A Layout Generation Method and Application of Jiangnan Private Garden Landscape

Abstract

The present invention discloses a method for generating the layout of a Jiangnan private garden landscape and its application. The steps of the method include: collecting images of Jiangnan private garden layout cases as original training data samples; generating target setting criteria to screen the original training data samples; cleaning the data of the original training data samples, unifying the styles of the Jiangnan private garden layout case images, annotating the screened data samples in combination with the knowledge of private garden design, and using them as training and test data sets; performing data preprocessing on the training and test data sets to meet the format conditions for inputting into the adversarial generation network model; training the adversarial generation network model to obtain a trained garden layout adversarial generation network model; inputting the test images in the test data set into the trained garden layout adversarial generation network model to obtain a garden layout scheme. The present invention realizes the automatic generation of the Jiangnan private garden landscape layout, greatly improving the quality and design efficiency of classical garden design work.

Description

Technical Field

[0001] The present invention relates to the technical field of computer-generated design, and particularly relates to a method for generating the layout of a Jiangnan private garden landscape and its application. Background Art

[0002] Jiangnan private gardens have gone through a long development process and have become an important part of the splendid Chinese culture and a precious historical and cultural heritage of all mankind. The Jiangnan private garden landscape adapts natural mountains, waters, forests and trees to local conditions and is skillfully designed according to the land use relationship, forming a spatial structure and layout full of twists and turns and artistic conception. Due to the difficulty in mastering the design methods and procedures, which are complex and changeable, the Jiangnan private garden landscape is often difficult to be formally generalized. In previous Chinese works on the design of Jiangnan private gardens, most scholars tend to analyze the characteristics of Jiangnan gardens using descriptive language based on observation and feeling. However, the traditional way of thinking to interpret gardens is based on observation and experience, and designers still need to understand by themselves how to design a new Jiangnan private garden space, and there is no intuitive landscape layout generation method provided for designers. Summary of the Invention

[0003] In order to overcome the defects and deficiencies of the prior art, the present invention provides a method for generating the layout of a Jiangnan private garden landscape. The present invention can obtain the generation result of a feasible landscape layout plan under the condition of meeting the characteristics of a specific garden style, assist garden designers to complete the design process of classical garden space. Under the background of the increasing construction volume, the generation design idea provided by the present invention can greatly improve the quality and efficiency of garden design work.

[0004] The second object of the present invention is to provide a system for generating the layout of a Jiangnan private garden landscape.

[0005] The third object of the present invention is to provide a storage medium.

[0006] The fourth object of the present invention is to provide a computing device.

[0007] In order to achieve the above objects, the present invention adopts the following technical solutions:

[0008] The present invention provides a method for generating the layout of a Jiangnan private garden landscape, including the following steps:

[0009] Collect images of Jiangnan private garden layout cases as original training data samples;

[0010] Screen the original training data samples according to the criteria set for the target of generating the Jiangnan private garden layout;

[0011] Clean the original training data samples, unify the styles of the images of Jiangnan private garden layouts, label the filtered data samples in combination with the knowledge of private garden design, and use them as training and test data sets;

[0012] Perform data preprocessing on the training and test data sets to meet the format conditions for input into the adversarial generation network model;

[0013] Train the adversarial generation network model to obtain the trained garden layout adversarial generation network model;

[0014] Input the test images in the test data set into the trained garden layout adversarial generation network model to obtain the garden layout scheme.

[0015] As a preferred technical solution, the step of labeling the filtered data samples in combination with the knowledge of private garden design specifically includes:

[0016] Simultaneously map the element plane layout information and height information in the garden layout to the image through the information mapping method. The objects to be labeled include one or more of roads, architectural features, water bodies, rockeries, or plants in the park.

[0017] As a preferred technical solution, the information mapping method includes channel-by-channel labeling and height information mapping. The channel-by-channel labeling separates the three channels in the RGB image, and each channel represents a kind of information separately. The height information mapping includes grayscale image mapping and differential labeling method mapping.

[0018] As a preferred technical solution, the data preprocessing of the training and test data sets specifically includes unifying the proportions and sizes of the images of Jiangnan private garden layouts, labeling the same functional elements in the images of Jiangnan private garden layouts with the same color, and at the same time, expanding the learning samples by rotating or flipping.

