Building change detection method and apparatus, electronic device, and storage medium

Abstract

The application relates to a building change detection method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a first remote sensing image of a to-be-detected area; performing image processing on the first remote sensing image to generate a second remote sensing image; identifying a building in the second remote sensing image based on a pre-trained network model to generate an identification result, wherein the identification result comprises a third remote sensing image; determining a building contour in the third remote sensing image, converting image coordinates of the building contour into geographical coordinates; generating first shp data based on the geographical coordinates of the building contour; acquiring second shp data matched with the first shp data in a database; comparing the first shp data and the second shp data to generate a matching result; and the matching result is used to determine whether the building has changed. The application improves the detection efficiency and accuracy of building changes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-precision maps, and in particular to a building change detection method and device, an electronic device, and a storage medium. BACKGROUND

[0002] Due to rapid economic development, buildings, as the main activity places of humans, are also constantly iterated and updated. In order to improve the accuracy of high-precision maps, the buildings in the high-precision maps also need to be updated regularly, and consequently building change detection is faced.

[0003] In related technologies, building change detection is performed manually. The manual detection method needs to consume a long period of time, and for dense urban buildings, it is very easy to miss, resulting in slow detection efficiency and low accuracy. SUMMARY

[0004] To overcome the problems in related technologies, the present application provides a building change detection method and device, an electronic device, and a storage medium to improve the detection efficiency and accuracy of building changes.

[0005] The first aspect of the present application provides a building change detection method, comprising:

[0006] obtaining a first remote sensing image of a region to be detected;

[0007] performing image processing on the first remote sensing image to generate a second remote sensing image;

[0008] identifying a building in the second remote sensing image based on a pre-trained network model to generate an identification result, the identification result comprising a third remote sensing image;

[0009] determining a building contour in the third remote sensing image, and converting an image coordinate of the building contour into a geographic coordinate;

[0010] generating first shp data based on the geographic coordinate of the building contour;

[0011] obtaining second shp data in a database that matches the first shp data;

[0012] comparing the first shp data and the second shp data to generate a matching result; wherein the matching result is used to determine whether the building has changed.

[0013] As a possible implementation manner of the present application, in this implementation manner, the image processing on the first remote sensing image to generate a second remote sensing image comprises:

[0014] perform cloud removal on the first remote sensing image to generate a second remote sensing image.

[0015] As a possible implementation of the present application, in this implementation, the performing cloud removal on the first remote sensing image to generate a second remote sensing image comprises:

[0016] determining a ground reflectance of each pixel in the first remote sensing image based on a pre-established physical model;

[0017] determining a target pixel in the ground reflectance interval of the cloud reflectance;

[0018] replacing a pixel value of the target pixel with the ground reflectance corresponding to the target pixel to generate a second remote sensing image.

[0019] As a possible implementation of the present application, in this implementation, the method further comprises:

[0020] performing atmospheric correction on the first remote sensing image;

[0021] Correspondingly, the determining a ground reflectance of each pixel in the first remote sensing image based on a pre-established physical model comprises:

[0022] determining a ground reflectance of each pixel in the first remote sensing image after atmospheric correction based on a pre-established physical model.

[0023] As a possible implementation of the present application, in this implementation, before the determining the building outline in the third remote sensing image and converting the image coordinates of the building outline into geographic coordinates, the method further comprises:

[0024] filtering out the building represented by the first color in the third remote sensing image with an area less than a specified area threshold.

[0025] As a possible implementation of the present application, in this implementation, the first remote sensing image comprises at least two periods of first remote sensing images; each period corresponds to a third remote sensing image, so that the number of third remote sensing images is a first number; the identification result further comprises a confidence of the building represented by the first color.

[0026] After the filtering out the building represented by the first color in the third remote sensing image with an area less than a specified area threshold, the method further comprises:

[0027] determining a same pixel position identified as a building in a second number of third remote sensing images; wherein the second number is greater than or equal to 2 and less than or equal to the first number;

[0028] determine a total confidence that the same pixel position is identified as a building based on a confidence of the third remote sensing image at the same pixel position of the building of the first quantity;

[0029] identify a building at the same pixel position with the total confidence greater than a specified confidence threshold as a target building;

[0030] Correspondingly, the determining the building contour in the third remote sensing image comprises:

[0031] determining the building contour of the target building in the third remote sensing image.

[0032] As a possible implementation of the present application, in this implementation, the obtaining the second shp data matched with the first shp data in the database comprises:

[0033] calculating a similarity of the first shp data and the shp data corresponding to each building in the database;

[0034] determining the shp data in the database with the highest similarity to the first shp data and the highest similarity greater than a specified similarity threshold as the second shp data matched with the first shp data.

[0035] As a possible implementation of the present application, in this implementation,

[0036] The comparing the first shp data and the second shp data to generate a matching result comprises:

[0037] comparing the first shp data and the second shp data by using the Hungarian algorithm to generate a matching result.

