Building height detection method and device, storage medium and electronic equipment

Abstract

The application discloses a building height detection method and device, a storage medium and an electronic device, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height; inputting the target remote sensing image into a target detection model for detection processing, and outputting height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model by using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used for representing height information of the building, and M is a positive integer. Through the application, the problem that the effect of detecting the height of a building is poor in the related art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a building height detection method and device, a storage medium and an electronic device. BACKGROUND

[0002] In the city building detection scene and the loan risk control scene of financial institutions, obtaining the height information of the building is an important goal. In the related art, the algorithm for detecting the height of the building mainly has the following problems:

[0003] (1) The cost of obtaining the height of the object in the remote sensing image by using a hardware device is high and time-consuming, and the collection technology of such data has high requirements, and the accuracy of the result is highly dependent on the professional background of the operator and the instrument.

[0004] (2) It is easier, faster and less costly to infer these data from remote sensing RGB images. However, predicting the height value from a single remote sensing image is limited because it is an ill-posed problem, and there can be countless heights for the same image.

[0005] (3) The current detection model of the height of the building mainly detects the building and the shadow, and then inversely calculates the height of the building by using the geometric relationship model of the building, the shadow, the sun and the satellite. This method not only needs a post-processing step, but also the accuracy of the shadow contour directly affects the final calculation result.

[0006] In view of the problem that the effect of detecting the height of the building in the related art is poor, an effective solution has not been proposed yet. SUMMARY

[0007] The main purpose of the present application is to provide a building height detection method and device, a storage medium and an electronic device to solve the problem that the effect of detecting the height of the building in the related art is poor.

[0008] In order to achieve the above purpose, according to one aspect of the present application, a building height detection method is provided. The method comprises: acquiring a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height; inputting the target remote sensing image into a target detection model for detection processing, and outputting the height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent the height information of the building, and M is a positive integer.

[0009] Further, the target detection model is obtained by: obtaining the M remote sensing images, the M shadow maps and the M target elevation maps; obtaining a data set according to the M remote sensing images and the M shadow maps; obtaining a training set for model training based on the data set; taking the M target elevation maps as sample labels; learning and training the target neural network model by using the training set and the sample labels to obtain the target detection model.

[0010] Further, learning and training the target neural network model by using the training set and the sample labels to obtain the target detection model includes: determining a loss function according to the training set and the sample labels; learning and training the target neural network model by using the loss function to obtain the target detection model.

[0011] Further, learning and training the target neural network model by using the loss function to obtain the target detection model includes: determining initial parameters of the target neural network model and determining an iteration number of learning and training the target neural network model; updating the initial parameters by using the loss function and obtaining a target number of updating the initial parameters by using the loss function; obtaining target parameters obtained by updating the initial parameters by using the loss function when the target number is the same as the iteration number; replacing the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0012] Further, obtaining the M shadow maps includes: obtaining the M remote sensing images; performing enhancement processing on the contrast of each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set includes at least M remote sensing images after enhancement processing; converting the remote sensing image set into a gray image set, wherein the gray image set includes at least M gray images; determining a threshold set according to the M gray images, wherein the threshold set includes at least M separation thresholds; performing segmentation processing on the M gray images by using the threshold set to obtain a shadow map set; and obtaining the M shadow maps from the shadow map set.

[0013] Further, the M target height maps are obtained by determining M target areas, wherein the target areas are areas where the buildings are located; obtaining a radar image set according to the M target areas, wherein the radar image set includes at least M radar images; determining M first height maps and M second height maps based on the radar image set, wherein the first height maps include at least elevation information of the ground and height information of target objects, the target objects are objects on the ground, and the second height maps include at least the elevation information of the ground; obtaining a height map set according to the M first height maps and the M second height maps; and obtaining the M target height maps from the height map set.

[0014] Further, before learning and training the target neural network model by using the training set and the sample labels to obtain the target detection model, the method further includes: determining a first sub-module of a first neural network model, wherein the first sub-module is at least one of an encoder module and a decoder module; and replacing the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0015] Further, the training set for model training is obtained based on the data set by performing patch processing on the data set to obtain a patch-processed data set, and obtaining the training set from the patch-processed data set according to a preset ratio.

[0016] Further, after inputting the target remote sensing image into the target detection model for detection processing and outputting the height information of the target building, the method further includes: determining a construction progress of the target building according to the height information of the target building; determining whether the construction progress reaches a target progress; if the construction progress reaches the target progress, indicating that a target object does not have a repayment risk, wherein the target object is an object that applies for a loan from a financial institution, and the loan is used for building the target building; and if the construction progress does not reach the target progress, indicating that the target object has the repayment risk.

[0017] In order to achieve the above object, according to another aspect of the present application, a building height detection device is provided. The device comprises: a first acquisition unit configured to acquire a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height; and a first processing unit configured to input the target remote sensing image into a target detection model for detection processing, and output height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent height information of the building, and M is a positive integer.

[0018] Further, the target detection model is obtained by: a second acquisition unit configured to acquire the M remote sensing images, the M shadow maps and the M target elevation maps; a first determination unit configured to obtain a data set according to the M remote sensing images and the M shadow maps; a second determination unit configured to obtain a training set for model training based on the data set; a third determination unit configured to take the M target elevation maps as sample labels; and a first training unit configured to learn and train the target neural network model using the training set and the sample labels to obtain the target detection model.