[0019] As a preferred technical solution, there is also a step of verifying and evaluating the garden layout scheme, which specifically includes: functional integrity analysis, plant generation situation analysis, height rationality analysis, and layout rationality analysis;

[0020] The functional integrity analysis performs integrity analysis on various functions in the garden layout scheme, including the block shapes, average quantities, average areas, distance relationships of each function, and the plane layout effects of each element, and checks whether there are consistent plane layout rules;

[0021] The plant generation situation analysis checks the distribution of plant generation;

[0022] The above-mentioned height rationality analysis includes the analysis of the learning situation of function markers with a fixed height in the training dataset in test cases and the analysis of the learning situation of function markers with a non-fixed height in the training dataset in test cases;

[0023] The layout rationality analysis examines the layout effect of each element and the distribution of plants.

[0024] As a preferred technical solution, there is also a correction step to correct the verified garden landscape generation scheme according to the user's design scheme to generate the final layout scheme for the landscape land. Specifically, the generated image is micro-processed and optimized through the OpenCv computer vision software library, the boundaries of some irregular buildings are regularized into rectangles, and the crown boundaries of plants are regularized into circles, so as to optimize the generation result.

[0025] As a preferred technical solution, there is also a display step to process the garden layout generation image obtained by the adversarial generation network through a script file, and combine it with the Unity or Grasshopper 3D modeling software to obtain the real-time 3D visualization effect of the garden space layout.

[0026] To achieve the above second objective, the present invention adopts the following technical solutions:

[0027] A layout generation system for Jiangnan private garden landscapes includes: an original training data sample construction module, a sample screening module, a data cleaning and annotation module, a data preprocessing module, a model training module, and an output module;

[0028] The original training data sample construction module is used to collect Jiangnan private garden layout case images as original training data samples;

[0029] The sample screening module is used to screen the original training data samples according to the target setting criteria for the generation of Jiangnan private garden layouts;

[0030] The data cleaning and annotation module is used to clean the original training data samples, unify the styles of Jiangnan private garden layout case images, annotate the screened data samples in combination with the knowledge of private garden design, and use them as training and test data sets;

[0031] The data preprocessing module is used to perform data preprocessing on the training and test data sets to meet the format conditions for inputting into the adversarial generation network model;

[0032] The model training module is used to train the adversarial generation network model to obtain a trained garden layout adversarial generation network model;

[0033] The output module is used to input the test images in the test dataset into the trained garden layout adversarial generation network model and output the garden layout plan.

[0034] To achieve the above third objective, the present invention adopts the following technical solutions:

[0035] A computer-readable storage medium stores a program, and when the program is executed by a processor, it implements the layout generation method of the Jiangnan private garden landscape as described above.

[0036] To achieve the above fourth objective, the present invention adopts the following technical solutions:

[0037] A computing device includes a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, it implements the layout generation method of the Jiangnan private garden landscape as described above.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] The present invention selects the layout of the Jiangnan private garden landscape as the generation design object and uses machine learning as the generation tool. It can obtain the generation result of a feasible landscape layout plan while meeting the characteristics of a specific garden style, assisting garden designers to complete the design process of classical garden spaces. In the context of the increasing construction volume, it can greatly improve the quality and efficiency of garden design work. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flowchart of the layout generation method of the Jiangnan private garden landscape of the present invention;

[0041] Figure 2 It is a schematic diagram of the channel annotation method of the present invention;

[0042] Figure 3 (a) It is a schematic diagram of the height mapping method of the present invention

[0043] Figure 3 (b) It is a schematic diagram of the gap annotation method of the present invention;

[0044] Figure 4 It is an extraction schematic diagram of the verification and evaluation analysis method of the present invention;

[0045] Figure 5 (a) It is a plan view of the garden layout generation result and the three-dimensional effect of the present invention

[0046] Figure 5 (b) It is a top view of the garden layout generation result and the three-dimensional effect of the present invention;

[0047] Figure 5(c) Perspective view of the generated result of the garden layout of the present invention and the 3D effect;

[0048] Figure 5 (d) Partial perspective view of the generated result of the garden layout of the present invention and the 3D effect;

[0049] Figure 5 (e) Elevation view of the generated result of the garden layout of the present invention and the 3D effect. Detailed implementation manners

[0050] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] Embodiment 1

[0052] This embodiment provides an automatic generation method for the layout of a Jiangnan private garden, including the following steps:

[0053] Step 1, data collection, collect cases of Jiangnan private garden layouts through at least one of python data crawling technology or manual collection of samples as the original training data.