[0038] As a possible implementation of the present application, in this implementation, the comparing the first shp data and the second shp data by using the Hungarian algorithm to generate a matching result comprises:

[0039] converting the first shp data into a first bipartite graph, converting the second shp data into a second bipartite graph, and determining edge weight information of the first bipartite graph and the second bipartite graph; wherein the first shp data comprises shp data of at least one building surface;

[0040] performing maximum weight matching by using the Hungarian algorithm based on the first bipartite graph, the second bipartite graph, and respective node information and the edge weight information to generate a matching result;

[0041] The matching result includes at least one of at least one matching pair, an unmatched building surface, and an overlapped building surface. One building surface in the matching pair is from the first shp data, and the other building surface is from the second shp data. The unmatched building surface includes building surfaces in the first shp data and the second shp data that are not matched as the matching pair. The overlapped building surface includes building surfaces that overlap in the first shp data and the second shp data.

[0042] As a possible implementation of the present application, in this implementation, further comprising:

[0043] Merging the matching pair based on the first shp data and the second shp data to generate a target shp file.

[0044] As a possible implementation of the present application, in this implementation, the network model corresponds to at least two update frequency ranges during the training process, and different update frequency ranges correspond to different learning rates.

[0045] And / or, a preset weight decay coefficient is added to the loss function in the network model.

[0046] And / or, a BN layer is set after each convolution layer of the network model.

[0047] And / or, the network layer number of the deep residual network module in the network model is set to a specified number of layers.

[0048] And / or, the activation function in the network model is set to a Mish activation function.

[0049] And / or, the loss function in the network model is set to a cross-entropy loss function and an IoU loss function.

[0050] The second aspect of the present application provides a building change detection device, comprising:

[0051] An image acquisition unit is configured to acquire a first remote sensing image of a region to be detected.

[0052] An image processing unit is configured to perform image processing on the first remote sensing image to generate a second remote sensing image.

[0053] A model recognition unit is configured to recognize buildings in the second remote sensing image based on a pre-trained network model to generate a recognition result, wherein the recognition result includes a third remote sensing image.

[0054] A coordinate conversion unit is configured to determine the outline of a building in the third remote sensing image and convert the image coordinates of the building outline into geographic coordinates.

[0055] generating a first shp data based on geographic coordinates of the building contour;

[0056] matching a second shp data in the database with the first shp data;

[0057] generating a matching result by comparing the first shp data and the second shp data, wherein the matching result is used to determine whether the building has changed.

[0058] The third aspect of the present application provides an electronic device, comprising:

[0059] a processor; and

[0060] a memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method as described above.

[0061] The fourth aspect of the present application provides a non-transitory machine-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method as described above.

[0062] The technical solution provided by the present application can include the following beneficial effects:

[0063] By using the pre-trained network model to identify the building in the remote sensing image, the building contour is determined, the image coordinates of the building contour are converted into geographic coordinates, the first shp data is generated based on the geographic coordinates of the building contour, the second shp data in the database that matches the first shp data is obtained, the first shp data and the second shp data are compared, and the matching result is generated. The matching result can be used to determine whether the building has changed. Therefore, the present application can compare the shp data generated based on the identified building with the shp data in the database to determine whether the building in the detection area has changed, thereby improving the detection efficiency and accuracy of the building change.

[0064] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0065] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings, in which like reference characters refer to like parts throughout the figures, and wherein:

[0066] Figure 1is a flowchart of a building change detection method according to an embodiment of the present application;

[0067] Figure 2 is a flowchart of a second remote sensing image generation method according to an embodiment of the present application;

[0068] Figure 3 is a flowchart of a shp data matching method according to an embodiment of the present application;

[0069] Figure 4 is a flowchart of a matching result generation method according to an embodiment of the present application;

[0070] Figure 5 is a structural diagram of a building change detection device according to an embodiment of the present application;

[0071] Figure 6 is a structural diagram of an image processing unit according to an embodiment of the present application;

[0072] Figure 7 is a structural diagram of a matching data unit according to an embodiment of the present application;

[0073] Figure 8 is a structural diagram of a generation result unit according to an embodiment of the present application;

[0074] Figure 9 is a structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0075] The preferred embodiments of the present application will be described herein below with reference to the accompanying drawings. While the preferred embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0076] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0077] It should be understood that although the terms "first", "second", "third", etc. can be used in this application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information without departing from the scope of the application. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0078] The technical solutions of the embodiments of the application are described in detail below with reference to the drawings.

[0079] Figure 1 is a flowchart of the building change detection method shown in the embodiments of the application, referring to Figure 1 , the method can include the following steps:

[0080] Step 101: acquiring a first remote sensing image of a to-be-detected area;

[0081] The first remote sensing image is an image of the to-be-detected area, which can be a top view of the to-be-detected area. Generally, the first remote sensing image of the to-be-detected area can be acquired by a collection device. In order to improve the accuracy of image collection, the first remote sensing image of the to-be-detected area can be acquired by a satellite map such as Goole Earth, a commercial map, or by satellite shooting.

