Land space planning data acquisition method and system based on big data

Abstract

The invention relates to the field of data transmission processing, in particular to a land space planning data acquisition method and system based on big data, and the method comprises the steps: obtaining a foreground region and a background region of a ground photographic image, and carrying out the splicing of the foreground region and the background region; dividing a smooth area in the area of the splicing part, calculating the smoothness of the smooth area, and responding to the condition that the smoothness is greater than a preset threshold value, determining that the terrestrial photographic image is successfully acquired; wherein the smoothness calculation method comprises the following steps: taking an intersection of a smooth region and a foreground region as a first region, taking an intersection of the smooth region and a background region as a second region, and dividing the first region and the second region into a plurality of grids; and calculating the similarity between the grids, and taking the mean value of the similarity as the smoothness of the smooth region. The method has the effect of monitoring whether ground camera image data acquisition is abnormal or not.

Description

Technical Field

[0001] The present application relates to the field of data acquisition and transmission, and in particular to a method and system for acquiring land space planning data based on big data. Background Art

[0002] National land space planning is an arrangement made in space and time for the development and protection of national land space in a certain area. It is a guide for national space development, a spatial blueprint for sustainable development, and the basic basis for various development, protection and construction activities. National land space planning data collection is an important foundation for national land space planning work. National land space planning data collection covers natural resources (such as topography, water resources, vegetation coverage, etc.), social economy (population, transportation, public service facilities, etc.), spatiotemporal big data (Internet spatiotemporal data, mobile phone signaling data, Internet map data, etc.) and urban operation data (infrastructure, urban safety, etc.). These data are collected and integrated through various technical means such as satellite remote sensing, drone aerial photography, geographic information system (GIS), and Internet data mining, providing comprehensive and dynamic data support for national land space planning.

[0003] Image data plays an important role in the data collection process of national land space planning. Drone aerial images provide high-resolution small-scale ground images, which are suitable for high-precision surveys of urban blocks and ecological protection areas; ground photography images record the current status of specific areas, such as buildings and public service facilities. When using drones to collect images, it is sometimes necessary to transmit the collected images back to the ground control station or other equipment, so that the collected images can be viewed in real time and the flight path or mission parameters can be adjusted in time.

[0004] During the transmission of ground photographic image data, low transmission efficiency may occur when the image is large. In this case, the foreground area and the background area in the image can be transmitted separately, and then spliced ​​or fused to reduce the demand for transmission bandwidth. However, the splicing effect may be affected by many factors, such as network fluctuations, signal interference, equipment failure, etc. These factors may cause abnormalities in the data after transmission. For example, ground photographic image data may lose some key information during transmission, or the quality of ground photographic images may be reduced due to improper use of compression algorithms, or even data may be tampered with or damaged. These anomalies will not only affect the availability of ground photographic image data, but may also lead to errors in subsequent analysis and decision-making. Summary of the invention

[0005] In order to solve the above-mentioned technical problem that the collected data is prone to anomalies, the present application provides a land space planning data collection method and system based on big data.

[0006] In the first aspect, the present application provides a method for collecting land space planning data based on big data, which adopts the following technical solutions: A land space planning data collection method based on big data comprises the steps of: obtaining a foreground area and a background area of ​​a ground photographic image and splicing them; dividing a smooth area in the splicing area, calculating the smoothness of the smooth area, and responding that the smoothness is greater than a preset threshold, the ground photographic image is successfully collected; wherein the smoothness is calculated by: taking the intersection of the smooth area and the foreground area as the first area, taking the intersection of the smooth area and the background area as the second area, and dividing the first area and the second area into a plurality of grids; calculating the similarity between the grids, and taking the mean of the similarity as the smoothness of the smooth area; the similarity calculation formula is: ; In the formula, Representation Grid With Grid The similarity of the first area grid and the grid of the second region Adjacent, Representation Grid The mean value of all pixel values ​​in , Representation Grid The mean value of all pixel values ​​in , Representation Grid and Grid The covariance of pixel values, Representation Grid The variance of the pixel values ​​in , Representation Grid The variance of the pixel values ​​in , is the first constant, is the second constant.

[0007] The beneficial effects are: the image is divided into the foreground area and the background area for transmission, and the images are spliced ​​at the receiving end. The splicing effect is evaluated by calculating the smoothness to ensure the natural transition of the ground photography image. The success of the ground photography image transmission is judged by the preset threshold to ensure that the transmission quality meets the application requirements. A real-time feedback mechanism is provided to quickly detect anomalies during the transmission process to ensure the reliability of the transmission.