[0019] Further, the first training unit comprises: a first determination module configured to determine a loss function according to the training set and the sample labels; and a first training module configured to learn and train the target neural network model using the loss function to obtain the target detection model.

[0020] Further, the first training module comprises: a first determination sub-module configured to determine initial parameters of the target neural network model and determine an iteration number of learning and training the target neural network model; a first processing sub-module configured to update the initial parameters using the loss function, and acquire a target number of times of updating the initial parameters using the loss function; a first acquisition sub-module configured to acquire target parameters obtained by updating the initial parameters using the loss function when the target number of times is the same as the iteration number; and a first replacement sub-module configured to replace the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0021] Further, the second obtaining unit comprises: a first obtaining module, configured to obtain the M remote sensing images; a first processing module, configured to perform enhancement processing on the contrast of each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set comprises at least M remote sensing images after the enhancement processing; a first conversion module, configured to convert the remote sensing image set into a gray image set, wherein the gray image set comprises at least M gray images; a second determining module, configured to determine a threshold set according to the M gray images, wherein the threshold set comprises at least M separation thresholds; a second processing module, configured to perform segmentation processing on the M gray images by using the threshold set to obtain a shadow map set; and a second obtaining module, configured to obtain the M shadow maps from the shadow map set.

[0022] Further, the second obtaining unit comprises: a third determining module, configured to determine M target regions, wherein the target regions are regions where the buildings are located; a fourth determining module, configured to obtain a radar image set according to the M target regions, wherein the radar image set comprises at least M radar images; a fifth determining module, configured to determine M first elevation maps and M second elevation maps based on the radar image set, wherein the first elevation map comprises at least the height information of a target object and the elevation information of the ground, the target object is an object on the ground, and the second elevation map comprises at least the elevation information of the ground; a sixth determining module, configured to obtain an elevation map set according to the M first elevation maps and the M second elevation maps; and a third obtaining module, configured to obtain the M target elevation maps from the elevation map set.

[0023] Further, the device further comprises: a fourth determining unit, configured to determine a first sub-module of a first neural network model before learning and training the target neural network model by using the training set and the sample label to obtain the target detection model, wherein the first sub-module is at least one of the following: an encoder module, a decoder module; and a first replacing unit, configured to replace the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0024] Further, the second determining unit comprises: a third processing module, configured to perform block processing on the data set to obtain a block-processed data set; and a fourth obtaining module, configured to obtain the training set from the block-processed data set according to a preset proportion.

[0025] Further, the device further comprises: a fifth determination unit, configured to, after inputting the target remote sensing image into a target detection model for detection processing and outputting height information of the target building, determine construction progress of the target building according to the height information of the target building; a first judgment unit, configured to judge whether the construction progress reaches a target progress; a sixth determination unit, configured to, if the construction progress reaches the target progress, indicate that a target object does not have a repayment risk, wherein the target object is an object applying for a loan from a financial institution, and the loan is used for building the target building; and a seventh determination unit, configured to, if the construction progress does not reach the target progress, indicate that the target object has the repayment risk.

[0026] In order to achieve the above object, according to another aspect of the present application, a computer readable storage medium is provided, the storage medium storing a program, wherein the program performs any one of the above building height detection methods.

[0027] In order to achieve the above object, according to another aspect of the present application, an electronic device is provided, the electronic device comprising one or more processors and a memory, the memory being configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above building height detection methods.

[0028] By the present application, the following steps are adopted: obtaining a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected for height; inputting the target remote sensing image into a target detection model for detection processing and outputting height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent height information of the building, and M is a positive integer, thereby solving the problem of poor effect of detecting building height in related technologies. By obtaining the target remote sensing image corresponding to the target building and inputting the target remote sensing image into the target detection model for detection processing, the height information of the target building is directly outputted, thereby improving the effect of detecting building height. BRIEF DESCRIPTION OF DRAWINGS

[0029] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and are used to interpret the illustrative embodiments of the present application and their descriptions, and shall not constitute improper limitation to the present application. In the drawings:

[0030] Figure 1is a flowchart of a method for detecting a building height according to an embodiment of the present application;

[0031] Figure 2 is a schematic diagram of a U-Net network structure with residual connection introduced in an embodiment of the present application;

[0032] Figure 3 is a flowchart of an optional method for detecting a building height according to an embodiment of the present application;

[0033] Figure 4 is a schematic diagram of a device for detecting a building height according to an embodiment of the present application;

[0034] Figure 5 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0038] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0039] For ease of description, the following describes some terms or terms related to the embodiments of the present application:

[0040] Remote sensing image refers to the image of the earth's surface obtained by remote sensing technology such as aviation or satellite, including images of different frequency bands such as visible light, infrared and microwave.

[0041] Radar image is the visualization data formed by the reflected signals received in the radar system.

[0042] DSM is the Chinese name of digital surface model, and DTM is the Chinese name of digital terrain model.

[0043] The full name of GIS in Chinese is geographic information system, and the full name in English is Geographic Information System.

[0044] U-net is a convolutional neural network structure for image segmentation tasks.

[0045] opencv is an open source library for image processing and computer vision.

