Method for constructing a database of green space structure types in a block based on deep learning

Abstract

The application discloses a deep learning-based street block green space structure type database construction method, relates to the technical field of artificial intelligence, city planning and landscape gardening, and comprises the following steps: firstly, pre-processing city remote sensing images, checking and identifying the green space and the processed city remote sensing images, constructing a street block green space structure type system according to the green space identification result, constructing a street block green space structure training data set based on the street block green space structure type system, constructing a preliminary street block green space structure screening model, training the preliminary street block green space structure screening model by using the street block green space structure training data set, completing the construction of the street block green space structure screening model, outputting a screening result, obtaining a plurality of street block green space structure type data, and establishing a street block green space structure type database. The method can efficiently and widely analyze a large amount of information samples across time periods, thereby effectively improving the stability and accuracy of the evaluation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, urban planning and landscape architecture, in particular to a street green space structure type database construction method based on deep learning. BACKGROUND

[0002] Urban development has entered a new stage of stock optimization from large-scale incremental construction, and urban renewal planning work focusing on "connotation development", "vitalizing stock" and "intensive and efficient" has begun to spread in large and medium-sized cities. The spatial unit of urban green space planning and regulation has shifted from traditional large-scale "city" and "urban area" to more refined small-scale "street" and "community". The structure of urban green space is an important factor affecting the service level of green space. Under this background, identifying the green space structure type at the street level is the basis for judging the service mode of street green space, and further provides direction for subsequent urban green space planning optimization and regulation.

[0003] Due to the large number of streets in the city and the great differences in internal functions, land structure and development intensity, relying on traditional on-site recording, investigation and analysis methods usually requires a lot of manpower, material resources and financial resources, and the spatial scope and information processing scale have great limitations, making it difficult to effectively meet the accuracy and breadth requirements of this task. SUMMARY

[0004] To solve the problems mentioned in the background, the purpose of the present application is to provide a street green space structure type database construction method based on deep learning.

[0005] The purpose of the present application can be achieved by the following technical scheme: a street green space structure type database construction method based on deep learning, the method comprising the following steps:

[0006] Filter and collect urban remote sensing images, and perform image correction processing on the urban remote sensing images;

[0007] Based on the processed urban remote sensing images, the green space is verified with the urban remote sensing images, and the green space recognition is completed;

[0008] Based on the result of green space recognition, the result of green space recognition is divided into grid units, and the street green space unit is obtained;

[0009] The street green space structure in the street green space unit is digitally extracted, the indicators are selected to describe the street green space structure type, and the street green space structure indicators are obtained;

[0010] Based on the different street green space structure types corresponding to the selected street green space structure indicators, a street green space structure type system is constructed;

[0011] Based on the constructed type system of the block green space structure, the unit sample labels are marked in the block green space unit sample according to the types of the block green space structure, and the unit sample labels are data enhanced to construct a block green space structure training dataset;

[0012] Based on the VGG16 convolutional neural network model, the convolution kernel size and step are modified and debugged according to the block green space structure discrimination needs to construct a preliminary block green space structure discrimination model;

[0013] The constructed block green space structure training dataset is used to train the preliminary block green space structure discrimination model, and the convolution kernel size and step of the preliminary block green space structure discrimination model are adjusted until the accuracy reaches the best value, so as to complete the construction of the block green space structure discrimination model;

[0014] The block green space structure training dataset is used to optimize and adjust the block green space structure discrimination model, and the block green space structure discrimination model is retrained;

[0015] Through multiple retraining of the block green space structure discrimination model, the model parameters are adjusted until the accuracy reaches the best value, so as to complete the construction of the block green space structure discrimination model;

[0016] The block green space unit sample is input into the block green space structure discrimination model, and the discrimination result is output;

[0017] The output discrimination result is the block green space structure type data of the input region, and the discrimination results are summarized to establish a block green space structure type database of the input region.

[0018] Preferably, the urban remote sensing image is derived from remote sensing satellite data of Landsat, Sentinel and Spot; the screening requirements of the urban remote sensing image include that the cloud cover of the urban remote sensing image is less than 5%, the shooting time is from June to October, and the accuracy of the panchromatic band of the remote sensing image is 10m, and the accuracy of other bands is at least 30m; the image correction processing of the urban remote sensing image includes geographic calibration, radiation calibration, image registration, image mosaic and image cropping.