[0008] The smoothness is quantified by calculating the mean of the similarity between the grids. The similarity calculation takes into account the mean, variance, and covariance information, and can fully reflect the quality of the stitching area. The mean parameter quantifies the brightness or color consistency between the grids. If the means of two grids are close, it means that their overall brightness or color is similar. Otherwise, the overall brightness or color is quite different. The covariance parameter quantifies the structural similarity between the grids. If the covariance of the two grids is large, it means that their pixel value change trends are similar. Otherwise, the pixel value change trends are very different.

[0009] Optionally, the expression for the first constant is ; The expression of the second constant is: ;in, is the first constant, is the second constant, Indicates the dynamic range of pixel values; represents the first adjustment coefficient, Represents the second adjustment coefficient.

[0010] The beneficial effects are: by introducing constants, the robustness of similarity calculation is enhanced, and the problem of denominator being zero or numerical instability is avoided. The value of the constant is dynamically adjusted according to the dynamic range of pixel values ​​to adapt to different ground photography image characteristics and application requirements.

[0011] Optionally, the dynamic range of pixel values ​​is .

[0012] Optionally, obtaining the foreground area and the background area of ​​the ground photographic image includes: obtaining the collected ground photographic image and dividing the foreground area and the background area; using different 5G slice networks to transmit the foreground area and the background area of ​​the ground photographic image respectively.

[0013] The beneficial effect is: by allocating the foreground area and the background area to different 5G slice networks for transmission, the high bandwidth and low latency characteristics of the 5G network are fully utilized, and network resources are dynamically allocated according to the importance of the content of the ground photographic image, avoiding the waste of network resources.

[0014] Optionally, the method of dividing the foreground area and the background area is: setting a target detection model, inputting the ground photographic image into the target detection model for target detection, and if there is a target, dividing the area where the target is located into the foreground area, and the rest into the background area.

[0015] The beneficial effects are: the foreground area and background area are automatically divided through the target detection model, which reduces human intervention and improves the intelligence level of ground photography image processing.

[0016] Optionally, different 5G slice networks are used to transmit the foreground area and background area of ​​the ground photographic image respectively, including: slicing the 5G network to obtain a first slice network and a second slice network, the transmission priority of the first slice network is higher than that of the second slice network; using the first slice network to transmit the foreground area of ​​the ground photographic image, and the second slice network to transmit the background area of ​​the ground photographic image.

[0017] The beneficial effects are: dynamically allocating network slice resources according to the target detection results, ensuring that important areas (foreground areas) can be transmitted first, thereby improving transmission efficiency.

[0018] Optionally, a first slicing network is used to transmit a foreground area of ​​a terrestrial photographic image, and a second slicing network is used to transmit a background area of ​​the terrestrial photographic image, including: inputting the foreground area into a preset autoencoding network to obtain a first coding vector; similarly, obtaining a second coding vector for the background area; transmitting the first coding vector through the first slicing network, and transmitting the second coding vector through the second slicing network; obtaining the first coding vector and the second coding vector, and decoding them using a decoder in the autoencoding network to obtain the foreground area and the background area.

[0019] The beneficial effects are: it provides a coding and decoding method, and the self-encoding network encodes the foreground and background areas into low-dimensional coding vectors respectively. Compared with directly transmitting the original image data, the amount of data is greatly reduced. In this way, during the transmission process, the network bandwidth occupancy can be effectively reduced and the transmission speed can be improved. It is especially suitable for the collection and transmission scenarios of large-scale national land space planning data.

[0020] Optionally, the loss function for autoencoder network training is: ;in, represents the loss function, represents the loss weight of the smooth area, represents the mean square error loss in the smooth region, represents the loss weight of the foreground area, represents the mean square error loss of the foreground area, represents the loss weight of the background area, represents the mean square error loss of the background area.

[0021] The beneficial effect is: through the multi-weighted loss function, the fidelity of the reconstructed ground photography image in important areas (foreground) and smooth areas is ensured. The loss weights can be adjusted according to actual needs to adapt to different application scenarios.

[0022] Optionally, the object detection model is deployed on the edge node.

[0023] In the second aspect, the present application provides a national land space planning data collection system based on big data, which adopts the following technical solutions: A land space planning data collection system based on big data comprises: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the land space planning data collection method based on big data is implemented.

[0024] The beneficial effect is: the above-mentioned land space planning data collection method based on big data is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a system based on the memory and the processor for easy use.

