A rapid digital design method for garden landscape

Abstract

The application discloses a kind of garden scene digital quick design method, belong to garden design field, and the specific steps of this design method are as follows: (1) gather regional image information to build environment model;(2) receive design demand and carry out regional planning;(3) the environment model after planning is optimized in depth;(4) the garden model is marked feedback;The application can effectively improve the efficiency of garden landscape adjustment by block processing to the environment model, while simplifying the garden landscape design steps, improve the work efficiency of designer, by real-time self-checking to depth optimization network, it is convenient for maintenance personnel to repair, can be more intuitive for maintenance personnel to view depth optimization network operation information, it is convenient for maintenance personnel to analyze depth optimization network, improve the maintenance efficiency of maintenance personnel.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of garden design, and in particular to a digital rapid design method for garden landscape. BACKGROUND

[0002] The basic components of garden landscape can be divided into two categories: one is soft things, such as trees, water, wind, drizzle, sunlight, and sky; the other is hard things, such as paving, walls, railings, and landscape structures. Soft things are called soft landscape, which is usually natural; hard things are called hard landscape, which is usually man-made. Landscape design refers to the planning and design of scenery and gardens, which includes natural landscape elements and artificial landscape elements. It is a cross between planning, ecology, geography, and other disciplines, and has different meanings in different disciplines. Landscape design mainly serves urban landscape design, residential area landscape design, urban park planning and design, waterfront green space planning and design, tourism resort and scenic area planning and design, etc. Its main performance is to add strong artistic atmosphere to cultural squares, parks, and communities. Protecting the limited forest resources on earth is to protect the home of mankind. As a new type of ecological landscape product, the application and promotion of wood-like series of garden landscape products are our practical actions to protect the environment and cherish nature.

[0003] Through retrieval, a landscape garden design method is disclosed in Chinese Patent No. CN114528621A. Although this invention can improve the design efficiency and scientificity of designers to a certain extent, the garden landscape design steps are complicated, the garden landscape adjustment efficiency is low, and it is not convenient for maintenance personnel to repair. In addition, the existing digital rapid design method for garden landscape cannot intuitively provide maintenance personnel with deep optimization network operation information, it is difficult for maintenance personnel to analyze the deep optimization network, and the maintenance efficiency of maintenance personnel is low. Therefore, we propose a digital rapid design method for garden landscape. SUMMARY

[0004] The purpose of the present application is to solve the defects in the prior art and to provide a digital rapid design method for garden landscape.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0006] A digital rapid design method for garden landscape, the specific steps of which are as follows:

[0007] (1) Collecting regional image information to construct an environment model: the computer receives the regional image transmitted by the remote sensing satellite, and processes the collected regional image, and then constructs the corresponding environment model according to the image processing result;

[0008] (2) receiving design requirements and planning a region: a designer uploads design requirements to a computer, and then the computer analyzes the design requirements and records the planning of the environment model according to the analysis results;

[0009] (3) depth optimization of the planned environment model: a depth optimization network is constructed, which is in communication connection with the scheme sharing platform, and each group of landscape design schemes is collected for training and optimization, and then the completed environment model is imported into the depth optimization network for adjustment to generate a landscape model;

[0010] (4) labeling and feedback of the landscape model: each landscape and building in the landscape model is labeled, and the landscape model is divided into regions, and then the divided landscape model is fed back to the designer for checking and modifying.

[0011] As a further scheme of the present application, the specific steps of the image processing in step (1) are as follows:

[0012] Step one: detecting whether the regional image transmitted by the remote sensing satellite is a color image, if it is a color image, carrying out gray processing and detecting the display ratio of the regional image, if it is a gray image, directly detecting the display ratio of the regional image;

[0013] Step two: according to the detected display ratio, the regional image is divided into blocks, and then the divided regional image is converted from image space to frequency space through Fourier forward and inverse transformation to reduce noise through filtering processing, and the region required by the designer is found through gray threshold value, and the fixed region is extracted through feature point extraction.

[0014] As a further scheme of the present application, the specific calculation formula of the Fourier forward and inverse transformation in step two is as follows:

[0015]

[0016] Wherein, u and v are frequency variables, x and y are the coordinates of a pixel point of the tongue coating image, formula (1) is Fourier forward transformation, and formula (2) is Fourier inverse transformation.

[0017] As a further scheme of the present application, the specific steps of the planning record in step (2) are as follows:

[0018] First step: keyword extraction is performed on the design requirements uploaded by the designer, and then the type of the landscape design is determined according to the extracted keywords, and each group of landscape information in the design requirements is extracted;

[0019] Second step: the collected garden landscape information is classified according to soft landscape and hard landscape, and the corresponding garden landscape is constructed according to the garden landscape parameters and through related three-dimensional software;

[0020] Third step: according to the design requirements, the garden landscape is spliced in the corresponding position of the environment model, and the coordinates of the garden landscape are recorded, and then a landscape record table is generated to record the garden landscape parameter information and coordinate information.