[0019] Preferably, the process of green space recognition is as follows:

[0020] According to the processed urban remote sensing image, samples of green land, water body and building area categories are selected, the separation degree between sample categories is verified, and the separation degree parameter value is greater than or equal to 1.8, otherwise the sample is reselected until the separation degree parameter value between sample categories is greater than or equal to 1.8, and the qualified sample is output; the qualified sample is input, and the support vector machine is used for supervised classification of the urban remote sensing image; the confusion matrix is used to compare the green land supervised classification result with the measured value, and when the Kappa coefficient is greater than or equal to 95%, the urban green space recognition result is output.

[0021] Preferably, the extraction process of the street green space unit is as follows:

[0022] According to the urban green space recognition result, data is imported into the Arcgis platform, the reclassification tool is used to assign the green space as 1 and the other parts as 0, and the green space raster image is output; the pixel grid segmentation tool is used in the Arcgis platform to perform pixel grid segmentation on the green space raster image, and the segmented street green space unit sample is output.

[0023] Preferably, according to the obtained street green space unit, the street green space structure characteristics are divided into 1, 2, …, n aspects; n different indexes are selected for digital description for each aspect; n indexes are used to calculate the street green space unit sample, and the collinearity evaluation is performed on the calculation result; if the selected n indexes are not collinear, the n street green space structure indexes are output, otherwise the indexes are reselected until the collinearity evaluation is not collinear, and the n street green space structure indexes are output.

[0024] Preferably, according to the n street green space structure indexes, each index is represented as 1, 2, …, m grades; n index types and m grades are combined to finally form m n street green space structure types.

[0025] Preferably, the process of constructing the street green space structure training data set is as follows:

[0026] Based on the m n street green space structure types, in the street green space unit sample, the classification result label is marked, the number is 10% of the total sample, and the marked sample is data enhanced by cropping and rotating, and the sample amount is increased to n times of the original, to construct the training street green space structure discrimination data set.

[0027] Preferably, a street green space structure type database construction system comprises:

[0028] A preprocessing module: filtering and collecting the urban remote sensing image, and performing image correction processing on the urban remote sensing image;

[0029] a green space identification module: based on the processed urban remote sensing image, the green space is identified after mutual checking with the urban remote sensing image;

[0030] a green space structure type system construction module: a street block green space unit sample is extracted, sample characteristics are summarized, street block green space structure indexes are selected, and a street block green space structure type system is constructed;

[0031] a green space structure identification model construction module: a street block green space structure training data set is constructed; based on a VGG16 convolutional neural network model, the convolution kernel size and step are modified and debugged according to the street block green space structure identification needs, a preliminary street block green space structure identification model is constructed; the street block green space structure training data set is used to train the preliminary street block green space structure identification model, the convolution kernel size and step and other parameters of the preliminary street block green space structure identification model are adjusted until the accuracy rate reaches the best value, so that the street block green space structure identification model construction is completed.

[0032] Preferably, an apparatus comprising:

[0033] one or more processors;

[0034] a memory for storing one or more programs;

[0035] when one or more of the programs are executed by one or more of the processors, so that one or more of the processors implement the deep learning-based street block green space structure type database construction method as described above.

[0036] Preferably, a storage medium containing computer executable instructions for executing the deep learning-based street block green space structure type database construction method as described above when executed by a computer processor.

[0037] The beneficial effects of the present application are:

[0038] The present application breaks through the scale limitation of traditional green space structure type identification by means of machine learning technology, can efficiently analyze a large amount of information samples across time periods, and effectively improves the stability and accuracy of the evaluation results. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows, and obviously, other drawings can also be obtained by those skilled in the art without creative labor on the premise of not paying creative labor;

[0040] Figure 1 is the overall framework diagram of the present application;

[0041] Figure 2 is the green space recognition module framework diagram of the present application;

[0042] Figure 3 is the block green space structure type system construction module schematic diagram of the present application;

[0043] Figure 4 is the block green space structure discrimination model construction process schematic diagram of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described 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. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0045] As shown in Figure 1 , the block green space structure type database construction method based on deep learning comprises the following steps:

[0046] S1: city remote sensing image preprocessing

[0047] The city remote sensing image is screened and collected, and the city remote sensing image is processed by image correction;

[0048] The city remote sensing image preprocessing specifically includes:

[0049] Landsat, Sentinel, Spot and other common remote sensing satellite data are obtained; the screening requirements include that the cloud cover of the remote sensing image is less than 5%, the shooting time is summer (June-October), and the full-color band accuracy of the remote sensing image is 10 m, and the accuracy of other bands is at least 30 m; the remote sensing image data preprocessing includes geographic calibration, radiation calibration, image registration, image inlay, and image cropping.

