Urban building unit data acquisition method based on machine learning

Abstract

The invention discloses an urban building unit data acquisition method based on machine learning, which comprises an optical remote sensing camera, a ground control station, a data processing center, a point cloud scanner, a three-dimensional ground penetrating radar and a remote sensing data system composed of a deep learning system, the optical remote sensing camera, the ground control station and the data processing center. The beneficial effects of the invention are that the method employs a satellite remote sensing technology to carry out the three-dimensional image of the building in the urban district, obtains the detail data of the existing building through point cloud three-dimensional scanning, builds a three-dimensional building model, and employs an AI algorithm to carry out the intelligent recognition of a building unit; underground pipe network distribution is identified through a three-dimensional ground penetrating radar, an underground pipeline model is established in combination with a BIM technology, a building model and the underground pipeline model are integrated, a city unit gene pool is established, machine learning is performed through a neural network model, city building unit identification efficiency is improved, and city update and development are powerfully promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of urban building data acquisition, and in particular to a method for acquiring urban building unit data based on machine learning. Background Art

[0002] With the acceleration of urbanization, the number of urban buildings has increased dramatically, and the demand for accurate and efficient acquisition of urban building unit data has become increasingly urgent. Traditional data acquisition methods, such as manual field measurement and mapping based on simple image recognition, have many problems such as low efficiency, difficulty in ensuring accuracy, and being greatly restricted by environmental factors. For example, manual field measurement requires a lot of manpower, material resources and time costs, and it is difficult to fully implement in complex terrain or dangerous areas.

[0003] In urban planning, construction and management, it is crucial to accurately obtain urban building unit data. Traditional data acquisition methods have problems such as low efficiency, low accuracy and incomplete data.

[0004] For example, a Chinese patent discloses an intelligent city planning system and method based on urban big data analysis (publication number: CN115984075A), the system includes: a data acquisition unit, configured to acquire urban planning data; the urban planning data includes at least three dimensions of data, namely: physical parameters, operating parameters and environmental parameters; the physical parameters include at least: the area and height of each building unit in the city; the operating parameters include at least: the type, flow of people and direction of flow of people of each building unit in the city; the environmental parameters include at least: the average temperature and meteorological information of each building unit in the city. It uses the parameters of various dimensions of building units in the city to construct the aggregate of the city, and then constructs the link according to the information flow of the aggregate, realizing the automation and intelligence of urban planning. Compared with the existing technology, the urban planning scheme formulated is more reasonable, more efficient and more applicable.

[0005] The vigorous development of machine learning technology has provided an innovative way to solve this dilemma. It has powerful data processing capabilities and can automatically mine valuable information from massive and complex data, and is expected to achieve efficient and accurate collection of urban building unit data. With the development of machine learning technology, it has provided a new way to efficiently and accurately obtain urban building unit data.

[0006] Therefore, in response to the above technical problems, it is necessary to propose a method for acquiring urban building unit data based on machine learning. Summary of the invention

[0007] The purpose of the present invention is to provide a method for acquiring urban building unit data based on machine learning to solve the problems raised in the above background technology.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] A method for acquiring urban building unit data based on machine learning, the method steps are as follows:

[0010] S1, the optical remote sensing camera takes pictures of urban buildings and transmits the acquired image data to the ground control station;

[0011] S2, the ground control station forwards the image data to the data processing center for storage and preliminary processing;

[0012] S3, point cloud scanner scans urban buildings, obtains three-dimensional point cloud data, and transmits it to the data processing center;

[0013] S4, 3D ground penetrating radar detects underground building structures and pipelines, and transmits data to the data processing center;

[0014] S5, the data processing center integrates and preprocesses the data from different devices to remove noise and abnormal data;

[0015] S6, inputting the preprocessed data into the deep learning system for analysis and processing;

[0016] S7. The deep learning system uses a neural network model to extract and classify data features and automatically identify various features of urban buildings;

[0017] S8. Based on the recognition results, the urban building unit data is extracted, stored and output for use by urban planning, construction and management departments.

[0018] Preferably, it also includes an optical remote sensing camera, a ground control station, a data processing center, a point cloud scanner, a three-dimensional ground penetrating radar and a deep learning system. The optical remote sensing camera, the ground control station and the data processing center constitute a remote sensing data system.

[0019] Preferably, the remote sensing data system uses a remote sensing camera to acquire high-resolution visible light or infrared band images, and the ground control station sends control instructions through a communication link, while receiving and processing remote sensing data, and performs data decompression, radiation correction, geometric correction and image enhancement through a data processing center.