[0025] This application has the following technical effects: 1. The image is divided into foreground area and background area for transmission, and spliced ​​at the receiving end. The splicing effect is evaluated by calculating the smoothness to ensure the natural transition of the ground photography image. The success of the ground photography image transmission is judged by the preset threshold to ensure that the transmission quality meets the application requirements. A real-time feedback mechanism is provided to quickly detect anomalies during the transmission process to ensure the reliability of the transmission.

[0026] 2. Quantify the smoothness by calculating the mean of the similarity between grids. The similarity calculation takes into account the mean, variance and covariance information, and can fully reflect the quality of the spliced ​​area. The mean parameter quantifies the brightness or color consistency between grids. If the means of two grids are close, it means that their overall brightness or color is similar. Otherwise, the overall brightness or color is quite different. The covariance parameter quantifies the structural similarity between grids. If the covariance of two grids is large, it means that their pixel value change trends are similar. Otherwise, the pixel value change trends are very different. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a method flow chart of a method for collecting land space planning data based on big data in an embodiment of the present application.

[0028] Figure 2 It is a method flow chart of step S1 in a method for collecting land space planning data based on big data in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The present application embodiment discloses a method for collecting land space planning data based on big data, referring to Figure 1 , comprising steps S1-S2: S1: Obtain the foreground area and background area of ​​the ground photography image and stitch them together.

[0030] Reference Figure 2, obtaining the foreground area and the background area of ​​the ground photographic image includes steps S10 and S11, which are specifically as follows: S10: Acquire the collected ground photographic image and divide it into a foreground area and a background area.

[0031] The method of dividing the foreground area and the background area is: setting up a target detection model, inputting the ground photography image into the target detection model for target detection, and if there is a target, the area where the target is located is divided into the foreground area, and the rest is divided into the background area.

[0032] Specifically, ground photographic images are taken by drones and other equipment, and the target areas in the ground photographic images are marked, and the types of the target areas are marked. The target detection dataset is constructed based on the ground photographic images and the annotation results corresponding to the ground photographic images.

[0033] The target detection model is trained using the target detection dataset, and the model is deployed on the edge node after training. The training of the target detection model is an existing technology and will not be described in detail.

[0034] In one embodiment, an existing target detection model such as YOLO (You Only Look Once) that can be applied in the application scenario of the present application can be selected, and a labeling tool (such as LabelImg or VIA, etc.) can be used to label the target area in the ground photography image. The prior art will not be described in detail.

[0035] When labeling, you need to label the bounding box of the target and the category of the target (such as vehicles, pedestrians, traffic signs, etc.). After labeling is completed, the ground photography image and the corresponding annotation file constitute the target detection dataset. Use the target detection dataset to train the selected target detection model. During the training process, the model will learn how to identify the target area from the ground photography image and output the category and bounding box of the target.

[0036] In the process of real-time ground photography image transmission, the captured ground photography image is first transmitted to the edge node, input into the target detection model deployed in the edge node, and outputs the target area in the ground photography image, i.e., the foreground and background. The trained target detection model is deployed on the edge node. Edge nodes usually have lower computing resources and power consumption, and are suitable for reasoning in real-time scenarios.

[0037] During the real-time ground photography image transmission process, the captured ground photography images will be transmitted to the edge node. The target detection model on the edge node will infer the input ground photography images and output the target area (foreground) and background in the ground photography images. The foreground ground photography images and background ground photography images of the ground photography images are labeled to facilitate the subsequent transmission to the server and combine the foreground ground photography images and background ground photography images into one ground photography image.

[0038] S11: Use different 5G slice networks to transmit the foreground area and background area of ​​the ground photography image respectively.

[0039] The 5G network is sliced ​​to obtain a first slice network and a second slice network, and the transmission priority of the first slice network is higher than that of the second slice network; the first slice network is used to transmit the foreground area of ​​the ground photography image, and the second slice network is used to transmit the background area of ​​the ground photography image.

[0040] The foreground area is input into a preset autoencoder network to obtain a first encoding vector; similarly, a second encoding vector of the background area is obtained; the first encoding vector is transmitted through a first slice network, and the second encoding vector is transmitted through a second slice network; the first encoding vector and the second encoding vector are obtained, and the foreground area and the background area are obtained after decoding using a decoder in the autoencoder network.