[0046] The present application will be described below in conjunction with preferred implementation steps, Figure 1 The flowchart of the building height detection method provided by the embodiments of the present application is shown in Figure 1 The method comprises the following steps:

[0047] Step S101, obtaining a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height.

[0048] For example, the building whose height needs to be measured (the above-mentioned target building) can be determined first, and then the remote sensing image corresponding to the building whose height needs to be measured (the above-mentioned target building) is obtained.

[0049] In step S102, the target remote sensing image is input into the target detection model for detection processing, and height information of a target building is output, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent height information of the building, and M is a positive integer.

[0050] For example, the target neural network model described above can be a U-Net network with residual connection. The U-Net network with residual connection (the target neural network model described above) can be learned and trained using a plurality of remote sensing images (the M remote sensing images described above) corresponding to buildings and shadow maps (the M shadow maps described above) thereof, and an elevation map (the M target elevation maps described above) obtained by subtracting a digital terrain model (DTM) from a digital surface model (DSM) generated by a radar image. Then, the image to be predicted (the target remote sensing image described above) is input into the trained network, and the height information of the building (the target building described above) whose height needs to be measured is output.

[0051] Through steps S101-S102 described above, the target remote sensing image corresponding to the target building is obtained, and the target remote sensing image is input into the target detection model for detection processing, and the height information of the target building is directly output, thereby improving the effect of detecting the height of the building.

[0052] Optionally, in the method for detecting the height of the building provided in the embodiments of the present application, the target detection model is obtained by: obtaining M remote sensing images, M shadow maps and M target elevation maps; obtaining a data set according to the M remote sensing images and the M shadow maps; obtaining a training set for model training based on the data set; taking the M target elevation maps as sample labels; learning and training the target neural network model using the training set and the sample labels to obtain the target detection model.

[0053] For example, the target neural network model described above can be a U-Net network with a residual connection form. The data set described above can be remote sensing images (M remote sensing images described above) corresponding to 20 buildings and their shadow maps (M shadow maps described above), and remote sensing images corresponding to 8 buildings and their shadow maps are selected from the data set as the training set. Then the remote sensing images corresponding to the 8 buildings and their shadow maps can be sent into the U-Net network (the target neural network model described above) with the residual connection together as training data (the training set described above), and the elevation map (M target elevation maps described above) obtained by the difference between the digital terrain model (DTM) and the digital surface model (DSM) generated by the radar image is used as a sample label to train the network, and the trained network is used as the target detection model.

[0054] In summary, by training the neural network model with the residual connection, a trained neural network model can be quickly and accurately obtained.

[0055] Optionally, in the building height detection method provided in the embodiments of the present application, the M shadow maps are obtained by: obtaining M remote sensing images; performing contrast enhancement processing on each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set at least includes M remote sensing images after the enhancement processing; converting the remote sensing image set into a gray image set, wherein the gray image set at least includes M gray images; determining a threshold set according to the M gray images, wherein the threshold set at least includes M separation thresholds; performing segmentation processing on the M gray images by using the threshold set to obtain a shadow map set; and obtaining the M shadow maps from the shadow map set.

[0056] For example, for a remote sensing high-resolution image (M remote sensing images described above), opencv can be used to obtain the corresponding shadow map (M shadow maps described above). Since the significant feature of the shadow on the image is that the pixel gray value is relatively low, thresholding can be used for segmentation, and considering that the image only has two features of interest: shadow and non-shadow, the gray histogram bimodal method can be used to obtain the threshold. Specifically, first, the contrast of the RGB image (M remote sensing images described above) is enhanced, then it is converted into a gray image (M gray images described above), and the average value of the two peaks is taken as the separation threshold (M separation thresholds described above), that is, the required building shadow map (M shadow maps described above) can be obtained.

[0057] Through the above scheme, according to the processing of the remote sensing image, the shadow map corresponding to the building can be quickly and accurately obtained.

[0058] Optionally, in the building height detection method provided in the embodiments of the present application, the M target elevation maps are obtained by: determining M target areas, wherein the target area is an area where the building is located; obtaining a radar image set according to the M target areas, wherein the radar image set at least includes M radar images; determining M first elevation maps and M second elevation maps based on the radar image set, wherein the first elevation map at least includes the altitude information of the ground and the height information of the target object, the target object is an object on the ground, and the second elevation map at least includes the altitude information of the ground; obtaining the elevation map set according to the M first elevation maps and the M second elevation maps; and obtaining the M target elevation maps from the elevation map set.

[0059] For example, the images used in the embodiments can be remote sensing high-resolution RGB images taken regularly by professional satellite manufacturers and radar images of the same area processed, wherein the radar images contain various echo information, so that various elevation maps such as DSM (the first elevation map described above) and DTM (the second elevation map described above) can be simulated by a specific algorithm, and then the DSM (the first elevation map described above) is subtracted from the DTM (the second elevation map described above) to obtain a new elevation map (the target elevation map described above).

[0060] In summary, by processing the obtained radar images, the elevation map for representing the altitude can be quickly and accurately obtained.

[0061] Optionally, in the building height detection method provided in the embodiments of the present application, the training set for model training is obtained based on the data set by: performing block processing on the data set to obtain a block-processed data set; and obtaining the training set from the block-processed data set according to a preset ratio.