[0020] Preferably, the point cloud scanner comprises a laser transmitting and receiving module, a scanning module, a control module, a positioning and navigation module, and a data processing module.

[0021] Preferably, the point cloud scanner realizes effective scanning of targets of different distances and materials through laser emission and receiving modules, and the scanning module realizes rapid scanning of target objects through a rotating platform and scanning mirror using high-precision motors, transmission machines, optical controllers and high-precision driving elements. The control module accurately controls the overall operation according to the preset scanning parameters such as range, resolution and scanning speed, and simultaneously performs preliminary data analysis and processing.

[0022] Preferably, the point cloud scanner obtains three-dimensional acceleration and angular velocity through a positioning and navigation system, obtains scanner posture information in real time, improves data registration accuracy, and can obtain absolute coordinate positions at the same time, effectively matching point cloud data with geographic coordinate systems to achieve data integration and application.

[0023] Preferably, the data processing module further processes and analyzes the acquired data, and generates a high-quality three-dimensional model library by combining it with BIM technology.

[0024] Preferably, the three-dimensional ground penetrating radar system includes a DXV1808 ground coupling antenna, a Geoscope MK IV host, an iRTK5, a photoelectric encoder, an acquisition computer, and a power supply. The DXV1808 ground coupling antenna, the Geoscope MK IV host, the iRTK5, and the photoelectric encoder are all connected to the acquisition computer, and the power supply provides power for the DXV1808 ground coupling antenna, the GeoscopeMK IV host, the iRTK5, the photoelectric encoder, and the acquisition computer respectively.

[0025] Among them, the Geoscope MK IV host has deeper detection capabilities and higher resolution while achieving high-density and high-speed data acquisition. It does not need to replace radar antennas of different frequencies when detecting at different depths like traditional pulse radars. The continuous frequency range of the DXV1808 ground-coupled antenna is 200MHz-3GHz, which is the widest-band antenna array on the market. It can collect 1.5m wide high-density three-dimensional data at a time. iRTK5 has full constellation, multi-channel, Maxwell7 technology, supports full constellation and full-band resolution, fast and accurate positioning, supports satellite station differential services, and global single-machine accurate positioning of 4 cm, making radar detection and positioning more reliable; the photoelectric encoder can use the principle of grating diffraction to realize displacement-digital conversion. Through photoelectric conversion, the mechanical geometric displacement on the output shaft is converted into a pulse digital quantity to accurately measure and control the displacement of the radar antenna.

[0026] The deep learning system adopts the OWM algorithm. The OWM algorithm can make the optimization process of the neural network finally stop at a solution that achieves good performance for both new and old tasks.

[0027] For the learning of the nth task, since the input space cannot be guaranteed to be full rank, the OWM algorithm gives the correction matrix P of the nth task OWM is the approximate projection operator of the orthogonal subspace of the input space, and the calculation formula is shown in formula (1):

[0028]

[0029] Among them, I p with I m is a p-order and m-order unit matrix, α is a sufficiently small positive number to ensure Full rank, from which we can get the calculation formula for updating the gradient using the OWM algorithm:

[0030] W n =W n-1 -λΔW n P OWM (2)

[0031] Among them, W n is the weight of the neural network after completing the nth task learning, λ is the learning rate,

[0032] ΔW n It is the gradient generated by back propagation when the neural network learns the nth task;

[0033] The α required for calculating the projection operator is obtained adaptively through the mean and variance of the input data. The calculation formula is shown in formula (3):

[0034] α=Average(xx T )+Cov(x) (3)

[0035] Among them, x represents the input data of the neural network, Average() means calculating the mean, and Cov() means calculating the variance.

[0036] Beneficial effects of the present invention:

[0037] 1. The present invention comprehensively utilizes a variety of data acquisition devices and machine learning technologies to efficiently and accurately obtain urban building unit data.

[0038] 2. The deep learning system of the present invention can automatically identify architectural features, reduce human intervention, and improve the efficiency and accuracy of data processing.