[0082] Step 102: image processing the first remote sensing image to generate a second remote sensing image;

[0083] In order to improve the accuracy of building change detection, the first remote sensing image needs to be image processed, which can be referred to as denoising processing. As a possible embodiment of the application, the image processing of the first remote sensing image to generate a second remote sensing image can include:

[0084] The first remote sensing image is cloud-removed to generate a second remote sensing image.

[0085] Cloud removal refers to removing the cloud layer in the first remote sensing image.

[0086] Specifically, as shown in Figure 2 , the cloud removal of the first remote sensing image to generate a second remote sensing image can include the following steps:

[0087] Step 201: determining the ground reflectance of each pixel in the first remote sensing image based on a pre-established physical model;

[0088] The physical model in the present application can be various models capable of determining the surface reflectivity. For example, the physical model can be an atmospheric radiation transfer model, such as a MODTRAN (moderate spectral resolution atmospheric transmittance) model.

[0089] Step 202: determining a target pixel of the surface reflectivity in the cloud reflectivity interval;

[0090] By judging whether the surface reflectivity of the pixel is in the cloud reflectivity interval, it can be determined whether the pixel is a cloud. The pixel that is a cloud is determined as a target pixel.

[0091] In the implementation process, the pixel value of the surface reflectivity in the cloud reflectivity interval in the first remote sensing image can be marked as 1, and the pixel value not in the cloud reflectivity interval can be marked as 0. That is, the pixel value of 1 represents that the pixel is a cloud, and the pixel value of 0 represents that the pixel is not a cloud.

[0092] Step 203: replacing the pixel value of the target pixel with the surface reflectivity corresponding to the target pixel to generate a second remote sensing image.

[0093] By replacing the pixel value, the cloud reconstruction is realized, and the surface information in the area blocked by the cloud can be restored, which is convenient for detecting the building change.

[0094] In order to eliminate the influence of atmospheric shielding and scattering on the surface reflectivity to obtain accurate surface reflectivity, in some embodiments of the present application, before step 201, the method can further include:

[0095] Atmospheric correction is performed on the first remote sensing image.

[0096] Correspondingly, determining the surface reflectivity of each pixel in the first remote sensing image based on the pre-established physical model includes:

[0097] Determining the surface reflectivity of each pixel in the first remote sensing image after atmospheric correction based on the pre-established physical model.

[0098] Step 103: identifying the building in the second remote sensing image based on the pre-trained network model to generate an identification result;

[0099] The network model in the present application is a neural network learning model, which can be a Mask R-CNN network model.

[0100] The identification result includes a third remote sensing image.

[0101] The third remote sensing image can distinguish between buildings and non-buildings, and in some embodiments, the third remote sensing image has a first color and a second color different from the first color, the first color representing buildings and the second color representing non-buildings.

[0102] The first color can be white and the second color can be black, or the first color can be black and the second color can be white.

[0103] Step 104: Determine the building outline in the third remote sensing image, and convert the image coordinates of the building outline to geographic coordinates.

[0104] In this application, the building outline of the first color in the third remote sensing image can be determined by an image processing algorithm. As a specific example, the building outline of the first color in the third remote sensing image can be determined by the cv2 library, such as the cv2.findContours() function in the cv2 library for obtaining the building outline.

[0105] The pyproj (Python interface to PROJ library) library can convert image coordinates to geographic coordinates, which can be latitude and longitude coordinates.

[0106] The pyproj library is a Python interface library for the PROJ.4 library, which can be used for geospatial coordinate conversion and projection. It provides the functions of the Python version of the PROJ.4 library, which can convert geographic coordinate systems and projection coordinate systems, supports a large number of geospatial coordinate systems and projection methods, and provides a Python version of the interface, which can facilitate the use of the functions of the PROJ.4 library for geospatial coordinate conversion and projection.

[0107] Of course, this application is not limited to using the pyproj library to convert coordinates, and other ways to convert image coordinates to geographic coordinates are within the scope of this application.

[0108] Step 105: Generate first shp data based on the geographic coordinates of the building outline.

[0109] In some embodiments, the QGIS (Quantum GIS) plug-in can be used to generate shp (Shape) data based on the geographic coordinates of the building. The shp format data is an open standard in the field of geographic information software.

[0110] Step 106: Obtain second shp data matching the first shp data in the database.

[0111] The database stores shapefile data of historical buildings. By comparing the first shapefile data with the shapefile data in the database, a second shapefile data that matches the first shapefile data can be determined.

[0112] In one approach, shapefile data can be manually compared to determine the second shapefile data that matches the first shapefile data.