[0062] For example, after the remote sensing images (the M remote sensing images described above) corresponding to the plurality of buildings and the shadow maps (the M shadow maps described above) thereof are taken as the data set, the data set can be block-processed with a step size of 64 and a size of 256x256, and then the block-processed data set can be randomly divided into the training set, the validation set and the test set according to the ratios of 85%, 10% and 5%.

[0063] In summary, by block processing the data set, the difficulty of network training can be reduced.

[0064] Optionally, in the building height detection method provided in the embodiments of the present application, before the target neural network model is learned and trained by using the training set and the sample label to obtain the target detection model, the method further includes: determining a first sub-module of a first neural network model, wherein the first sub-module is at least one of the following: an encoder module, a decoder module; and replacing the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0065] For example, the U-Net network with the residual connection can be built by using the Tenserflow framework (an open source framework encapsulating steps required by a deep learning algorithm), that is, the sub-module of the U-Net network is replaced with the form with the residual connection. Specifically, the residual connection can be introduced in the encoder and the decoder when the U-Net network is built by using the Tenserflow framework (an open source framework encapsulating steps required by a deep learning algorithm), and a schematic diagram of the U-Net network structure with the residual connection is as shown in FIG. 2. Figure 2 Moreover, Figure 2 The numbers above the arrows in FIG. 2 represent the sizes of features, Figure 2 and the numbers below the arrows represent the numbers of channels of the network.

[0066] Through the above scheme, the residual connection form can be quickly and accurately added to the neural network model.

[0067] Optionally, in the building height detection method provided in the embodiments of the present application, the target neural network model is learned and trained by using the training set and the sample label to obtain the target detection model, which includes: determining a loss function according to the training set and the sample label; and learning and training the target neural network model by using the loss function to obtain the target detection model.

[0068] For example, the MSE (the square of the difference between the output of the network and the label) can be used as the loss function of the training process to iteratively train the network parameters, and a trained network model is obtained.

[0069] Through the above scheme, the neural network model can be iteratively trained quickly and accurately.

[0070] Optionally, in the building height detection method provided in the embodiments of the present application, the target neural network model is learned and trained by using the loss function to obtain the target detection model, which includes: determining the initial parameters of the target neural network model, and determining the iteration number of learning and training the target neural network model; updating the initial parameters by using the loss function, and obtaining the target number of updating the initial parameters by using the loss function; when the target number is the same as the iteration number, obtaining the target parameters obtained by updating the initial parameters by using the loss function; replacing the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0071] For example, 64 pictures can be randomly selected from the training set as a batch of training data to train the network. The training data is randomly flipped when input into the network for training to improve the robustness of the network. The flipping angles can be: 0°, 90°, 180° and 270°. The Adam algorithm (an algorithm for gradient optimization of the loss function) can be used as the optimizer of the network, and the initial learning rate can be set to 0.0001, and the iteration is 500000 times, and the learning rate becomes 0.8 of the original every 20000 times. Until the number of iteration training reaches 500000 times, the current parameters are obtained, and the parameters of the network model are replaced with the current parameters.

[0072] Through the above scheme, the trained neural network model can be quickly and accurately obtained.

[0073] Optionally, in the building height detection method provided in the embodiments of the present application, after the target remote sensing image is input into the target detection model for detection processing and the height information of the target building is output, the method further includes: determining the construction progress of the target building according to the height information of the target building; determining whether the construction progress reaches the target progress; if the construction progress reaches the target progress, it indicates that the target object does not have a repayment risk, wherein the target object is an object that applies for a loan from a financial institution, and the loan is used to build the target building; if the construction progress does not reach the target progress, it indicates that the target object has a repayment risk.

[0074] For example, after the neural network outputs the height of the building, the obtained height information of the building can be used to monitor the progress of the construction in the risk control scene. Specifically, the current construction progress can be determined according to the obtained height of the building, and it is determined whether the current construction progress reaches the target construction progress. If the current construction progress reaches the target construction progress, it is determined that the object that applies for a loan from a financial institution and uses the loan to build a building does not have a repayment risk; if the current construction progress does not reach the target construction progress, it is determined that the object that applies for a loan from a financial institution and uses the loan to build a building has a repayment risk.

[0075] Through the above scheme, the construction progress of the building can be effectively monitored, so that the phenomenon of overdue repayment of loans can be avoided, and the safety of loans of financial institutions is ensured.

[0076] For example, Figure 3 An optional building height detection method provided by the embodiment of the application is shown in FIG. Figure 3 The flow of the optional building height detection method includes the following steps:

[0077] 1. The image used in the embodiment can come from a remote sensing high-resolution RGB image taken periodically by a professional satellite manufacturer and a radar image of the same area processed, wherein the RGB image has characteristics such as high resolution and wide radiation range, and the radar image contains various echo information, and various elevation maps such as DSM and DTM required by the algorithm can be simulated through a specific algorithm.

[0078] 2. The remote sensing high-resolution image in step 1 is processed using opencv to obtain a corresponding shadow map. The significant feature of the shadow on the image is that the pixel gray value is relatively low, and the threshold method can be used for segmentation. Considering that the image has only two features of interest: shadow and non-shadow, the gray histogram bimodal method can be used to obtain the threshold value. First, enhance the contrast of the RGB image, and then convert it to a gray image, and take the average value of the two peaks as the separation threshold, that is, the building shadow map required.