[0039] 3. The present invention provides comprehensive and accurate building data support for urban planning, construction and management, which helps to improve the level of urban management and the scientific nature of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a flow chart of the unit data acquisition method of the present invention;

[0042] Figure 2 A schematic diagram of a remote sensing data system of the present invention;

[0043] Figure 3 It is a schematic diagram of a point cloud scanner of the present invention;

[0044] Figure 4 It is a schematic diagram of optimizing the OWM algorithm of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] like Figures 1 to 4 As shown, an embodiment of the present invention provides:

[0047] A method for acquiring urban building unit data based on machine learning, the method steps are as follows:

[0048] S1, the optical remote sensing camera takes pictures of urban buildings and transmits the acquired image data to the ground control station;

[0049] S2, the ground control station forwards the image data to the data processing center for storage and preliminary processing;

[0050] S3, point cloud scanner scans urban buildings, obtains three-dimensional point cloud data, and transmits it to the data processing center;

[0051] S4, 3D ground penetrating radar detects underground building structures and pipelines, and transmits data to the data processing center;

[0052] S5, the data processing center integrates and preprocesses the data from different devices to remove noise and abnormal data;

[0053] S6, inputting the preprocessed data into the deep learning system for analysis and processing;

[0054] S7. The deep learning system uses a neural network model to extract and classify data features and automatically identify various features of urban buildings;

[0055] S8. Based on the recognition results, the urban building unit data is extracted, stored and output for use by urban planning, construction and management departments.

[0056] Furthermore, it also includes optical remote sensing cameras, ground control stations, data processing centers, point cloud scanners, three-dimensional ground penetrating radars and deep learning systems. The optical remote sensing cameras, ground control stations and data processing centers constitute the remote sensing data system.

[0057] Furthermore, the remote sensing data system uses a remote sensing camera to obtain high-resolution visible light or infrared band images, and the ground control station sends control instructions through a communication link, while receiving and processing remote sensing data, and performs data decompression, radiation correction, geometric correction and image enhancement through a data processing center.

[0058] Furthermore, the point cloud scanner includes a laser transmitting and receiving module, a scanning module, a control module, a positioning and navigation module, and a data processing module.

[0059] Furthermore, the point cloud scanner can effectively scan targets of different distances and materials through the laser emission and receiving modules, and the scanning module can quickly scan the target object through the rotating platform and scanning mirror using high-precision motors, transmission machines, optical controllers and high-precision drive components. The control module accurately controls the overall operation according to the preset scanning parameters such as range, resolution and scanning speed, and simultaneously performs preliminary data analysis and processing.

[0060] Furthermore, the point cloud scanner obtains three-dimensional acceleration and angular velocity through the positioning and navigation system, obtains scanner posture information in real time, improves data registration accuracy, and can obtain absolute coordinate position at the same time, effectively matching point cloud data with the geographic coordinate system to achieve data integration and application.

[0061] Furthermore, the data processing module further processes and analyzes the acquired data, and generates a high-quality three-dimensional model library by combining it with BIM technology.

[0062] Furthermore, the three-dimensional ground penetrating radar system includes a DXV1808 ground coupling antenna, a Geoscope MK IV host, an iRTK5, a photoelectric encoder, an acquisition computer, and a power supply. The DXV1808 ground coupling antenna, the Geoscope MK IV host, the iRTK5, and the photoelectric encoder are all connected to the acquisition computer, and the power supply provides power for the DXV1808 ground coupling antenna, the GeoscopeMK IV host, the iRTK5, the photoelectric encoder, and the acquisition computer respectively.

[0063] Among them, the Geoscope MK IV host has deeper detection capabilities and higher resolution while achieving high-density and high-speed data acquisition. It does not need to replace radar antennas of different frequencies when detecting at different depths like traditional pulse radars. The continuous frequency range of the DXV1808 ground-coupled antenna is 200MHz-3GHz, which is the widest-band antenna array on the market. It can collect 1.5m wide high-density three-dimensional data at a time. iRTK5 has full constellation, multi-channel, Maxwell7 technology, supports full constellation and full-band resolution, fast and accurate positioning, supports satellite station differential services, and global single-machine accurate positioning of 4 cm, making radar detection and positioning more reliable; the photoelectric encoder can use the principle of grating diffraction to realize displacement-digital conversion. Through photoelectric conversion, the mechanical geometric displacement on the output shaft is converted into a pulse digital quantity to accurately measure and control the displacement of the radar antenna.

[0064] The deep learning system adopts the OWM algorithm. The OWM algorithm can make the optimization process of the neural network finally stop at a solution that achieves good performance for both new and old tasks.