[0054] In this embodiment, the collected data should be in picture format. The garden cases should be distributed in the Yangtze River Delta region, with a scale between 2,000 ㎡ and 52,000 ㎡, and the images should contain clear, distinct, and easily distinguishable main components of Jiangnan gardens.

[0055] Step 2, data screening, formulate criteria for the generation target to screen the collected samples;

[0056] Step 3, data processing, perform data cleaning on the collected flat drawings to ensure the unity of the training data style. On the basis of completing data screening, further perform knowledge annotation processing in combination with the design of private gardens. Through the information mapping method provided in this embodiment, the element plane layout information and height information can be simultaneously mapped into the pictures, and the machine can obtain complete 3D information by reading the picture data and perform statistics and learning on the feature space of the information data. The annotation scheme includes at least one of roads, architectural features, water bodies, rockeries, and plants in the park, and the annotated data is used as the training / test data set for machine learning.

[0057] In this embodiment, data processing performs data cleaning on the collected flat drawings, so that the drawing data has a training format with consistent style. On the basis of completing data screening, further annotation processing is performed on the functional elements included in the data. Through the information mapping method, the spatial layout information of different elements is mapped into the R, G, and B channel data of the picture. At the same time, the numerical values of the R and G channels are used to identify different functional elements, and the numerical value of the B channel is used to represent the height of different elements. The mapping relationship is B = H / 25 (H is the true height of the element).

[0058] The three-dimensional information mapping method includes the following two aspects:

[0059] (1) Channel-by-channel annotation. As Figure 2 shown, an RGB picture contains the R, G, and B channel data. The color of each pixel point is a vector with RGB three values. Using color blocks to represent functions is equivalent to annotating the same information with the combination of the R, G, and B channel data. When multiple pieces of information need to be annotated, the three channels can be separated, and each channel represents a piece of information alone, which is equivalent to the superposition of multiple layers of information. For example, the R channel data represents the category of the element, the G channel data represents the function of the element, and the B channel data represents the height of the element, etc. When the three channels of a picture are not enough to accommodate enough information, the number of pictures can be increased and the channels can be stacked to annotate more information, that is, two pictures are equivalent to six channels.

[0060] (2) Height information mapping. As Figure 3 (a) shows, it represents a typical height mapping method. The grayscale image maps the height information to between 0 and 255, and uses the corresponding grayscale value to represent the height of different elements, which is convenient for the modeling software to read. However, the grayscale image can only annotate the height of solid objects, and for objects with holes, such as chairs and rockeries, their true physical characteristics cannot be restored. Therefore, in addition to the grayscale image, the difference annotation method can also be used. As Figure 3 (b) shows, the annotation rule is: the pixel point R in the element represents the highest point of the solid from bottom to top, G represents the highest point of the hollow part above the height of the R value, B represents the highest point of the solid part above the height of the G value... For multiple channels, it is deduced according to this rule.

[0061] (1) Entrance. The entrance of the main park in the garden is marked with a triangle, and the RGB values are R = 255, G = 255, B = 0 respectively.

[0062] (2) Main hall. The most important hall building in the case is marked as the main hall, that is, the most important viewing point. The RG values for marking the main hall are R = 255 and G = 127, B = 150 respectively.

[0063] (3) Main landscape building. The important pavilions in the case are marked as the main landscape buildings. The marked RG values are R = 0, G = 127, B = 90 respectively.

[0064] (4) Other buildings. Except for the main hall and the main landscape building, other buildings such as pavilions and verandas in the garden are marked as other buildings, and the marked RG values are R = 127, G = 127, B = 100 respectively.

[0065] (5) Main hill. The most important landscape rockery in the case is marked as the main hill, and the RG values for marking the main hill are R = 0, G = 0, B = 135 respectively. The height of the main hill is mapped to the B channel. Based on the mountain height distribution law extracted from the analysis of the rockery information, the B values of the main hill are marked according to the mountain peak height levels of 17, 33, 40, 83, 135, 180, and 200.

[0066] (6) Main water. The main landscape water body in the case is marked as the main water, and the marked RG value is R = 255, G = 255, B = 0.