[0113] In another way, shapefile data can be compared automatically, such as... Figure 3 As shown, obtaining the second shapefile data that matches the first shapefile data from the database may include the following process:

[0114] Step 301: Calculate the similarity between the first shapefile data and the shapefile data corresponding to each building in the database;

[0115] The shapefile data includes at least one of the following: location attribute data, shape attribute data, area attribute data, etc. Matching is determined by calculating the similarity between the attributes of the first shapefile data and the attributes of the shapefile data of each building in the database. Specifically, similarity measurement algorithms, such as cosine similarity algorithm and Euclidean distance algorithm, can be used to calculate the similarity between the first shapefile data and the shapefile data corresponding to each building in the database.

[0116] Step 302: Determine the second shapefile data that matches the first shapefile data as having the highest similarity to the first shapefile data in the database, and whose highest similarity is greater than a specified similarity threshold.

[0117] Identify the shapefile in the database that has the highest similarity to the first shapefile and determine whether the highest similarity is greater than a specified similarity threshold. If so, identify the shapefile that has the highest similarity to the first shapefile as the second shapefile that matches the first shapefile. If not, determine that the match has failed and there is no shapefile in the database that matches the first shapefile.

[0118] If a match fails, a second shapefile that matches the first shapefile can be manually retrieved from the database to correct the match. If a matching second shapefile cannot be found manually either, the first shapefile can be added to the database as the shapefile for the new building.

[0119] Step 107: Compare the first shapefile data and the second shapefile data to generate a matching result.

[0120] The matching result is used for determining whether the building has changed.

[0121] In a possible implementation of the present application, the comparison of the first shp data and the second shp data to generate a matching result comprises:

[0122] The first shp data and the second shp data are compared by using a Hungarian algorithm to generate a matching result.

[0123] Specifically, as shown in Figure 4 The comparison of the first shp data and the second shp data by using a Hungarian algorithm to generate a matching result can include the following steps:

[0124] Step 401: converting the first shp data into a first bipartite graph and the second shp data into a second bipartite graph, and determining edge weight information of the first bipartite graph and the second bipartite graph;

[0125] The first shp data comprises shp data of at least one building surface, and the second shp data comprises shp data of at least one building surface. The building surface is used for representing a surface of a building. If the shp data comprises shp data of at least two building surfaces, the at least two building surfaces are surfaces of different buildings in the same projection plane, such as surfaces of different buildings in a top view.

[0126] The edge weight information is used for representing a similarity between one building surface in the first shp data and one building surface in the second shp data. In the process of converting the first shp data into a first bipartite graph and the second shp data into a second bipartite graph, the similarity between each two building surfaces can be calculated, and the similarity is taken as the edge weight information between the two building surfaces. It can be understood that one building surface in each two building surfaces is from the first shp data, and the other building surface is from the second shp data.

[0127] Step 402: performing maximum weight matching by using a Hungarian algorithm based on the first bipartite graph, the second bipartite graph, the respective node information, and the edge weight information to generate a matching result.

[0128] The first bipartite graph, the second bipartite graph, the node information of the first bipartite graph, the node information of the second bipartite graph, and the edge weight information are taken as input parameters, and the Hungarian algorithm is used to perform maximum weight matching to obtain the matching result.

[0129] The matching result comprises at least one of at least one matching pair, an unmatched building surface, and an overlapping building surface.

[0130] The matching pair is used to represent the best combination of matching, one building surface in the matching pair is from the first shp data, and the other building surface is from the second shp data.

[0131] The unmatched building surface includes the building surface in the first shp data and the second shp data that is not matched as the matching pair.

[0132] The overlapping building surface includes the building surface that has overlap in the first shp data and the second shp data.

[0133] The above-mentioned building surface matched as the matching pair, the unmatched building surface, and the overlapping building surface can be marked differently in the first shp data and the second shp data, so as to distinguish different building surfaces.

[0134] Based on the above-mentioned matching result, whether the building in the to-be-detected area changes can be intuitively reflected.

[0135] In some embodiments, the method further comprises:

[0136] Based on the first shp data and the second shp data, the matching pair is merged to generate a target shp file.

[0137] In the present application, the building in the remote sensing image is recognized through the pre-trained network model, and then the building contour is determined, the image coordinates of the building contour are converted into geographic coordinates, the first shp data is generated based on the geographic coordinates of the building contour, the second shp data matching the first shp data in the database is obtained, the first shp data and the second shp data are compared, and the matching result is generated. The matching result can be used to determine whether the building changes; it can be seen that the present application can compare the shp data generated based on the recognized building with the shp data in the database to determine whether the building in the to-be-detected area changes, thereby improving the detection efficiency and accuracy of the building change.

[0138] In some embodiments of the present application, before the building contour in the third remote sensing image is determined, and the image coordinates of the building contour are converted into geographic coordinates, the method further comprises:

[0139] The building represented by the first color in the third remote sensing image with an area less than a specified area threshold is filtered out.