[0065] For the learning of the nth task, since it is impossible to guarantee that the input space is full rank, the OWM algorithm gives the correction matrix P of the nth task OWM is the approximate projection operator of the orthogonal subspace of the input space, and the calculation formula is shown in formula (1):

[0066]

[0067] Among them, I p with I m is a p-order and m-order unit matrix, α is a sufficiently small positive number to ensure Full rank, from which we can get the calculation formula for updating the gradient using the OWM algorithm:

[0068] W n =W n-1 -λΔW n P OWM (2)

[0069] Among them, Wn is the weight of the neural network after completing the nth task learning, λ is the learning rate,

[0070] ΔW n It is the gradient generated by back propagation when the neural network learns the nth task;

[0071] The α required for calculating the projection operator is obtained adaptively through the mean and variance of the input data. The calculation formula is shown in formula (3):

[0072] α=Average(xx T )+Cov(x) (3)

[0073] Among them, x represents the input data of the neural network, Average() means calculating the mean, and Cov() means calculating the variance.

[0074] The most representative one is the orthogonal weight modification algorithm (OWM algorithm). In the scenario of multiple tasks, it is impossible for the neural network to learn all the mappings at the same time. Many mapping relationships cannot be set in advance and must be determined after encountering the corresponding data. In order to prevent the previously learned mappings from being erased by subsequent training, that is, to avoid catastrophic forgetting, the OWM algorithm was proposed. Its gradient correction strategy is to calculate the projection operator of the orthogonal subspace A of the input space through the input space of the neural network as the gradient correction matrix P of the OWM algorithm. OWM Then use P OWM The gradient is projected to A so that the weights are modified only in the orthogonal direction of the input space. This ensures that the learning of new tasks does not interfere with the old tasks that have been learned, because the weight changes in the entire network will not interact with the old inputs, thereby maintaining the mapping relationship of the old tasks. Therefore, combined with the optimal weight search based on stochastic gradient descent (SGD), OWM helps the network achieve better learning results on new tasks while maintaining the performance of old tasks unchanged.

[0075] Beneficial effects of the present invention:

[0076] 1. The present invention comprehensively utilizes a variety of data acquisition devices and machine learning technologies to efficiently and accurately obtain urban building unit data.

[0077] 2. The deep learning system can automatically identify building features, reduce human intervention, and improve the efficiency and accuracy of data processing.

[0078] 3. It provides comprehensive and accurate building data support for urban planning, construction and management, which helps to improve the level of urban management and the scientific nature of decision-making.

[0079] Working principle:

[0080] The optical remote sensing camera takes pictures of urban buildings and transmits the acquired image data to the ground control station; the ground control station forwards the image data to the data processing center for storage and preliminary processing; the point cloud scanner scans urban buildings, obtains three-dimensional point cloud data, and transmits it to the data processing center; the three-dimensional ground penetrating radar detects underground building structures and pipelines, and transmits the data to the data processing center; the data processing center integrates and preprocesses data from different devices to remove noise and abnormal data; the preprocessed data is input into the deep learning system for analysis and processing; the deep learning system uses a neural network model to extract and classify data features and automatically identify various features of urban buildings; based on the recognition results, the urban building unit data is extracted, stored and output for use by urban planning, construction and management departments.

[0081] 1. Optical remote sensing camera: used to obtain image information of urban buildings from high altitudes, with the characteristics of high resolution and wide coverage.

[0082] 2. Ground control station: Control and transmit data to optical remote sensing cameras and other equipment to achieve remote operation and monitoring.

[0083] 3. Data processing center: receives data from various devices and performs preliminary processing and storage.

[0084] 4. Point cloud scanner: The three-dimensional point cloud data of urban buildings is obtained through laser scanning technology, which can accurately reflect the geometric shape of the building.

[0085] 5. 3D ground penetrating radar: used to detect underground building structures, pipelines and other information, providing important data for a comprehensive understanding of urban buildings.

[0086] The deep learning system based on the neural network model comprehensively analyzes and processes data from optical remote sensing cameras, point cloud scanners and 3D ground penetrating radars. Through a large amount of training data, the system can automatically identify the characteristics of urban buildings, such as building type, height, area, etc., and extract useful building unit data.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for acquiring urban building unit data based on machine learning, characterized in that: The method steps are: S1, the optical remote sensing camera takes pictures of urban buildings and transmits the acquired image data to the ground control station; S2, the ground control station forwards the image data to the data processing center for storage and preliminary processing; S3, point cloud scanner scans urban buildings, obtains three-dimensional point cloud data, and transmits it to the data processing center; S4, 3D ground penetrating radar detects underground building structures and pipelines, and transmits data to the data processing center; S5, the data processing center integrates and preprocesses the data from different devices to remove noise and abnormal data; S6, inputting the preprocessed data into the deep learning system for analysis and processing; S7. The deep learning system uses a neural network model to extract and classify data features and automatically identify various features of urban buildings; S8. Based on the recognition results, the urban building unit data is extracted, stored and output for use by urban planning, construction and management departments.