[0041] The autoencoder network is an unsupervised learning neural network used for data compression and reconstruction. The autoencoder network consists of two parts: an encoder and a decoder. The encoder can compress the input data (such as ground photography images) into a low-dimensional encoding vector; the decoder can reconstruct the encoding vector into the original data. The encoder is deployed at the edge node and the decoder is deployed at the server center. The foreground ground photography image is transmitted through the network channel with a high priority after slicing, and the background ground photography image is transmitted through the network channel with a low priority after slicing. After being transmitted to the edge node, the ground photography image is encoded and processed, and the encoded ground photography image data is transmitted to the server center. After decoding, the foreground ground photography image and the background ground photography image are combined into the original ground photography image, and the foreground and background are judged to be correctly matched according to the splicing effect of the foreground and background.

[0042] The training process of the autoencoder network mainly includes the following steps: Input foreground ground photography image or background ground photography image. Encoding, the input ground photography image is compressed into a low-dimensional feature vector through the encoding network. Decoding, the encoded feature vector is reconstructed into the output ground photography image through the decoding network. Loss calculation, the difference between the input ground photography image and the output ground photography image is calculated and used as the loss function. During the training process, the goal of the model is to minimize the difference between the input ground photography image and the output ground photography image, so that the reconstructed ground photography image is as close to the original ground photography image as possible.

[0043] Due to the different importance of foreground ground photography images and background ground photography images, their loss functions need to be weighted when training the model: the foreground ground photography images contain important target information (such as buildings, etc.), so their loss weight is larger to ensure that the foreground area of ​​the reconstructed ground photography images has high fidelity. The background ground photography images contain less information, so their loss weight is smaller. By adjusting the loss weights of the foreground ground photography images and the background ground photography images, the model can pay more attention to the reconstruction quality of important areas.

[0044] The loss function for autoencoder network training is: ;in, represents the loss function, represents the loss weight of the smooth area, represents the mean square error loss in the smooth region, represents the loss weight of the foreground area, represents the mean square error loss of the foreground area, represents the loss weight of the background area, represents the mean square error loss of the background area.

[0045] Used to measure the difference between the reconstructed ground photographic image and the original ground photographic image. Used to measure the difference between the foreground area in the reconstructed ground photography image and the foreground area in the original ground photography image. Used to measure the difference between the background area in the reconstructed ground photography image and the background area in the original ground photography image. Used to measure the difference between the smooth area in the reconstructed ground photography image and the smooth area in the original ground photography image.

[0046] By adjusting the weights, you can control the degree to which the autoencoder network focuses on different areas during training. For example, if the foreground area is more important, you can increase , so that the network pays more attention to the reconstruction quality of the foreground area. If the importance of the background area is low, you can reduce , reducing the impact of the background area on the total loss. For example, , , .

[0047] S2: Divide the area at the joint into smooth areas, calculate the smoothness of the smooth areas, and in response to the smoothness being greater than a preset threshold, the ground photography image is successfully acquired.

[0048] The foreground area and the background area are stitched together. After the stitching is completed, a transition area is divided at the stitching point as a smooth area. Part of the smooth area is located in the foreground area, and the other part is located in the background area. The purpose is to make the stitched ground photography image transition naturally and avoid obvious boundaries or discontinuities. The smoothness of the smooth area is calculated, and the stitching effect of the foreground area and the background area is judged based on the smoothness. Smoothness is used to quantify the quality of the stitching area. The higher the smoothness, the better the stitching effect; the lower the smoothness, the worse the stitching effect.

[0049] The calculation method of smoothness is as follows: the intersection of the smooth area and the foreground area is taken as the first area, the intersection of the smooth area and the background area is taken as the second area, the first area and the second area are divided into multiple grids; the similarity between the grids is calculated, and the mean of the similarity is taken as the smoothness of the smooth area. The greater the similarity, the smoother it is, and the better the splicing effect of the foreground area and the background area is. Conversely, the splicing effect of the foreground area and the background area is worse, and the mean of the similarity is taken as the smoothness of the smooth area.

[0050] The similarity calculation formula is: ; In the formula, Representation Grid With Grid The similarity of the first area grid and the grid of the second region Adjacent, Representation Grid The mean value of all pixel values ​​in , Representation Grid The mean value of all pixel values ​​in , Representation Grid and Grid The covariance of pixel values, Representation Grid The variance of the pixel values ​​in , Representation Grid The variance of the pixel values ​​in , is the first constant, is the second constant.

[0051] Among them, the mean parameter quantifies the brightness or color consistency between grids. If the means of two grids are close, it means that their overall brightness or color is similar. Otherwise, the overall brightness or color is quite different. The covariance parameter quantifies the structural similarity between grids. If the covariance of two grids is large, it means that their pixel value change trends are similar. Otherwise, the pixel value change trends are very different. The mean and variance are normalized by the denominator to ensure that the similarity value range is between 0 and 1.