[0050] S2: green space recognition model construction

[0051] Based on the city remote sensing image preprocessed by S1, the city green space is identified, and the city green space recognition is completed after mutual verification with the city remote sensing image;

[0052] The green space recognition model construction specifically includes:

[0053] According to the city remote sensing image data preprocessing result in S1, the samples of green space, water body, building area and other categories are manually selected, the separation degree between sample categories is verified, that is, the Jeffries-Matustia and Transfromed Divergence parameter values are greater than or equal to 1.8, otherwise the samples are reselected until the separation degree parameter values between sample categories are greater than or equal to 1.8, and then the samples are output; the qualified samples are input, and the support vector machine is used for supervised classification of the city remote sensing image; the confusion matrix is used to compare the green space supervised classification result with the measured value, and when the Kappa coefficient is greater than or equal to 95%, the city green space recognition result can be output.

[0054] S3: Construction of street block green space structure type system

[0055] The construction of the street block green space structure type system includes S31, extraction of street block green space unit samples, S32, selection of street block green space structure indexes, and S33, completion of the construction of the street block green space structure type system.

[0056] S31: Extraction of street block green space unit samples

[0057] Based on the green space recognition result of S2, the grid unit division is performed, and the extraction of the street block green space unit samples is completed.

[0058] Specifically, it includes:

[0059] According to the green space recognition result, the data is imported into the Arcgis platform, the reclassification tool is used to assign the green space as 1 and the other parts as 0, and the green space raster image is output; the pixel grid division is performed on the green space raster image by using the grid division tool in the Arcgis platform, and the divided street block green space unit samples are output, and the extraction of the city green space unit samples is completed. In this embodiment, the grid division is taken as an example, and the size of the unit grid should be reasonably selected according to the actual research.

[0060] S32: Selection of green space structure indexes

[0061] The green space structure in S31 unit is digitally extracted, and appropriate indexes are selected to describe the green space structure type in S31 unit.

[0062] Specifically, it includes:

[0063] According to the subjective perception of the block green space unit sample, the artificial block green space form feature is divided into 1, 2, …, n aspects; for each aspect, select appropriate indicators for digital description, a total of n different indicators; use n indicators to calculate the block green space unit sample, and evaluate the collinearity of the calculation results. If the selected n indicators are not collinear, output n block green space form indicators, otherwise reselect indicators until the collinearity evaluation is not collinear, and output n block green space form indicators. The block green space form indicator selection is completed.

[0064] In this embodiment, the artificial block green space structure feature is divided into 3 aspects; for each aspect, select appropriate indicators for digital description, which are green amount, shape index, and distribution uniformity, corresponding to green coverage rate, related circumscribed circle index, and mixed standard deviation, a total of 3 different indicators; use the selected 3 indicators to calculate the block green space unit sample, and evaluate the collinearity of the calculation results. If the selected 3 indicators are not collinear, output 3 block green space structure indicators, and the block green space structure indicator selection is completed.

[0065] S33: The block green space structure type system construction is completed

[0066] Based on the selected indicator combination of S32, different block green space structure types are selected to complete the block green space structure type system construction;

[0067] Specifically, it includes:

[0068] For the selected n block green space structure indicators, each indicator is represented as 1, 2, …, m grades according to the actual situation; combine n indicator types with m grades to finally form m n block green space structure types

[0069] It should be noted that for the selected green coverage rate, related circumscribed circle index, and mixed standard deviation, a total of 3 block green space structure indicators; each indicator is represented as low (L) and high (H) according to the actual situation, a total of 2 grades; wherein, Lc represents low green amount, the value range is (0-0.3], Hc represents high green amount, the value range is (0.3-1], Ls represents low shape index, the value range is (0-0.7], Hs represents high shape index, the value range is (0.7-1), Lt represents low distribution uniformity, the value range is (0-0.3], Ht represents high distribution uniformity, the value range is (0.3-0.5); combine 3 indicator types with 2 grades to finally form Lc-Ls-Lt, Lc-Ls-Ht, Lc-Hs-Ht, Lc-Hs-Lt, Hc-Ls-Lt, Hc-Ls-Ht, Hc-Hs-Lt, Hc-Hs-Ht a total of 8 block green space structure types.

[0070] S4: Construction of a block green space structure discrimination model

[0071] The construction of the block green space structure discrimination model includes S41: construction of a block green space structure training dataset, S42: construction of a preliminary block green space structure discrimination model, S43: training and testing of the preliminary block green space structure discrimination model, and S44: completion of the construction of the block green space structure discrimination model.