[0140] The network model may misidentify some features that are not buildings as buildings when identifying the second remote sensing image to generate the third remote sensing image. Therefore, to solve this problem, the application can filter out the buildings represented by the first color with an area less than a specified area threshold in the third remote sensing image before converting the coordinates, thereby improving the accuracy of building change detection.

[0141] The specified area threshold can be set based on actual conditions, which is not limited by the application.

[0142] In some embodiments of the application, to further improve the accuracy of building change detection, the first remote sensing image of the to-be-detected region in different periods can be obtained. That is, the first remote sensing image includes at least two first remote sensing images of different periods. Correspondingly, each period corresponds to a first remote sensing image, and one first remote sensing image corresponds to one third remote sensing image, so that the number of third remote sensing images is a first number; the number of third remote sensing images is consistent with the number of first remote sensing images.

[0143] The identification result further includes a confidence of the building represented by the first color.

[0144] Correspondingly, after filtering out the buildings represented by the first color with an area less than a specified area threshold in the third remote sensing image, the method can further include the following steps:

[0145] (1.1) determining the same pixel position identified as a building in the second number of third remote sensing images;

[0146] Wherein, the second number is greater than or equal to 2, and less than or equal to the first number.

[0147] It can be understood that the same pixel position refers to a position that is the same in the second number of third remote sensing images and is identified as a building.

[0148] (1.2) determining the total confidence of the same pixel position identified as a building based on the confidence of the building at the same pixel position in the first number of third remote sensing images;

[0149] The total confidence can be determined by the following formula.

[0150]

[0151] Wherein, n is the number of acquisition periods of the first remote sensing image, score is the confidence of the building at the same pixel position in each third remote sensing image of the first number, and confidence is the total confidence.

[0152] (1.3) The buildings identified at the same pixel location with a total confidence score greater than the specified confidence score threshold are identified as target buildings.

[0153] This application pre-sets a confidence threshold, thereby identifying buildings at the same pixel location with a total confidence score greater than the confidence threshold as target buildings.

[0154] Accordingly, determining the building outlines in the third remote sensing image includes:

[0155] The building outline of the target building in the third remote sensing image is determined.

[0156] In this application, the accuracy of building change detection can be further improved by using area screening and confidence level screening.

[0157] In some embodiments of this application, in order to improve the convergence speed of the network model, the network model in this application can be an optimized network model. Specifically:

[0158] The network model corresponds to at least two update count ranges during training, and different update count ranges correspond to different learning rates.

[0159] And / or, a preset weight decay coefficient is added to the loss function in the network model;

[0160] And / or, a BN layer is set after each convolutional layer of the network model;

[0161] And / or, set the number of network layers of the deep residual network module in the network model to a specified number of layers;

[0162] And / or, set the activation function in the network model to the Mish activation function;

[0163] And / or, set the loss function in the network model to the cross-entropy loss function and the IoU loss function.

[0164] In the above network model optimization, the number of times the network model is updated during training is epoch. This application sets different learning rates for different epoch ranges. Specifically, the learning rates are set from large to small as the update number range is sorted from small to large.

[0165] For example, the learning rate is set to 0.01 for the first 30 epochs; 0.001 for epochs 30-60; and 0.0001 for epochs after 60. By setting different learning rates, the training speed and accuracy of the model can be better balanced.

[0166] In some embodiments, the complexity of the network model can be controlled by adding a weight decay coefficient to the loss function to avoid overfitting. For example, the weight decay coefficient can be set to 0.001, and of course the coefficient can be adjusted based on the training results. The addition of the weight decay coefficient can be implemented through the optimizer SGD (Stochastic Gradient Descent).

[0167] In some embodiments, a BN (Batch Normalization) layer is correspondingly arranged after each convolutional layer of the network model, which can accelerate the training process and improve the generalization ability of the model.

[0168] In some embodiments, the number of network layers of the ResNet (Residual Network) module in the network model can be set to a specified number; for example, ResNet101 in the original network model is replaced by ResNet50 to reduce the number of network layers of ResNet.

[0169] Corresponding to the foregoing application function implementation method embodiments, the present application also provides a building change detection device, an electronic device and corresponding embodiments.

[0170] Figure 5 is a structural schematic diagram of the building change detection device shown in the embodiments of the present application. Referring to Figure 5 The device can include an image acquisition unit 110, an image processing unit 120, a model recognition unit 130, a coordinate conversion unit 140, a generated data unit 150, a matching data unit 160, and a generated result unit 170; wherein:

[0171] The image acquisition unit 110 is configured to acquire a first remote sensing image of a detection area;

[0172] The image processing unit 120 is configured to perform image processing on the first remote sensing image to generate a second remote sensing image;

[0173] The model recognition unit 130 is configured to recognize buildings in the second remote sensing image based on a pre-trained network model to generate a recognition result, wherein the recognition result includes a third remote sensing image;

[0174] The third remote sensing image can distinguish between buildings and non-buildings. In some embodiments, the third remote sensing image has a first color and a second color different from the first color, wherein the first color represents buildings and the second color represents non-buildings.