2. The method for acquiring urban building unit data based on machine learning according to claim 1, characterized in that: It also includes optical remote sensing cameras, ground control stations, data processing centers, point cloud scanners, three-dimensional ground penetrating radars and deep learning systems, among which the optical remote sensing cameras, ground control stations and data processing centers constitute the remote sensing data system.

3. The method for acquiring urban building unit data based on machine learning as claimed in claim 2, characterized in that: The remote sensing data system uses remote sensing cameras to obtain high-resolution visible light or infrared band images. The ground control station sends control instructions through communication links, while receiving and processing remote sensing data, and performs data decompression, radiation correction, geometric correction and image enhancement through the data processing center.

4. The method for acquiring urban building unit data based on machine learning according to claim 1, characterized in that: The point cloud scanner comprises a laser transmitting and receiving module, a scanning module, a control module, a positioning and navigation module, and a data processing module.

5. The method for acquiring urban building unit data based on machine learning as claimed in claim 4, characterized in that: The point cloud scanner realizes effective scanning of targets of different distances and materials through laser emission and receiving modules. The scanning module realizes rapid scanning of target objects through a rotating platform and a scanning mirror using a high-precision motor, a transmission machine, an optical controller and high-precision driving elements. The control module accurately controls the overall operation according to the preset scanning parameters such as the range, resolution and scanning speed, and simultaneously performs preliminary data analysis and processing.

6. The method for acquiring urban building unit data based on machine learning as claimed in claim 4, characterized in that: The point cloud scanner obtains three-dimensional acceleration and angular velocity through the positioning and navigation system, obtains scanner posture information in real time, improves data registration quarter, and can obtain absolute coordinate position at the same time, effectively matches point cloud data with the geographic coordinate system, and realizes data integration and application.

7. The method for acquiring urban building unit data based on machine learning according to claim 4, characterized in that: The data processing module further processes and analyzes the acquired data and generates a high-quality three-dimensional model library by combining it with BIM technology.

8. The method for acquiring urban building unit data based on machine learning according to claim 1, characterized in that: The three-dimensional ground penetrating radar system includes a DXV1808 ground coupling antenna, a Geoscope MKIV host, an iRTK5, a photoelectric encoder, an acquisition computer, and a power supply. The DXV1808 ground coupling antenna, the Geoscope MK IV host, the iRTK5, and the photoelectric encoder are all connected to the acquisition computer, and the power supply provides power for the DXV1808 ground coupling antenna, the Geoscope MK IV host, the iRTK5, the photoelectric encoder, and the acquisition computer respectively.

9. The method for acquiring urban building unit data based on machine learning according to claim 1, characterized in that: The Geoscope MK IV host has deeper detection capabilities and higher resolution while achieving high-density and high-speed data acquisition. The DXV1808 ground-coupled antenna has a continuous frequency range of 200MHz-3GHz and can collect 1.5m wide high-density three-dimensional data in a single shot; The photoelectric encoder can use the principle of grating diffraction to realize digital conversion of displacement. Through photoelectric conversion, the mechanical geometric displacement on the output shaft is converted into a pulse digital quantity, thereby accurately measuring and controlling the displacement of the radar antenna.

10. The method for acquiring urban building unit data based on machine learning according to claim 1, characterized in that: The deep learning system adopts the OWM algorithm. The OWM algorithm can make the optimization process of the neural network finally stop at a solution that achieves good performance for both new and old tasks. For the learning of the nth task, since the input space cannot be guaranteed to be full rank, the OWM algorithm gives the correction matrix P of the nth task OWM is the approximate projection operator of the orthogonal subspace of the input space, and the calculation formula is shown in formula (1): Among them, I p with I m is a p-order and m-order unit matrix, α is a sufficiently small positive number to ensure Full rank, from which we can get the calculation formula for updating the gradient using the OWM algorithm: W n =W n-1 -λΔW n P OWM (2) Among them, W n is the weight of the neural network after learning the nth task, λ is the learning rate, ΔW n It is the gradient generated by back propagation when the neural network learns the nth task; The α required for calculating the projection operator is obtained adaptively through the mean and variance of the input data. The calculation formula is shown in formula (3): α=Average(xx T )+Cov(x) (3) Among them, x represents the input data of the neural network, Average(·) means calculating the mean, and Cov(·) means calculating the variance.

Images