[0072] S41: Construction of a block green space structure training dataset

[0073] Based on the block green space form classification model established in S32, the unit sample labels are marked in the block green space unit samples extracted in S31 according to each form, and the extracted samples are data enhanced to construct a training block green space form classification dataset;

[0074] Specifically, it includes:

[0075] Based on the eight block green space form classifications output by S32, in the entire block green space unit sample extracted in S31, eight labels are manually marked, including Lc-Ls-Lt, Lc-Ls-Ht, Lc-Hs-Ht, Lc-Hs-Lt, Hc-Ls-Lt, Hc-Ls-Ht, Hc-Hs-Lt, and Hc-Hs-Ht. The number of training samples is usually 10% of the total number of samples. In this embodiment, there are 3996 unit samples in the urban central city, so 50 representative samples are selected under each of the eight labels, and the selected samples are data enhanced through cropping, rotation, and other operations to increase the sample size to 10 times the original, to construct a block green space form classification training dataset containing 4000 samples covering eight types;

[0076] S42: Construction of a preliminary block green space structure discrimination model

[0077] The VGG16 convolutional neural network model proposed by the Visual Geometry Group (VGG) of Oxford University is selected to preliminarily construct a block green space structure discrimination model using the original parameters and structure of the model;

[0078] S43: Training and testing of the preliminary block green space structure discrimination model

[0079] Based on the preliminary block green space structure discrimination model constructed in S42, the model is trained and tested using the training dataset in S41, and the convolution kernel size and step parameters in the preliminary block green space structure discrimination model are adjusted;

[0080] Specifically, it includes:

[0081] Taking the block green space structure classification in the embodiment as an example, a preset algorithm with image classification processing capability is selected, and the VGG16 convolutional neural network model proposed by the Visual Geometry Group (VGG) of the University of Oxford is selected in this example. The model is preliminarily trained and tested by using the S41 training data set, and the parameters such as the size and step of the convolution kernel are adjusted.

[0082] S44: The green space structure discrimination model is constructed

[0083] Through multiple S42 block green space structure discrimination model training, the model parameters are adjusted. If the accuracy of Lc-Ls-Lt is low, the samples in the Lc-Ls-Lt category can be considered for adjustment, and the samples are further characterized and extracted until the accuracy reaches 85%, and the block green space structure discrimination model is constructed;

[0084] S5: Preprocessing data block green space structure discrimination and result output

[0085] The block green space unit samples extracted by S31 are input into the block green space structure discrimination model constructed by S5, and the classification results are output.

[0086] Specifically, it includes:

[0087] The total block green space unit samples to be analyzed, a total of 3996, are classified by using the block green space structure discrimination model adjusted in S44, and the classification data is output.

[0088] S6: Constructing a block green space structure type database

[0089] Based on the classification of S5, a plurality of block green space structure type data is obtained, and a block green space structure type database is established.

[0090] It should be further noted that based on the same inventive concept, the present application further provides a computer storage medium, which stores a computer program, and the computer program is run by a processor to execute the above method. The storage medium can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples (non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus.

[0091] In the description of the present application, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0092] The basic principles, main features and advantages of the present disclosure are shown and described above. It should be understood by those skilled in the art that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, various changes and improvements can be made to the present disclosure, and these changes and improvements all fall within the scope of the claimed present disclosure.