[0067] (7) Corridor. In Jiangnan private gardens, there are two situations for the viewing routes: one is the corridors, houses, and roads corresponding to the mountains and ponds; the other is the mountain paths, caves, and bridges for climbing mountains and crossing waters. The first type of route is marked as the corridor, while the second type of route is marked as the internal flow line. The corridor is marked with R = 255, G = 0, B = 75.

[0068] (8) Internal flow line. The internal flow line in the garden is arranged in the most basic form of surrounding mountains and waters. To clarify the learning points, an internal flow line that surrounds the main hill and the main water is marked for all cases, and some of its branches are connected to the external flow line corridors and buildings. The marked R = 255, G = 127, B = 17 for the internal flow line.

[0069] (9) Mezzanine. In the layout form where the corridor is close to the outer boundary and extremely tortuous, some small courtyard spaces will be enclosed with the outer boundary. Small landscapes are often set in these spaces, or the scenery outside the boundary is borrowed through louvers and door openings to enrich the viewing experience during the tour. These spaces are also part of the garden landscape space. Therefore, these spaces are marked as the mezzanine to distinguish them from other landscape spaces, and the marked RG value is R = 0, G = 127, B = 10.

[0070] (10) Grassland. The other areas in the garden except for the above elements are marked as the grassland, that is, the areas mainly landscaped by planting fruit trees and flowers. The marked RG value for the grassland is R = 0, G = 255, B = 10.

[0071] (11) Clumped trees. The information of different trees is marked by drawing circles. The diameter of the circle corresponds to the crown width of different trees, the center of the circle corresponds to the planting position of the tree, and the r value of the circle color corresponds to the height of the tree. The mapping method is r = h / 15 (h is the actual height of the tree).

[0072] (12) Spotted trees. The spotted trees are also mapped by drawing circles. The g value of the circle color corresponds to the height of the spotted tree, and g = h / 15 (h is the actual height of the tree).

[0073] In this embodiment, data processing also includes unifying the scale and size of the drawing, and performing consistent color marking on the same functional elements in the drawing. First, the case data plan view needs to be scaled to a canvas of 512px * 512px (18cm * 18cm) in size, and the plane scale is set to 1:830, that is, the 512px * 512px canvas corresponds to a real range of 130m * 130m. Then, based on the case plan view, marking is performed in Photoshop. At the same time, the expansion of machine learning samples can also be achieved through rotation, flipping, etc.

[0074] Step 4, Dataset preprocessing, realizing image compression and image combination through python code to meet the specific format requirements for input into the adversarial generation network model;

[0075] In this embodiment, the generative adversarial network model is a type of machine learning model. Since Goodfellow et al. pioneered the generative adversarial network in 2014, its application has been extended from the initial image generation to various fields of computer vision. The present invention first applies the generative model to the automatic generation of garden layouts, which plays a great role in promoting the automatic generation design of traditional garden spaces.

[0076] Step 5, Training the adversarial generation network, adjusting the artificially controlled variables, learning rate, and iteration number parameters to complete the training process of machine learning and obtain an optimized adversarial generation network model for garden layout generation;

[0077] In this embodiment, the training process of the machine learning model includes the input of the plane drawing dataset, adjusting the artificially controlled variables, learning rate, and iteration number parameters. In the established model running environment, the network parameters of the model generator and discriminator are updated through machine learning algorithms, and the continuous repetition of the update process is the model training process, completing the training process of the adversarial generation network.

[0078] The machine learning model in this embodiment can select derivative models based on the generative adversarial network (GAN) such as pix2pix, pix2pix HD, GauGAN, Cycle GAN, etc. as training tools.

[0079] Step 6, result output: Input the test images containing landscape land in the test set into the trained adversarial generation network model to obtain the garden layout plan;

[0080] As Figure 4 shown, this embodiment can perform verification and evaluation analysis on the garden generation plan according to user needs. According to the relevant treatises and reviews of Jiangnan classical gardens, this embodiment transforms the typical features in the garden space layout into index parameters on the plane graph, and analyzes and statistics the generation results based on the OpenCV software library to obtain the inspection and evaluation of the generation results. Specifically, the evaluation and analysis can be carried out from the following aspects:

[0081] (1) Analysis of functional integrity

[0082] Conduct an integrity analysis on various functions in the garden generation plan, including the color block shapes, average quantities, average areas, distance relationships of each function, and the plane layout effects of each element, and detect whether the functional elements in the layout generation results include complete corridors, mountains, waters, rich mezzanine courtyards, and a certain number of reasonably distributed buildings.