[0079] 3. The shadow map obtained in step 2 is spliced with the high-resolution RGB image in step 1 in the channel dimension to obtain a training data set.

[0080] 4. Subtract the DTM from the DSM to obtain the label corresponding to the training set.

[0081] 5. The data set is cut into blocks with a step size of 64 and a size of 256*256 to reduce the difficulty of network training.

[0082] 6. The data set in step 5 can be randomly divided into a training set, a validation set and a test set according to the proportions of 85%, 10% and 5%.

[0083] 7. A U-Net network with residual connection is built using the Tenserflow framework (an open source framework that encapsulates the steps required by the deep learning algorithm), and the schematic diagram of the U-Net network structure with residual connection is shown in FIG. Figure 2 That is, the sub-modules of the U-Net are replaced by the form with residual connection. The size of the network convolution kernel is set to 3*3, the step size is 1, the parameter is initialized to a normal distribution form with a mean of 0 and a standard deviation of 0.05, and relu is used as the activation function between layers.

[0084] 8、From the training set randomly selected 64 pictures as a batch of training data to train the network. When sent into the network training, the training data is randomly flipped to improve the robustness of the network, the angle of flip can be: 0°, 90°, 180° and 270°. Using Adam algorithm (gradient optimization algorithm for loss function) as the optimizer of the network, the initial learning rate is set to 0.0001, a total of 500000 iterations, every 20000 times learning rate becomes 0.8. The MSE (the square of the difference between the output of the network and the label) is used as the loss function of the training process to iterate the network parameters.

[0085] 9、After the network training is completed, the network parameters are saved, and the input is the image to be predicted. The output is the target building height map.

[0086] In this embodiment, a building height prediction model based on remote sensing images and building shadows is designed. Compared with the existing building height detection method, the main innovations of this method are as follows:

[0087] (1) The building height of the target area can be obtained end-to-end, and the post-processing step of the building detection model is saved, which enhances the expandability of the model.

[0088] (2) The model considers both shadow information and structural features of objects in RGB images to estimate building height information, while also reducing prediction error. The results show that the model has good robustness for inconsistent errors in the data set.

[0089] In addition, remote sensing images obtained by satellite sensors are a kind of image data with rich information and high geographical positioning accuracy. These images are widely used in geographic information systems (GIS) and have made significant contributions in urban planning, resource investigation, agriculture, forestry and animal husbandry. The image types in the field of remote sensing are diverse, including high-resolution remote sensing images containing rich ground texture information, and radar images capable of generating various elevation images.

[0090] Moreover, laser radar is an optical remote sensing technology that uses laser to densely sample the earth's surface. By recording the return of a single pulse, high-precision x, y, z measurements are obtained after analysis and processing. After professional processing, digital terrain model (DTM) and digital surface model (DSM) can be obtained. DTM is a digital representation of the morphological properties of the terrain surface, generally including the ground or bare ground, and DSM is a ground elevation model that includes the height information of buildings, bridges and trees. The difference between DSM and DTM can provide very useful information, such as building height and vegetation canopy height.

[0091] Therefore, in the embodiment, a building height prediction model based on remote sensing image and building shadow is proposed to solve the problems of the prior art. Moreover, with the rapid development of deep neural networks, great potential has been shown in the field of computer vision (cv). It can guide the model to solve non-deterministic problems through learning with high accuracy. After the building height detection model is trained, the high-resolution remote sensing image to be predicted can be input to obtain the height information of the target building directly without other operations.

[0092] In addition, the method inputs the remote sensing high-resolution RGB image and its shadow map as training data into the U-Net network with residual connection, uses the difference between the digital terrain model (DTM) and the digital surface model (DSM) generated by the radar image as the sample label to train the network. After the network is trained, the image to be predicted is input into the network, and the output is the required building height prediction map.

[0093] Moreover, in the embodiment, the remote sensing high-resolution image is first subjected to simple image processing to obtain the corresponding shadow map, and the result shadow map is spliced with the RGB image in the channel dimension to serve as the input of the neural network. In the method, the requirement for the accuracy of the shadow contour is not very high, because the neural network will directly map the high-resolution RGB image into a height map through learning, and the shadow here serves as an attention mechanism, the purpose of which is to make the network pay more attention to some areas where the shadow exists during learning, so that they can be better estimated. Next, the DSM is subtracted from the DTM to obtain the height map of the building, which is used as the sample label for training the network. In order to reduce the difficulty of model training and improve the processing time, residual connection is introduced in the encoder and decoder parts when the U-Net is built using the Tenserflow framework (an open source framework that encapsulates the steps required for deep learning algorithms). After the training is completed, the network parameters are saved, which is the building height prediction model.

[0094] In conclusion, the building height detection method provided in the embodiments of the present application comprises the following steps: obtaining a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height; inputting the target remote sensing image into a target detection model for detection processing, and outputting height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of a building, the target elevation map is used to represent height information of a building, and M is a positive integer. The problem of poor detection effect of building height in the related art is solved. By obtaining the target remote sensing image corresponding to the target building, inputting the target remote sensing image into the target detection model for detection processing, and directly outputting the height information of the target building, the detection effect of building height is improved.

