second score being used to characterize the road network data, green space data and open space area of ​​the initial landscape plan; According to the first score and the second score to obtain the design quality score of the initial landscape plan : In the formula, , , are constants, and , where M represents the quality scoring model, L represents the road network, and G represents the green space. is the score corresponding to the road network density formed by the roads on the initial landscape plan, is the score corresponding to the green coverage rate on the initial landscape plan, is the area of ​​the open space, For the second score.

2. The method for generating a landscape plan as claimed in claim 1, characterized in that: The method further comprises: adjusting the initial landscape plan map by using a trained deep reinforcement learning model according to the design quality score of the initial landscape plan map to obtain a target landscape plan map of the area to be designed; Collecting the number of iterations corresponding to the landscape plan update map, and determining an incentive score according to the number of iterations and the design quality score of the landscape plan update map; The landscape plan update map is continuously updated by using the incentive score and the trained deep reinforcement learning model until the updated landscape plan update map meets the set conditions, thereby obtaining a target landscape plan map.

3. The method for generating a landscape plan as claimed in claim 2, characterized in that: The step of continuing to update the landscape plane update map by using the incentive score and the trained deep reinforcement learning model includes: Applying a trained deep reinforcement learning model to the incentive scores and the landscape plan update map; Random sampling is performed in the design extension area through the trained deep reinforcement learning model to obtain a random sampling result, and the landscape plane update map is iteratively updated according to the relative position between the random sampling result and the design area.

4. The method for generating a landscape plan as claimed in claim 3, characterized in that: The iterative updating of the landscape plane update map according to the relative position between the random sampling result and the design area comprises: Determine a number of design grids included in the design area, and determine a random sampling grid in the random sampling result; When the random sampling grid is located in a plurality of the design grids, updating the landscape plane update map by deleting the random sampling grid; When the random sampling grid is located outside the plurality of design grids, the landscape plane update map is updated by adding the random sampling grid.

5. The method for generating a landscape plan as claimed in claim 1, characterized in that: The training methods of the trained quality scoring model and the trained deep reinforcement learning model include: Obtaining a landscape plan sample image and a benchmark score corresponding to the landscape plan sample image; Applying a quality scoring model to the landscape plan sample image to obtain a training score output by the quality scoring model; Determining a difference between the training score and the benchmark score, and adjusting parameters of the quality scoring model according to the difference to obtain a trained quality scoring model; Based on the trained quality scoring model, a deep reinforcement learning model is trained to obtain a trained deep reinforcement learning model.

6. A landscape plan generating device, characterized in that: The device comprises the following components: An initial landscape plan generation module is used to obtain layout data of the area to be designed, initialize the positions of the design elements according to the layout data, and obtain an initial landscape plan, and is also used to apply the trained deep reinforcement learning model to the design quality score of the initial landscape plan and the initial landscape plan, so as to iteratively update the initial landscape plan through the trained deep reinforcement learning model to obtain an updated landscape plan, obtain the design area occupied by the design elements on the updated landscape plan, and iteratively update the design area through the trained deep reinforcement learning model; A scoring module, used to determine a design quality score of the initial landscape plan; An updating module, configured to adjust the initial landscape plan according to the design quality score of the initial landscape plan through a trained deep reinforcement learning model to obtain a target landscape plan of the area to be designed; Initializing the positions of the design elements according to the layout data to obtain an initial landscape plan includes: Identify pathways and scenic views as design elements; Determining the site boundaries and entrance and exit locations and internal building locations in the layout data; Initializing the starting position and the ending position of the road according to the entrance and exit positions; Randomly initializing the relative position of the viewing landscape between the interior of the site boundary and the interior building position; Generate an initial landscape plan according to the site boundary, the entrance and exit locations, the internal building locations, the roads after the initialization location, and the viewing landscape; Determining the design quality score of the initial landscape plan includes: Applying the trained quality scoring model to the initial landscape plan to obtain a first score for the initial landscape plan, wherein the first score is used to characterize the road traffic information and the viewing landscape information presented by the initial landscape plan; Determine a second score of the initial landscape plan, the second score being used to characterize the road network data, green space data and open space area of ​​the initial landscape plan; According to the first score and the second score to obtain the design quality score of the initial landscape plan : In the formula, , , are constants, and , where M represents the quality scoring model, L represents the road network, and G represents the green space. is the score corresponding to the road network density formed by the roads on the initial landscape plan, is the score corresponding to the green coverage rate on the initial landscape plan, is the area of ​​the open space, For the second score.

7. A terminal device, characterized in that: The terminal device includes a memory, a processor, and a landscape plan generation program stored in the memory and executable on the processor. When the processor executes the landscape plan generation program, the steps of the landscape plan generation method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a landscape plan generation program, and when the landscape plan generation program is executed by the processor, the steps of the landscape plan generation method according to any one of claims 1 to 5 are implemented.

Images