[0021] As a further scheme of the application, the garden design type in the first step is specifically European classical garden landscape, European classical garden landscape form, British rural garden landscape, Chinese classical garden landscape, Chinese classical garden landscape or modern garden landscape.

[0022] As a further scheme of the application, the specific steps of the depth optimization network training optimization in step (3) are as follows:

[0023] S1.1: the depth optimization network collects each group of garden landscape design scheme from the scheme sharing platform, extracts the design information in each group of design scheme, and classifies the design information according to the test set and the training set;

[0024] S1.2: the accuracy of the depth optimization network is verified by repeating the test set for multiple times, the root mean square error of each group of data in the test set is counted, each group of data in the test set is predicted once, the prediction value meeting the expectation value is collected, the best data in the prediction value is output as the optimal parameter, and the training is stopped, and the specific calculation formula of the root mean square error is as follows:

[0025]

[0026] Wherein, E(y i ) represents the ith actual observation value, y i is the ith prediction value of the model, and n is the total number of data in the test set;

[0027] S1.3: according to the optimal parameter, the training set is trained to generate training samples, finally the training samples are transported to the depth optimization network, and the long-term iteration method is used to optimize the depth optimization network in real time, and the performance of the depth optimization network meeting the expectation value is evaluated, that is, the accuracy, detection rate and false alarm rate are evaluated, and the evaluation results are fed back to the maintenance personnel for checking.

[0028] As a further scheme of the application, the specific steps of the environment model adjustment in step (3) are as follows:

[0029] S2.1: The depth optimization network receives the completed environment module, and the environment model is divided into blocks according to the size ratio, and each group of environment models after block processing is numbered;

[0030] S2.2: Then, according to the numbering order, each group of environment models is extracted, and according to the landscape record table, the corresponding garden landscape in the module is optimized through input, convolution, pooling, full connection and output, and the road width direction and road boundary line in the environment model are adjusted according to the surrounding structure change;

[0031] S2.3: The computer evaluates the performance of the depth optimization network in real time through the focal loss function, and if the depth optimization network loss value does not reach the expected value, an alarm is sent, and the abnormal position of the depth optimization network is located and fed back to the relevant maintenance personnel for maintenance.

[0032] As a further scheme of the application, the focal loss function in S2.3 has the following specific calculation formula:

[0033] FL(pi)=-α(1-pi) γ log(pi) (4)

[0034] Wherein, pi represents the predicted value, alpha represents the weight factor, and gamma represents the focus parameter.

[0035] Compared with the prior art, the application has the following advantages:

[0036] 1、The garden landscape digital rapid design method compared with the previous single design method, the application receives the environment module completed by the depth optimization network, and the environment model is divided into blocks and numbered according to the size ratio, then according to the numbering order, each group of environment models is extracted, and according to the landscape record table, the corresponding garden landscape in the module is optimized, and the road width direction and road boundary line in the environment model are adjusted according to the surrounding structure change, in the optimization process, the computer evaluates the performance of the depth optimization network in real time through the focal loss function, and if the depth optimization network loss value does not reach the expected value, an alarm is sent, and the abnormal position of the depth optimization network is located and fed back to the relevant maintenance personnel for maintenance, by block processing of the environment model, the garden landscape adjustment efficiency can be effectively improved, the garden landscape design steps are simplified, the design personnel work efficiency is improved, and the depth optimization network is self-checked in real time, which is convenient for maintenance personnel to repair;

[0037] 2、The garden landscape digital rapid design method collects each set of garden landscape design information from the scheme sharing platform through the depth optimization network, classifies each set of design information according to the test set and the training set, then repeatedly verifies the accuracy of the depth optimization network through the test set multiple times, simultaneously performs prediction on each set of data in the test set once, and outputs the best data in the predicted value as the optimal parameter, simultaneously stops training, transmits the training set through learning and training to generate a training sample according to the optimal parameter, finally delivers the training sample to the depth optimization network, and adopts the long-term iteration method to optimize the depth optimization network in real time, and evaluates the performance of the depth optimization network that meets the expected value, that is, the accuracy, detection rate and false alarm rate are evaluated, and the evaluation results are fed back to the maintenance personnel for viewing, which can more intuitively enable the maintenance personnel to view the depth optimization network running information, facilitate the maintenance personnel to analyze the depth optimization network, and improve the maintenance efficiency of the maintenance personnel. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, illustrate the application, and are used to explain the application together with the embodiments of the application, and do not constitute a limitation on the application.