[0095] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0096] The embodiments of the present application also provide a building height detection device. It should be noted that the building height detection device of the embodiments of the present application can be used to execute the building height detection method provided by the embodiments of the present application. The building height detection device provided by the embodiments of the present application is introduced as follows.

[0097] Figure 4 FIG. 1 is a schematic diagram of a building height detection device according to the embodiments of the present application. As shown in FIG. 1, the device comprises a first obtaining unit 401 and a first processing unit 402. Figure 4

[0098] Specifically, the first obtaining unit 401 is configured to obtain a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height.

[0099] The first processing unit 402 is configured to input the target remote sensing image into a target detection model for detection processing, and output height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of a building, the target elevation map is used to represent height information of a building, and M is a positive integer.

[0100] ​In conclusion, the building height detection device provided by the embodiments of the present application obtains the target remote sensing image corresponding to the target building through the first acquisition unit 401, wherein the target building is a building to be detected in height; the first processing unit 402 inputs the target remote sensing image into the target detection model for detection processing, and outputs the height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of a building, the target elevation map is used to represent the height information of a building, and M is a positive integer, thereby solving the problem of poor detection effect of building height in related technologies. By obtaining the target remote sensing image corresponding to the target building, and inputting the target remote sensing image into the target detection model for detection processing, the height information of the target building is directly outputted, thereby improving the detection effect of building height.

[0101] Optionally, in the building height detection device provided by the embodiments of the present application, the target detection model is obtained in the following manner: a second acquisition unit is configured to acquire M remote sensing images, M shadow maps and M target elevation maps; a first determination unit is configured to obtain a data set according to the M remote sensing images and the M shadow maps; a second determination unit is configured to obtain a training set for model training based on the data set; a third determination unit is configured to take the M target elevation maps as sample labels; and a first training unit is configured to learn and train the target neural network model using the training set and the sample labels to obtain the target detection model.

[0102] Optionally, in the building height detection device provided by the embodiments of the present application, the first training unit comprises: a first determination module configured to determine a loss function according to the training set and the sample labels; and a first training module configured to learn and train the target neural network model using the loss function to obtain the target detection model.

[0103] Optionally, in the building height detection device provided by the embodiments of the present application, the first training module comprises: a first determination sub-module configured to determine initial parameters of the target neural network model and determine the number of iterations for learning and training the target neural network model; a first processing sub-module configured to update the initial parameters using the loss function and obtain the target number of times of updating the initial parameters using the loss function; a first acquisition sub-module configured to obtain the target parameters obtained by updating the initial parameters using the loss function when the target number of times is equal to the number of iterations; and a first replacement sub-module configured to replace the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0104] Optionally, in the building height detection apparatus provided by the embodiment of the present application, the second obtaining unit comprises: a first obtaining module, configured to obtain M remote sensing images; a first processing module, configured to perform enhancement processing on the contrast of each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set comprises at least M remote sensing images after the enhancement processing; a first conversion module, configured to convert the remote sensing image set into a gray scale image set, wherein the gray scale image set comprises at least M gray scale images; a second determining module, configured to determine a threshold set according to the M gray scale images, wherein the threshold set comprises at least M separation thresholds; a second processing module, configured to perform segmentation processing on the M gray scale images by using the threshold set to obtain a shadow map set; and a second obtaining module, configured to obtain M shadow maps from the shadow map set.

[0105] Optionally, in the building height detection apparatus provided by the embodiment of the present application, the second obtaining unit comprises: a third determining module, configured to determine M target regions, wherein the target region is a region where the building is located; a fourth determining module, configured to obtain a radar image set according to the M target regions, wherein the radar image set comprises at least M radar images; a fifth determining module, configured to determine M first elevation maps and M second elevation maps based on the radar image set, wherein the first elevation map comprises at least the height information of the target object and the elevation information of the ground, the target object is an object on the ground, and the second elevation map comprises at least the elevation information of the ground; a sixth determining module, configured to obtain an elevation map set according to the M first elevation maps and the M second elevation maps; and a third obtaining module, configured to obtain M target elevation maps from the elevation map set.

[0106] Optionally, in the building height detection apparatus provided by the embodiment of the present application, the apparatus further comprises: a fourth determining unit, configured to determine a first sub-module of the first neural network model before learning and training the target neural network model by using the training set and the sample label to obtain the target detection model, wherein the first sub-module is at least one of the following: an encoder module, a decoder module; and a first replacing unit, configured to replace the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0107] Optionally, in the building height detection apparatus provided by the embodiment of the present application, the second determining unit comprises: a third processing module, configured to perform cutting block processing on the data set to obtain a cutting block processed data set; and a fourth obtaining module, configured to obtain the training set from the cutting block processed data set according to a preset ratio.

[0108] Optionally, in the building height detection device provided by the embodiment of the application, the device further comprises: a fifth determination unit configured to determine a construction progress of the target building according to the height information of the target building after inputting the target remote sensing image into the target detection model for detection processing and outputting the height information of the target building; a first judgment unit configured to judge whether the construction progress reaches a target progress; a sixth determination unit configured to indicate that a target object does not have a repayment risk if the construction progress reaches the target progress, wherein the target object is an object that applies for a loan from a financial institution, and the loan is used for building the target building; and a seventh determination unit configured to indicate that the target object has the repayment risk if the construction progress does not reach the target progress.