[0083] (2) Analysis of plant generation

[0084] The aforementioned functional integrity also includes whether the generation of plant information is reasonable, including the center points, crown widths, heights of arbors (marked in magenta), the dot-planting distribution of arbors, and the center points, crown widths, heights of shrubs (marked in light blue), and the group-planting distribution of shrubs.

[0085] (3) Analysis of height rationality

[0086] In this embodiment, the height rationality analysis is divided into two aspects: one is the learning situation analysis of function markers with fixed heights in the training set in the test cases, such as corridors, buildings, etc. with fixed heights; the other is the learning situation analysis of function markers with non-fixed heights in the training set in the test cases, such as rockeries.

[0087] (4) Analysis of layout rationality

[0088] In this embodiment, the layout rationality analysis refers to whether the layout effects of each element and the plant generation are reasonable, including whether the samples have a consistent plane layout rule:

[0089] A rule of "the main hall separates the main water and faces the main mountain"

[0090] The layout rule of "the main hall separates the main water and faces the main mountain" can be verified by the included angle of the main hall-main water-main mountain layout. Specifically, for all generated images, the included angle of the main water-main hall-main mountain and the viewing angle of the main hall-main mountain in their layout planes are statistically analyzed. Taking the center point of the main water, the center point of the main hall, and the center point of the main mountain as endpoints to form an angle, if this angle is obtuse, the overall layout is relatively reasonable at this time. The results are used to evaluate the rationality of the plane layout through a combination of mathematical, physical, and chemical statistics and plane diagram analysis.

[0091] B Rule of "Line-of-Sight Deviation Angle between the Main Hall and the Main Mountain"

[0092] If the garden layout conforms to the rule of "the main hall is the main viewing building", the line-of-sight deviation angle of the main hall looking far at the main mountain should be an acute angle. If the angle is too large, it indicates that the landscape orientation of the main hall is poor and it is not suitable as the main viewing building. Specifically, for all generated images, the line-of-sight deviation angle of the main hall-main mountain in their layout planes is statistically analyzed. The results are used to evaluate the rationality of the plane layout through a combination of mathematical, physical, and chemical statistics and plane diagram analysis.

[0093] In one embodiment, the method of the present invention further includes: correcting the verified garden landscape generation scheme according to the user's design scheme to generate the final layout scheme for the landscape land. Correcting the generation scheme includes micro-processing and optimizing the generated images through the OpenCv computer vision software library. The micro-processing and optimization include regularizing the boundaries of some irregular buildings into rectangles, and regularizing the crown boundaries of plants into circles, etc., so as to optimize the generation results;

[0094] In one embodiment, the method of the present invention further includes: performing a three-dimensional display of the garden layout through visualization software. Processing the garden layout generation images obtained by the adversarial generation network through a script file, and combining with three-dimensional modeling software such as Unity and Grasshopper, a real-time three-dimensional visualization effect of the garden space layout can be obtained ( Figure 5 )

[0095] Specifically, the overall effect of the garden space layout can be obtained for the generated garden layout results through the Grasshopper three-dimensional modeling software. The specific method is as follows:

[0096] (1) Read the picture information. After obtaining the garden space layout image through step 6, use the battery component in Grasshopper to separately read the b-channel height information of the 512-pixel × 512-pixel generated image, and convert it into a linear data structure to facilitate the transfer of data to the grid array;

[0097] (2) Process the picture information. The b-channel represents the size through integer values from 0 to 255. To ensure that the numerical information has practical significance, use Grasshopper to convert it to the range of 0 - 10m to represent the real height;

[0098] (3) Stretch the unit grid to increase the height. Create a 512×512 grid, match each grid with the corresponding height information, and stretch it into a unit block. The three-dimensional model of the real result can be fitted through the unit block array;

[0099] (4) Mapping functional images. Although the three-dimensional block array has a high three-dimensional meaning, users cannot distinguish and judge the distribution of each functional element, and lack a more complete architectural meaning. Therefore, Grasshopper is further used to extract the plane information of the r channel and the g channel, and the color blocks of each functional element are mapped to the three-dimensional model to provide a more intuitive three-dimensional perception, thereby generating a three-dimensional garden layout model.