[0039] Figure 1 A flow chart of a garden landscape digital rapid design method according to the present application. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments.

[0041] Embodiment 1

[0042] Reference Figure 1 A garden landscape digital rapid design method, the specific steps of which are as follows:

[0043] Collecting regional image information to construct an environment model: the computer receives the regional image transmitted by the remote sensing satellite, and performs image processing on the collected regional image, and then constructs a corresponding environment model according to the image processing result.

[0044] Specifically, it is detected whether the regional image transmitted by the remote sensing satellite is a color image, if it is a color image, it is processed to gray scale and the display ratio of the regional image is detected, if it is a gray image, the display ratio of the regional image is directly detected, then the regional image is processed by block according to the detected display ratio, and then the regional image after block processing is converted from image space to frequency space through Fourier forward and inverse transformation to reduce noise through filtering processing, and the required area of the designer is found through gray threshold value, and the fixed area is extracted through feature point extraction.

[0045] It should be further explained that the specific calculation formula of Fourier forward and inverse transformation is as follows:

[0046]

[0047] Wherein, u and v are frequency variables, x and y are the coordinates of a pixel point of the tongue coating image, formula (1) is Fourier forward transformation, and formula (2) is Fourier inverse transformation.

[0048] Receiving design requirements and planning regions: the designer uploads the design requirements to the computer, and then the computer detects and analyzes the design requirements, and plans and records the environment model according to the analysis results.

[0049] Specifically, first, the computer extracts keywords from the design requirements uploaded by the designer, then determines the garden design type according to the extracted keywords, extracts each group of garden landscape information in the design requirements, classifies the collected garden landscape information according to soft landscape and hard landscape, determines the corresponding garden landscape according to each group of garden landscape parameters and through related three-dimensional software, then splices each group of garden landscape in the corresponding position of the environment model according to the design requirements, records the coordinates of each group of garden landscape, and generates a landscape record table to record the garden landscape parameter information and coordinate information.

[0050] It should be further explained that the garden design type is specifically European classical garden landscape, European classical garden landscape form, British rural garden landscape, Chinese classical garden landscape, Chinese classical garden landscape or modern garden landscape.

[0051] Embodiment 2

[0052] Referring to Figure 1 A kind of digital rapid design method of garden landscape, the specific steps of the design method are as follows:

[0053] The planned environment model is deeply optimized: a deep optimization network is constructed, the deep optimization network is in communication connection with the scheme sharing platform, and each group of garden landscape design schemes is collected to train and optimize, and then the environment model after planning is imported into the deep optimization network to adjust to generate a garden model.

[0054] Specifically, the deep optimization network collects each group of garden landscape design schemes from the scheme sharing platform, extracts the design information in each group of design schemes, classifies each group of design information according to the test set and the training set, then repeatedly verifies the accuracy of the deep optimization network through the test set multiple times, and calculates the root mean square error of each group of data in the test set, and each group of data in the test set is predicted once, the predicted value meeting the expected value is collected, and the best data in the predicted value is output as the optimal parameter, and the training is stopped. According to the optimal parameter, the training set is trained to generate a training sample, and finally the training sample is transported to the deep optimization network, and the long-term iteration method is used to optimize the deep optimization network in real time, and the deep optimization network meeting the expected value is evaluated in performance, that is, the accuracy, detection rate and false alarm rate are evaluated, and the evaluation results are fed back to the maintenance personnel for viewing, which can more intuitively show the maintenance personnel the running information of the deep optimization network, facilitate the maintenance personnel to analyze the deep optimization network, and improve the maintenance efficiency of the maintenance personnel.

[0055] It should be further pointed out that the specific calculation formula of the root mean square error is as follows:

[0056]

[0057] Where, E(y i ) represents the ith actual observation value, y i is the ith predicted value of the model, and n is the total number of data in the test set.

[0058] Specifically, the deep optimization network receives the completed environment module, and blocks the environment model according to the size ratio, and numbers each group of environment models after blocking, then extracts each group of environment models according to the numbering order, and optimizes the corresponding landscape in the module through input, convolution, pooling, full connection and output according to the landscape record table, and adjusts the road width and road boundary line in the environment model according to the surrounding structure change. During the optimization process, the computer evaluates the performance of the deep optimization network in real time through the focal loss function. If the loss value of the deep optimization network does not reach the expected value, an alarm will be sent, and the abnormal position of the deep optimization network will be located and fed back to the relevant maintenance personnel for maintenance. By blocking the environment model, the efficiency of adjusting the landscape can be effectively improved, the design steps of the landscape can be simplified, and the work efficiency of the designer can be improved. Through real-time self-checking of the deep optimization network, it is convenient for maintenance personnel to repair.