[0175] The first color can be white and the second color can be black, or the first color can be black and the second color can be white.

[0176] The coordinate conversion unit 140 is configured to determine a building contour in the third remote sensing image, and convert image coordinates of the building contour into geographic coordinates.

[0177] The generation data unit 150 is configured to generate first shp data based on the geographic coordinates of the building contour.

[0178] The matching data unit 160 is configured to acquire second shp data matched with the first shp data in a database.

[0179] The generation result unit 170 is configured to compare the first shp data and the second shp data, and generate a matching result.

[0180] The matching result is used to determine whether the building has changed.

[0181] In some embodiments, the image processing unit performs image processing on the first remote sensing image to generate a second remote sensing image, specifically, performs cloud removal processing on the first remote sensing image to generate a second remote sensing image.

[0182] Specifically, as shown in Figure 6 The image processing unit 120 can include a first determination module 1201, a second determination module 1202, and a first generation module 1203.

[0183] The first determination module 1201 is configured to determine the ground reflectivity of each pixel in the first remote sensing image based on a pre-established physical model.

[0184] The second determination module 1202 is configured to determine a target pixel in which the ground reflectivity is in a cloud layer reflectivity interval.

[0185] The first generation module 1203 is configured to replace a pixel value of the target pixel with the ground reflectivity corresponding to the target pixel to generate a second remote sensing image.

[0186] In order to eliminate the influence of atmospheric obstruction and scattering on the ground reflectivity to obtain accurate ground reflectivity, in some embodiments of the present application, the image processing unit can further include:

[0187] The atmospheric correction unit is configured to perform atmospheric correction on the first remote sensing image.

[0188] Correspondingly, the first determination module is specifically configured to determine the ground reflectivity of each pixel in the first remote sensing image after atmospheric correction based on a pre-established physical model.

[0189] In some embodiments, as shown in Figure 7 The matching data unit 160 can include a similarity calculation module 1601 and a third determination module 1602.

[0190] The similarity calculating module 1601 is configured to calculate the similarity between the first shp data and the shp data corresponding to each building in the database.

[0191] The third determining module 1602 is configured to determine the shp data in the database with the highest similarity to the first shp data, and the highest similarity is greater than a specified similarity threshold value, as the second shp data matched with the first shp data.

[0192] In some embodiments, the result generating unit is specifically configured to utilize the Hungarian algorithm to compare the first shp data and the second shp data, and generate the matching result.

[0193] In some embodiments, as shown in Figure 8 The result generating unit 170 can include a conversion determining module 1701 and a maximum weight matching module 1702; wherein:

[0194] The conversion determining module 1701 is configured to convert the first shp data into a first bipartite graph, convert the second shp data into a second bipartite graph, and determine the edge weight information of the first bipartite graph and the second bipartite graph; wherein the first shp data includes shp data of at least one building surface.

[0195] The maximum weight matching module 1702 is configured to utilize the Hungarian algorithm to perform maximum weight matching based on the first bipartite graph, the second bipartite graph, the respective node information, and the edge weight information, and generate a matching result.

[0196] The matching result includes at least one of at least one matching pair, an unmatched building surface, and an overlapping building surface, one building surface in the matching pair comes from the first shp data, and the other building surface comes from the second shp data; the unmatched building surface includes building surfaces in the first shp data and the second shp data that are not matched as the matching pair; and the overlapping building surface includes building surfaces that overlap in the first shp data and the second shp data.

[0197] In some embodiments, the apparatus can further include:

[0198] The merging generating unit is configured to merge the matching pair based on the first shp data and the second shp data, and generate a target shp file.

[0199] In some embodiments, the apparatus further includes a first filtering unit configured to filter out the buildings represented by the first color in the third remote sensing image with an area less than a specified area threshold value.

[0200] In some embodiments, the first remote sensing image comprises at least two time periods of first remote sensing images; each time period corresponds to one of the third remote sensing images, so that the number of the third remote sensing images is a first number; the identification result further comprises a confidence of the building represented by the first color;

[0201] The apparatus further comprises:

[0202] A position determination unit is configured to determine, after the first filtering unit filters out the building represented by the first color in the third remote sensing image with an area less than a specified area threshold, a same pixel position identified as a building in a second number of the third remote sensing images;

[0203] The second number is greater than or equal to 2 and less than or equal to the first number.

[0204] A total confidence determination unit is configured to determine a total confidence of the same pixel position identified as a building based on the confidence of the building in the first number of the third remote sensing images at the same pixel position;

[0205] A target building determination unit is configured to determine the building identified by the same pixel position with a total confidence greater than a specified confidence threshold as a target building.

[0206] Correspondingly, the coordinate conversion unit determines the building contour of the building in the third remote sensing image, comprising: determining the building contour of the target building in the third remote sensing image.

[0207] In some embodiments, the network model corresponds to at least two ranges of update times during the training process, and different ranges of update times correspond to different learning rates.