[0052] In order to enhance the robustness of similarity calculation, the expression of the first constant is ; The expression of the second constant is: ;in, Represents the dynamic range of pixel values. For example, for an 8-bit image, the dynamic range of pixel values ​​is .

[0053] represents the first adjustment coefficient, represents the second adjustment coefficient. For example, , , can be adjusted according to the actual application scenario and The value of .

[0054] An embodiment of the present application also discloses a land space planning data collection system based on big data, including a processor and a memory, the memory storing computer program instructions, and when the computer program instructions are executed by the processor, a land space planning data collection method based on big data according to the present application is implemented.

[0055] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0056] In the present application, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory, a dynamic random access memory, a static random access memory, etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device.

[0057] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for collecting land space planning data based on big data, characterized in that: The method comprises the following steps: obtaining a foreground area and a background area of ​​a ground photographic image and splicing them; dividing the spliced ​​area into a smooth area, calculating the smoothness of the smooth area, and in response to the smoothness being greater than a preset threshold, the ground photographic image is successfully acquired; The smoothness is calculated by taking the intersection of the smooth area and the foreground area as the first area, taking the intersection of the smooth area and the background area as the second area, and dividing the first area and the second area into a plurality of grids; calculating the similarity between the grids, and taking the mean of the similarities as the smoothness of the smooth area; The similarity calculation formula is: ; In the formula, Representation Grid With Grid The similarity of the first area grid and the grid of the second region Adjacent, Representation Grid The mean value of all pixel values ​​in , Representation Grid The mean value of all pixel values ​​in , Representation Grid and Grid The covariance of pixel values, Representation Grid The variance of the pixel values ​​in , Representation Grid The variance of the pixel values ​​in , is the first constant, is the second constant.

2. The method for collecting data for national land space planning based on big data according to claim 1 is characterized in that: The expression of the first constant is ; The expression of the second constant is: ;in, is the first constant, is the second constant, Indicates the dynamic range of pixel values; represents the first adjustment coefficient, Represents the second adjustment coefficient.

3. The method for collecting data for national land space planning based on big data according to claim 2 is characterized in that: The dynamic range of pixel values ​​is .

4. The method for collecting data for national land space planning based on big data according to claim 1 is characterized in that: Obtaining a foreground area and a background area of ​​a ground photographic image, including: obtaining a collected ground photographic image and dividing the foreground area and the background area; and using different 5G slice networks to transmit the foreground area and the background area of ​​the ground photographic image respectively.

5. The method for collecting land space planning data based on big data according to claim 4 is characterized in that: The method of dividing the foreground area and the background area is: setting up a target detection model, inputting the ground photography image into the target detection model for target detection, and if there is a target, the area where the target is located is divided into the foreground area, and the rest is divided into the background area.

6. The method for collecting data for national land space planning based on big data according to claim 4 is characterized in that: Use different 5G network slices to transmit the foreground and background areas of ground photography images separately, including: The 5G network is sliced ​​to obtain a first slice network and a second slice network, and the transmission priority of the first slice network is higher than that of the second slice network; the first slice network is used to transmit the foreground area of ​​the ground photography image, and the second slice network is used to transmit the background area of ​​the ground photography image.

7. The method for collecting land space planning data based on big data according to claim 6 is characterized in that: The foreground area of ​​the ground photographic image is transmitted using the first slicing network, and the background area of ​​the ground photographic image is transmitted using the second slicing network, including: Input the foreground area into a preset autoencoder network to obtain a first encoding vector; similarly, obtain a second encoding vector for the background area; the first encoding vector is transmitted through a first slice network, and the second encoding vector is transmitted through a second slice network; The first encoding vector and the second encoding vector are obtained, and the foreground area and the background area are obtained after decoding using the decoder in the autoencoding network.

8. The method for collecting land space planning data based on big data according to claim 7 is characterized in that: The loss function for autoencoder network training is: ;in, represents the loss function, represents the loss weight of the smooth area, represents the mean square error loss in the smooth region, represents the loss weight of the foreground area, represents the mean square error loss of the foreground area, represents the loss weight of the background area, represents the mean square error loss of the background area.

9. The method for collecting data for national land space planning based on big data according to claim 5 is characterized in that: The object detection model is deployed on the edge nodes.

10. A national land space planning data collection system based on big data, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a land space planning data collection method based on big data is implemented according to any one of claims 1 to 9.

Images