[0059] It should be further pointed out that the specific calculation formula of the focal loss function is as follows:

[0060] FL(pi)=-α(1-pi) γ log(pi) (4)

[0061] Wherein, pi represents the predicted value, alpha represents the weight factor, and gamma represents the focusing parameter.

[0062] The garden model is labeled and fed back: the landscapes and buildings in the garden model are labeled, and the garden model is divided into regions, and then the divided garden model is fed back to the designer for checking and modifying.

[0063] The above describes only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for rapid digital design of a landscape, characterized in that, The design method comprises the following specific steps: (1) Collect regional image information to build an environment model: the computer receives the regional image transmitted by the remote sensing satellite, and processes the collected regional image, and then builds a corresponding environment model according to the image processing result; (2) Receive design requirements and plan the region: the designer uploads the design requirements to the computer, and then the computer analyzes the design requirements, and records the planning of the environment model according to the analysis result; (3) Depth optimization of the planned environment model: a depth optimization network is constructed, which is in communication connection with the scheme sharing platform, and collects various groups of landscape design schemes for training and optimization, and then the completed environment model is imported into the depth optimization network for adjustment to generate a landscape model; (4) Label the feedback of the landscape model: label each landscape and building in the landscape model, and divide the landscape model into regions, and then feed back the divided landscape model to the designer for checking and modifying; The planning record in step (2) comprises the following specific steps: First step: extract the keywords of the design requirements uploaded by the designer, and then determine the landscape design type according to the extracted keywords, and extract the landscape information in the design requirements; Second step: classify the collected landscape information according to soft landscape and hard landscape, and construct the corresponding landscape by related three-dimensional software according to the landscape parameters; Third step: splice each group of landscape in the corresponding position of the environment model according to the design requirements, record the coordinates of each group of landscape, and then generate a landscape record table to record the landscape parameter information and coordinate information; The environment model adjustment in step (3) comprises the following specific steps: S2.1: The depth optimization network receives the completed environment model, and processes the environment model according to the size ratio, and numbers each group of environment model after block processing; S2.2: Then extract each group of environment model according to the number sequence, and optimize the corresponding landscape in the module by input, convolution, pooling, full connection and output according to the landscape record table, and adjust the road width and boundary line of the environment model according to the surrounding structure change; S2.3: The computer evaluates the performance of the depth optimization network in real time through the focal loss function, and if the loss value of the depth optimization network does not reach the expected value, an alarm is sent, and the abnormal position of the depth optimization network is located and fed back to the relevant maintenance personnel for maintenance.

2. The method according to claim 1, wherein, The image processing in step (1) comprises the following specific steps: Step one: detect whether the regional image transmitted by the remote sensing satellite is a color image, if it is a color image, process it to gray scale and detect the display ratio of the regional image, if it is a gray image, directly detect the display ratio of the regional image; Step two: according to the detected display ratio, the image of the region is divided into blocks, and then the image of the region after block processing is converted into frequency space through Fourier forward and inverse transformation to reduce noise, and the required region of the designer is found through gray threshold value, and the fixed region is extracted through feature point extraction.

3. The method of claim 1, wherein, The garden design type in the first step is specifically a European classical garden landscape, a European classical garden landscape form, a British rural garden landscape, a Chinese classical garden landscape, a Chinese classical garden landscape, or a modern garden landscape.

4. The method of claim 1, wherein, The specific steps of the deep optimization network training optimization in step (3) are as follows: S1.1: The deep optimization network collects each group of garden landscape design schemes from the scheme sharing platform, extracts the design information in each group of design schemes, and classifies each group of design information according to the test set and the training set; S1.2: The accuracy of the deep optimization network is verified by repeating the test set multiple times, and the root mean square error of each group of data in the test set is counted, and each group of data in the test set is predicted once, the prediction value that meets the expected value is collected, and the best data in the prediction value is output as the optimal parameter, and the training is stopped, and the root mean square error is calculated as follows: where E(y i ) represents the ith actual observation, y i is the ith predicted value from the model, and n is the total number of data in the test set. S1.3: According to the optimal parameter, the training set is trained to generate a training sample, and finally the training sample is fed into the deep optimization network, and the long-term iteration method is used to optimize the deep optimization network in real time, and the performance of the deep optimization network that meets the expected value is evaluated, that is, the accuracy, detection rate and false alarm rate are evaluated, and the evaluation results are fed back to the maintenance personnel for checking.

5. The method of claim 1, wherein, The specific calculation formula of the focal loss function in S2.3 is as follows: FL(pi) = -a(l - pi) γ log(pi) (4) Where, pi represents the prediction value, alpha represents the weight factor, and gamma represents the focus parameter.

Images