[0100] Specifically, the generated garden layout result can also be used to obtain the overall effect of the garden space layout through the unity3D game engine. The specific method is as follows:

[0101] (1) Read the image information. In the results generated by the model, the r and g channel values ​​in the first image represent different functional elements, and the b channel represents the height of each element. In the second image, the r channel represents the position, height, and crown width of group-planted small trees, and the g channel represents the position, height, and crown width of point-planted large trees. By reading the RGB value of each pixel in the two images, the size, height, and position of each element can be obtained.

[0102] (2) Generate terrain. After obtaining the height and location information of the boundary, grassland, inner ring line, rockery, and water body in the image information, a terrain mesh is established based on the height information, and material coloring is performed according to different elements.

[0103] (3) Retrieving building materials. Based on the OpenCV library algorithm in Python, the rotation angles of each building in the first layout are calculated, and the results are connected to Unity3D. Combined with the building size and position information read from the image pixels, the prefabricated materials corresponding to the "main hall", "main landscape building", and "other buildings" are retrieved, scaled to the corresponding size, and set to the corresponding position. The algorithm identifies the trajectory of the corridor in the plane layout, and combined with the width and size information of the corridor, the "corridor" prefabricated parts are retrieved to generate the corridor of the entire garden.

[0104] (4) Retrieving plant materials. Obtain the size, height, crown width, and location information of the point-planted large trees and group-planted small trees in the second picture information, retrieve the corresponding "point-planted trees" and "group-planted small trees" prefabricated material, set the corresponding crown width and height, and finally place them in the corresponding position of the site.

[0105] (5) Perfect details. A certain number of dead trees are randomly generated between the corridor and the boundary, shrubs are randomly generated on the grass and rockery, and duckweed is randomly generated in the water to enrich the garden atmosphere.

[0106] (6) Model generation. The conversion script includes four files. After mounting is completed, the model generation result is input into Unity3D and run by clicking to obtain the corresponding 3D model in real time.

[0107] In one embodiment, the method of the present invention further includes: building the trained model on a computer platform adapted to the environment. After the user uploads / draws an image according to requirements and inputs it into the input end of the model for automatically generating a garden layout for testing, an image generated by the corresponding model can be obtained;

[0108] In one embodiment, the method of the present invention further includes: exporting the final building generation plan into a file of a preset type. The preset types in this embodiment include CAD geometric information, 3D model, building information model, presentation, table, bitmap, and vector graph.

[0109] Embodiment 2

[0110] This embodiment provides a layout generation system for a Jiangnan private garden landscape, including: an original training data sample construction module, a sample screening module, a data cleaning and annotation module, a data preprocessing module, a model training module, and an output module;

[0111] In this embodiment, the original training data sample construction module is used to collect images of Jiangnan private garden layout cases as original training data samples;

[0112] In this embodiment, the sample screening module is used to screen the original training data samples according to the target setting criteria for generating a Jiangnan private garden layout;

[0113] In this embodiment, the data cleaning and annotation module is used to clean the original training data samples, unify the styles of the images of Jiangnan private garden layout cases, annotate the screened data samples in combination with the knowledge of private garden design, and use them as training and test data sets;

[0114] In this embodiment, the data preprocessing module is used to perform data preprocessing on the training and test data sets to meet the format conditions for inputting into the adversarial generation network model;

[0115] In this embodiment, the model training module is used to train the adversarial generation network model to obtain a trained garden layout adversarial generation network model;

[0116] In this embodiment, the output module is used to input the test images in the test data set into the trained garden layout adversarial generation network model and output a garden layout plan.

[0117] Embodiment 3

[0118] This embodiment provides a storage medium, which can be a storage medium such as ROM, RAM, disk, optical disc, etc. The storage medium stores one or more programs, and when the programs are executed by a processor, the layout generation method of the Jiangnan private garden landscape in Embodiment 1 is implemented.

[0119] Embodiment 4

[0120] This embodiment provides a computing device, which can be a desktop computer, a notebook computer, a smart phone, a PDA handheld terminal, a tablet computer or other terminal devices with display functions. The computing device includes a processor and a memory. The memory stores one or more programs, and when the processor executes the programs stored in the memory, the layout generation method of the Jiangnan private garden landscape in Embodiment 1 is implemented.