[0109] The building height detection device comprises a processor and a memory, the first acquisition unit 401 and the first processing unit 402 and the like are stored in the memory as program units, and the corresponding functions are realized by the processor executing the program units stored in the memory.

[0110] The processor comprises a core, and the core retrieves the corresponding program units from the memory. The core can be one or more, and the effect of detecting the building height is improved by adjusting the core parameters.

[0111] The memory can include a non-permanent memory in a computer readable medium, a random access memory (RAM) and / or a non-volatile memory such as a read-only memory (ROM) or a flash memory (flash RAM), and the memory comprises at least one memory chip.

[0112] The embodiment of the application provides a computer readable storage medium, and a program is stored on the computer readable storage medium. The program is executed by the processor to realize the building height detection method.

[0113] The embodiment of the application provides a processor, and the processor is used to run a program. When the program is run, the building height detection method is executed.

[0114] As Figure 5As shown, the embodiment of the present application provides an electronic device, the device comprising a processor, a memory, and a program stored on the memory and executable on the processor, and the processor implements the following steps when executing the program: obtaining a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected in height; inputting the target remote sensing image into a target detection model for detection processing, and outputting height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps, and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent height information of the building, and M is a positive integer.

[0115] The processor further implements the following steps when executing the program: the target detection model is obtained by: obtaining the M remote sensing images, the M shadow maps, and the M target elevation maps; obtaining a data set according to the M remote sensing images and the M shadow maps; obtaining a training set for model training based on the data set; taking the M target elevation maps as sample labels; and learning and training the target neural network model using the training set and the sample labels to obtain the target detection model.

[0116] The processor further implements the following steps when executing the program: learning and training the target neural network model using the training set and the sample labels to obtain the target detection model comprises: determining a loss function according to the training set and the sample labels; and learning and training the target neural network model using the loss function to obtain the target detection model.

[0117] The processor further implements the following steps when executing the program: learning and training the target neural network model using the loss function to obtain the target detection model comprises: determining initial parameters of the target neural network model, and determining an iteration number of learning and training the target neural network model; updating the initial parameters using the loss function, and obtaining a target number of times of updating the initial parameters using the loss function; obtaining target parameters obtained by updating the initial parameters using the loss function when the target number of times is the same as the iteration number; and replacing the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0118] The processor further implements the following steps when executing the program: obtaining the M shadow maps comprises: obtaining the M remote sensing images; performing enhancement processing on the contrast of each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set at least includes M remote sensing images after enhancement processing; converting the remote sensing image set into a gray image set, wherein the gray image set at least includes M gray images; determining a threshold set according to the M gray images, wherein the threshold set at least includes M separation thresholds; performing segmentation processing on the M gray images by using the threshold set to obtain a shadow map set; and obtaining the M shadow maps from the shadow map set.

[0119] The processor further implements the following steps when executing the program: obtaining the M target elevation maps comprises: determining M target regions, wherein the target regions are regions where the buildings are located; obtaining a radar image set according to the M target regions, wherein the radar image set at least includes M radar images; determining M first elevation maps and M second elevation maps based on the radar image set, wherein the first elevation maps at least include the height information of the target objects and the elevation information of the ground, the target objects are objects on the ground, and the second elevation maps at least include the elevation information of the ground; obtaining an elevation map set according to the M first elevation maps and the M second elevation maps; and obtaining the M target elevation maps from the elevation map set.

[0120] The processor further implements the following steps when executing the program: before learning and training the target neural network model by using the training set and the sample label to obtain the target detection model, the method further comprises: determining a first sub-module of a first neural network model, wherein the first sub-module is at least one of the following: an encoder module, a decoder module; and replacing the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0121] The processor further implements the following steps when executing the program: obtaining a training set for model training based on the data set comprises: performing cut block processing on the data set to obtain a cut block processed data set; and obtaining the training set from the cut block processed data set according to a preset ratio.

[0122] The processor also implements the following steps when executing the program: after inputting the target remote sensing image into the target detection model for detection processing and outputting the height information of the target building, the method further comprises: determining the construction progress of the target building according to the height information of the target building; determining whether the construction progress reaches a target progress; if the construction progress reaches the target progress, it indicates that the target object does not have a repayment risk, wherein the target object is an object applying for a loan from a financial institution, and the loan is used for building the target building; if the construction progress does not reach the target progress, it indicates that the target object has a repayment risk.

[0123] The device herein can be a server, a PC, a PAD, a mobile phone, etc.

[0124] The application also provides a computer program product, which, when executed on a data processing device, is adapted to execute a program initialized with the following method steps: obtaining a target remote sensing image corresponding to a target building, wherein the target building is a building to be detected for height; inputting the target remote sensing image into a target detection model for detection processing, and outputting height information of the target building, wherein the target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps and M target elevation maps, the target neural network model is a neural network model with residual connection, the remote sensing image is a remote sensing image corresponding to a building, the shadow map is a shadow map of the building, the target elevation map is used to represent the height information of the building, and M is a positive integer.