[0208] And / or, a preset weight decay coefficient is added to the loss function in the network model.

[0209] And / or, a BN layer is set after each convolution layer of the network model.

[0210] And / or, the network layer number of the deep residual network module in the network model is set to a specified number of layers.

[0211] And / or, the activation function in the network model is set to a Mish activation function.

[0212] And / or, the loss function in the network model is set to a cross-entropy loss function and an IoU loss function.

[0213] As to the apparatus in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments of the method, and will not be described in detail here.

[0214] Figure 9 FIG. 1 is a structural schematic diagram of an electronic device according to an embodiment of the present application.

[0215] Referring to FIG. 1, Figure 9 The electronic device 1000 includes a memory 1010 and a processor 1020.

[0216] The processor 1020 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can be any conventional processor.

[0217] The memory 1010 can include various types of storage units, such as a system memory, a read-only memory (ROM), and a permanent storage device. Among them, the ROM can store static data or instructions required by the processor 1020 or other modules of the computer. The permanent storage device can be a read-write storage device. The permanent storage device can be a non-volatile storage device that does not lose stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a mass storage device (such as a magnetic or optical disk, a flash memory) as a permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, an optical drive). The system memory can be a read-write storage device or a volatile read-write storage device, such as a dynamic random access memory. The system memory can store some or all instructions and data required by the processor during runtime. In addition, the memory 1010 can include a combination of any computer readable storage media, including various types of semiconductor storage chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic disks and / or optical disks. In some embodiments, the memory 1010 can include a read and / or write removable storage device, such as a compact disc (CD), a read-only digital versatile disc (such as DVD-ROM, double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a min SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer readable storage medium does not include a carrier wave and a transient electronic signal transmitted through a wireless or wired transmission.

[0218] The executable code stored on the memory 1010 can cause the processor 1020 to perform part or all of the above-mentioned methods when the executable code is processed by the processor 1020.

[0219] The solutions of the present application have been described in detail above with reference to the accompanying drawings. In the above-described embodiments, the description of each embodiment is focused on respectively, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments. It should also be known to those skilled in the art that the actions and modules involved in the specification are not necessarily required by the present application. In addition, it can be understood that the steps in the method embodiments of the present application can be adjusted in sequence, combined and reduced according to actual needs, and the modules in the device embodiments of the present application can be combined, divided and reduced according to actual needs.

[0220] In addition, the method according to the present application can also be implemented as a computer program or computer program product, which includes computer program code instructions for executing part or all of the steps in the above-mentioned method of the present application.

[0221] Alternatively, the present application can also be implemented as a non-transitory machine readable storage medium (or computer readable storage medium, or machine readable storage medium) having executable code (or computer program, or computer instruction code) stored thereon, which, when executed by a processor of an electronic device (or electronic device, server, etc.), causes the processor to perform part or all of the steps of the above-mentioned method according to the present application.

[0222] Those skilled in the art will also appreciate that the various example logical blocks, modules, circuits, and algorithm steps described in connection with the present application herein can be implemented as electronic hardware, computer software, or a combination of the two.

[0223] The computer software can be coded using any suitable machine code or computer language that can be subject to well- defined and repetitive processing. Note that the computer software can be written in an interpreted language or can be written in a compiled language that is converted into a computer independent intermediate representation code or object code suitable for usage on a variety of computers (i.e., different machines) or that is converted into a proprietary

[0224] Embodiments of the present application have been described above, with the understanding that these embodiments are exemplary only, and not exhaustive, and are not limited to the embodiments disclosed. Many modifications and variations of the described embodiments are possible, without departing from the scope and spirit of the described embodiments. The selection of terms to be used in the description is intended to best explain the principles of the embodiments, practical application, or improvement over the technology in the art, or to enable other ordinary skilled in the art to understand the embodiments disclosed herein.

Claims

1. Method for detecting changes in a building, characterized in that, The method comprises the following steps: obtaining a first remote sensing image of a region to be detected; performing image processing on the first remote sensing image to generate a second remote sensing image; identifying buildings in the second remote sensing image based on a pre-trained network model to generate an identification result, the identification result comprising a third remote sensing image; determining the building outlines in the third remote sensing image and converting the image coordinates of the building outlines into geographic coordinates; generating first shp data based on the geographic coordinates of the building outlines; obtaining second shp data in a database that matches the first shp data; comparing the first shp data and the second shp data to generate a matching result; wherein the matching result is used to determine whether the buildings have changed, comprising: converting the first shp data into a first bipartite graph, converting the second shp data into a second bipartite graph, determining the edge weight information of the first bipartite graph and the second bipartite graph, the first shp data comprising shp data of at least one building surface; performing maximum weight matching using the Hungarian algorithm based on the first bipartite graph, the second bipartite graph, and the respective node information and edge weight information to generate a matching result, the matching result comprising at least one of a matching pair, an unmatched building surface, and an overlapping building surface, one building surface in the matching pair coming from the first shp data and the other building surface coming from the second shp data; the unmatched building surface comprising building surfaces in the first shp data and the second shp data that are not matched as the matching pair; the overlapping building surface comprising building surfaces that overlap in the first shp data and the second shp data.