[0121] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A method for generating the layout of a private garden landscape in the south of the Yangtze River, characterized in that, it includes the following steps: Collect images of private garden layouts in the south of the Yangtze River as the original training data samples; Screen the original training data samples according to the target setting criteria for generating the layout of private gardens in the south of the Yangtze River; Clean the original training data samples, unify the styles of the images of private garden layout cases in the south of the Yangtze River, label the screened data samples in combination with the knowledge of private garden design, and use them as the training and test data sets; The step of labeling the screened data samples in combination with the knowledge of private garden design specifically includes: Simultaneously map the plane layout information and height information of the elements in the garden layout into the image through the information mapping method. The objects to be labeled include one or more of roads, architectural features, water bodies, rockeries or plants in the park; The information mapping method includes channel-by-channel labeling and height information mapping. The channel-by-channel labeling separates the three channels in the RGB image, and each channel represents a kind of information alone. The height information mapping includes grayscale image mapping and differential labeling method mapping; Perform data preprocessing on the training and test data sets to meet the format conditions for inputting into the adversarial generation network model; Train the adversarial generation network model to obtain the trained garden layout adversarial generation network model; Input the test images in the test data set into the trained garden layout adversarial generation network model to obtain the garden layout plan; There is also a step of verifying and evaluating the garden layout plan, which specifically includes: functional integrity analysis, plant generation situation analysis, height rationality analysis and layout rationality analysis; The functional integrity analysis performs integrity analysis on various functions in the garden layout plan, including the color block shapes, average quantities, average areas, distance relationships of each function and the plane layout effects of each element, and checks whether there are consistent plane layout rules; The plant generation situation analysis checks the distribution of plant generation; The height rationality analysis includes the learning situation analysis of the function markers with fixed heights in the training data set in the test cases and the learning situation analysis of the function markers with unfixed heights in the training data set in the test cases; The layout rationality analysis checks the layout effects of each element and the distribution of plants.

2. The method for generating the layout of a private garden landscape according to claim 1, characterized in that, The data preprocessing of the training and test data sets specifically includes unifying the proportions and sizes of the images of private garden layout cases in the south of the Yangtze River, labeling the same functional elements in the images of private garden layout cases in the south of the Yangtze River with the same color, and at the same time, expanding the learning samples by rotating or flipping.

3. The method for generating the layout of a private garden landscape according to claim 1, characterized in that, There is also a correction step to correct the generated landscape plan after verification according to the user's design plan to generate the final layout plan for the landscape land. Specifically, the generated image is micro-processed and optimized by the OpenCv computer vision software library to regularize some irregular building boundaries into rectangles and the crown boundaries of plants into circles, so as to optimize the generation result.

4. The method for generating the layout of a Jiangnan private garden landscape according to claim 1, characterized in that, there is also a display step to process the generated garden layout image obtained by the adversarial generation network through a script file, and combine it with the Unity or Grasshopper 3D modeling software to obtain the real-time 3D visualization effect of the garden space layout.

5. A system for generating the layout of a Jiangnan private garden landscape, characterized in that, used to implement the method for generating the layout of a Jiangnan private garden landscape according to any one of claims 1-4, including: an original training data sample construction module, a sample screening module, a data cleaning and annotation module, a data preprocessing module, a model training module, and an output module; The original training data sample construction module is used to collect the Jiangnan private garden layout case images as the original training data samples; The sample screening module is used to screen the original training data samples according to the target setting criteria for generating the Jiangnan private garden layout; The data cleaning and annotation module is used to clean the original training data samples, unify the styles of the Jiangnan private garden layout case images, annotate the screened data samples in combination with the knowledge of private garden design, and use them as training and test data sets; The data preprocessing module is used to perform data preprocessing on the training and test data sets to meet the format conditions for inputting into the adversarial generation network model; The model training module is used to train the adversarial generation network model to obtain the trained garden layout adversarial generation network model; The output module is used to input the test images in the test data set into the trained garden layout adversarial generation network model and output the garden layout plan.

6. A computer-readable storage medium storing a program, characterized in that, when the program is executed by a processor, it implements the method for generating the layout of a Jiangnan private garden landscape according to any one of claims 1-4.

7. A computing device, including a processor and a memory for storing the processor-executable program, characterized in that, when the processor executes the program stored in the memory, it implements the method for generating the layout of a Jiangnan private garden landscape according to any one of claims 1-4.

Images