Claims

1. A method for constructing a database of green space structure types in a block based on deep learning, characterized by, The method comprises the following steps: The urban remote sensing image is screened and collected, and the urban remote sensing image is subjected to image correction processing; Based on the processed urban remote sensing image, the green space is verified with the urban remote sensing image, and the green space recognition is completed; Based on the result of the green space recognition, the result of the green space recognition is divided into grid units, and the block green space unit is obtained; The block green space structure in the block green space unit is digitally extracted, the index is selected to describe the block green space structure type, and the block green space structure index is obtained; Based on the selected block green space structure index corresponding to different block green space structure types, a block green space structure type system is constructed; Based on the constructed block green space structure type system, unit sample labels are selected in the block green space unit sample according to the block green space structure of each type, and the unit sample labels are subjected to data enhancement to construct a block green space structure training data set; Based on the VGG16 convolutional neural network model, the convolution kernel size and step are modified and debugged according to the block green space structure identification needs, and a preliminary block green space structure identification model is constructed; The constructed block green space structure training data set is used to train the preliminary block green space structure identification model, and the convolution kernel size and step of the preliminary block green space structure identification model are adjusted until the accuracy reaches the best value, so as to complete the construction of the block green space structure identification model; The block green space unit sample is input into the block green space structure identification model, and the identification result is output; The output identification result is the block green space structure type data of the input region, and a block green space structure type database of the input region is established based on the identification result. 2.The deep learning-based block green space structure type database construction method of claim 1, wherein, The urban remote sensing image is derived from Landsat, Sentinel and Spot remote sensing satellite data; the screening requirements of the urban remote sensing image include that the cloud cover of the urban remote sensing image is less than 5%, the shooting time is from June to October, and the accuracy of the urban remote sensing image is 10m; the image correction processing process of the urban remote sensing image includes geographic calibration, radiation calibration, image registration, image mosaic and image cropping. 3.The deep learning-based block green space structure type database construction method of claim 1, wherein, The green space recognition process is as follows: According to the processed urban remote sensing image, samples of green land, water body and building area are selected, the separation degree between sample categories is verified, and the separation degree parameter value is greater than or equal to 1.8; otherwise, the sample is reselected until the separation degree parameter value between the sample categories is greater than or equal to 1.8, and the qualified sample is output; the qualified sample is input, and the support vector machine is used for supervised classification of the urban remote sensing image; the confusion matrix is used to compare the green land supervised classification result with the measured value, and when the Kappa coefficient is greater than 95%, the urban green space recognition result is output. 4.The deep learning-based block green space structure type database construction method of claim 1, wherein, The extraction process of the block green space unit is as follows: According to the urban green space recognition result, the data is imported into the Arcgis platform, the reclassification tool is used, the green space is assigned as 1, and the other parts are assigned as 0, and a green space raster image is output; in the Arcgis platform, the raster segmentation tool is used, the green space raster image is subjected to pixel grid segmentation, and a segmented block green space unit sample is output. 5.The deep learning-based block green space structure type database construction method of claim 4, wherein, According to the obtained block green space unit, the block green space structure feature is divided into 1, 2, …, n aspects; for each aspect, an index is selected for digital description, and a total of n different indexes are selected; the n indexes are used to calculate the block green space unit sample, the calculation result is subjected to collinearity evaluation, the n selected indexes are not collinear, then n block green space structure indexes are output, otherwise the indexes are reselected until the collinearity evaluation is not collinear, and then the n block green space structure indexes are output. 6.The deep learning-based block green space structure type database construction method of claim 5, wherein, According to the n block green space structure indexes, each index is represented as 1, 2, …, m grades; n index types are combined with m grades to finally form m n block green space structure types. 7.The deep learning-based block green space structure type database construction method of claim 1, wherein, The process of constructing the block green space structure training data set is as follows: Based on m n The number of the marked classification result labels is 10% of the total samples. The marked samples are data enhanced by cropping and rotating to increase the sample quantity to n times of the original quantity, thereby constructing a training block green space structure discrimination dataset. 8.A system for constructing a database of block green space structure types, comprising: It includes: A preprocessing module: screening and collecting urban remote sensing images, and performing image correction processing on the urban remote sensing images; A green space recognition module: based on the processed urban remote sensing images, the green space is verified with the urban remote sensing images, and the urban green space recognition is completed; A green space structure type system construction module: extracting block green space unit samples, summarizing sample features, selecting block green space structure indexes, and constructing a block green space structure type system; A green space structure discrimination model construction module: constructing a block green space structure training data set; Based on the VGG16 convolutional neural network model, the convolution kernel size and step are modified and debugged according to the needs of block green space structure discrimination, and a preliminary block green space structure discrimination model is constructed; The constructed block green space structure training data set is used to train the preliminary block green space structure discrimination model, and the convolution kernel size and step of the preliminary block green space structure discrimination model are adjusted, until the accuracy reaches the best value, so as to complete the construction of the block green space structure discrimination model.

9. An apparatus, comprising: It includes: One or more processors; Memory for storing one or more programs; When one or more of the programs are executed by one or more of the processors, one or more of the processors implement the deep learning-based block green space structure type database construction method according to any one of claims 1-7.

10. A storage medium containing computer-executable instructions, wherein: The computer executable instructions, when executed by a computer processor, are used to execute the deep learning-based block green space structure type database construction method according to any one of claims 1-7. The computer executable instructions, when executed by a computer processor, are used to execute the deep learning-based block green space structure type database construction method according to any one of claims 1-7.

Images