2. The method of claim 1, wherein, The method further comprises the following steps: performing cloud removal processing on the first remote sensing image to generate a second remote sensing image.

3. The method of claim 2, wherein, The method further comprises the following steps: determining the surface reflectance of each pixel in the first remote sensing image based on a pre-established physical model; determining a target pixel whose surface reflectance is in the cloud reflectance interval; replacing the pixel value of the target pixel with the surface reflectance corresponding to the target pixel to generate a second remote sensing image.

4. The method of claim 3, wherein, The method further comprises the following steps: performing atmospheric correction on the first remote sensing image; Correspondingly, the method further comprises the following steps: determining the surface reflectance of each pixel in the first remote sensing image after atmospheric correction based on a pre-established physical model.

5. The method of claim 1, wherein, The method further comprises the following steps before the step of determining the building outlines in the third remote sensing image and converting the image coordinates of the building outlines into geographic coordinates: filtering out buildings represented by a first color in the third remote sensing image whose area is less than a specified area threshold.

6. The method of claim 5, wherein, The first remote sensing image comprises at least two first remote sensing images of different time periods; each time period corresponds to a third remote sensing image, so that the number of third remote sensing images is a first number; the identification result further comprises the confidence of the buildings represented by the first color. The filtering out of the third remote sensing image, the area of the first color represented by the building is less than the specified area threshold, further comprises: Determine a second number of the same pixel positions in the third remote sensing image identified as a building; wherein the second number is greater than or equal to 2, and less than or equal to the first number; Determine the total confidence of the same pixel position identified as a building based on the confidence of the first number of buildings in the third remote sensing image at the same pixel position; Identify the building in the same pixel position with a total confidence greater than a specified confidence threshold as a target building; Correspondingly, the determination of the building contour in the third remote sensing image comprises: Determine the building contour of the target building in the third remote sensing image.

7. The method of claim 1, wherein, The acquisition of the second shp data matching the first shp data in the database comprises: Calculate the similarity of the first shp data and the shp data corresponding to each building in the database; Determine the shp data in the database with the highest similarity to the first shp data, and the highest similarity is greater than a specified similarity threshold, as the second shp data matching the first shp data.

8. The method of claim 1, wherein, Further comprising: Merge the matching pair based on the first shp data and the second shp data to generate a target shp file.

9. The method of claim 1, wherein, The network model corresponds to at least two update frequency ranges during the training process, and different update frequency ranges correspond to different learning rates; And / or, a preset weight decay coefficient is added to the loss function in the network model; And / or, a BN layer is set after each convolutional layer of the network model; And / or, the network layer number of the deep residual network module in the network model is set to a specified number; And / or, the activation function in the network model is set to the Mish activation function; And / or, the loss function in the network model is set to the cross-entropy loss function and the IoU loss function.

10. A device for detecting changes in a building, characterized in that It comprises: An image acquisition unit for acquiring a first remote sensing image of a region to be detected; An image processing unit for image processing the first remote sensing image to generate a second remote sensing image; A model recognition unit for identifying buildings in the second remote sensing image based on a pre-trained network model to generate an identification result, the identification result comprising a third remote sensing image; A coordinate conversion unit for determining a building contour in the third remote sensing image and converting the image coordinates of the building contour into geographic coordinates; A data generation unit for generating first shp data based on the geographic coordinates of the building contour; A matching data unit for acquiring second shp data matching the first shp data in the database; The generation result unit is configured to compare the first shp data and the second shp data to generate a matching result, wherein the matching result is used to determine whether the building has changed, including: converting the first shp data into a first bipartite graph, converting the second shp data into a second bipartite graph, and determining edge weight information of the first bipartite graph and the second bipartite graph, wherein the first shp data includes shp data of at least one building surface; performing maximum weight matching based on the first bipartite graph, the second bipartite graph, and respective node information and the edge weight information by using a Hungarian algorithm to generate a matching result, wherein the matching result includes at least one of at least one matching pair, an unmatched building surface, and an overlapping building surface, one building surface in the matching pair is from the first shp data, and the other building surface is from the second shp data; the unmatched building surface includes building surfaces that are not matched into the matching pair in the first shp data and the second shp data; and the overlapping building surface includes building surfaces that overlap in the first shp data and the second shp data.

11. An electronic device, comprising: Comprise: a processor; and a memory having stored thereon executable code that, when executed by the processor, causes the processor to perform the method of any one of claims 1-9. 12.A non-transitory machine-readable storage medium having stored thereon executable code that, when executed by a processor of an electronic device, causes the processor to perform the method of any one of claims 1-9.

Images