[0125] When executed on a data processing device, it is also adapted to execute a program initialized with the following method steps: the target detection model is obtained by: obtaining the M remote sensing images, the M shadow maps and the M target elevation maps; obtaining a data set according to the M remote sensing images and the M shadow maps; obtaining a training set for model training based on the data set; taking the M target elevation maps as sample labels; learning and training the target neural network model using the training set and the sample labels to obtain the target detection model.

[0126] When executed on a data processing device, it is also adapted to execute a program initialized with the following method steps: learning and training the target neural network model using the training set and the sample labels to obtain the target detection model comprises: determining a loss function according to the training set and the sample labels; learning and training the target neural network model using the loss function to obtain the target detection model.

[0127] When executed on a data processing device, the program is further adapted to execute the method steps of initializing the program to: obtain the target detection model by performing learning training on the target neural network model using the loss function, comprising: determining initial parameters of the target neural network model, and determining an iteration number of performing learning training on the target neural network model; performing update processing on the initial parameters using the loss function, and obtaining a target number of performing update processing on the initial parameters using the loss function; when the target number is the same as the iteration number, obtaining target parameters obtained by performing update processing on the initial parameters using the loss function; and replacing the initial parameters of the target neural network model with the target parameters to obtain the target detection model.

[0128] When executed on a data processing device, the program is further adapted to execute the method steps of initializing the program to: obtain the M shadow maps, comprising: obtaining the M remote sensing images; performing enhancement processing on the contrast of each remote sensing image to obtain a remote sensing image set, wherein the remote sensing image set at least includes M remote sensing images after enhancement processing; converting the remote sensing image set into a gray image set, wherein the gray image set at least includes M gray images; determining a threshold set according to the M gray images, wherein the threshold set at least includes M separation thresholds; performing segmentation processing on the M gray images using the threshold set to obtain a shadow map set; and obtaining the M shadow maps from the shadow map set.

[0129] When executed on a data processing device, the program is further adapted to execute the method steps of initializing the program to: obtain the M target elevation maps, comprising: determining M target regions, wherein the target regions are regions where the buildings are located; obtaining a radar image set according to the M target regions, wherein the radar image set at least includes M radar images; determining M first elevation maps and M second elevation maps based on the radar image set, wherein the first elevation maps at least include ground elevation information and target object height information, the target objects are objects on the ground, and the second elevation maps at least include the ground elevation information; obtaining an elevation map set according to the M first elevation maps and the M second elevation maps; and obtaining the M target elevation maps from the elevation map set.

[0130] When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: before learning training the target neural network model using the training set and the sample label, the method further comprises: determining a first sub-module of a first neural network model, wherein the first sub-module is at least one of: an encoder module, a decoder module; replacing the first sub-module with a second sub-module to obtain the target neural network model, wherein the second sub-module is a sub-module with a residual connection.

[0131] When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: obtaining a training set for model training based on the data set comprises: performing patch processing on the data set to obtain a patch-processed data set; and obtaining the training set from the patch-processed data set according to a preset proportion.

[0132] When executed on a data processing device, it is also suitable to execute a program initialized with the following method steps: after inputting the target remote sensing image into the target detection model for detection processing and outputting the height information of the target building, the method further comprises: determining the construction progress of the target building according to the height information of the target building; determining whether the construction progress reaches a target progress; if the construction progress reaches the target progress, indicating that the target object does not have a repayment risk, wherein the target object is an object applying for a loan from a financial institution, and the loan is used for building the target building; and if the construction progress does not reach the target progress, indicating that the target object has a repayment risk.

[0133] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) having computer-usable program code embodied therein.

[0134] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a machine that implements the functions described in the flowcharts and / or block diagrams.Figure 1 apparatuses that perform the functions recited in flow Figure 1

[0135] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 apparatuses that perform the functions recited in flow Figure 1

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 apparatuses that perform the functions recited in flow Figure 1

[0137] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0138] The memory can include non-persistent memory and / or volatile memory, such as a random access memory (RAM) including a cache area for the temporary storage of data. The memory can also include non-volatile memory, such as read only memory (ROM), electrically programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), flash memory, or non-volatile random access memory (NVRAM). The memory can also inoperable with a computer-readable medium storing the instructions.

[0139] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for the storage of information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0140] ​​​It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0141] Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0142] The embodiments of the present application are only illustrative and are not intended to limit the present application. Various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for detecting the height of a building, characterized in that, include: Acquire a remote sensing image of the target building, wherein the target building is the building whose height is to be detected; The target remote sensing image is input into a target detection model for detection processing, and the height information of the target building is output. The target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps, and M target elevation maps. The target neural network model is a neural network model with residual connections. The remote sensing images are remote sensing images corresponding to the building. The shadow maps are shadow maps of the building. The target elevation maps are used to represent the height information of the building. M is a positive integer. The target detection model is obtained in the following way: Acquire the M remote sensing images, the M shadow maps, and the M target elevation maps; A dataset is obtained based on the M remote sensing images and the M shadow maps; Based on the dataset, a training set for model training is obtained; Use the M target elevation maps as sample labels; The target neural network model is trained using the training set and the sample labels to obtain the target detection model. Obtaining the M shadow images includes: Acquire the M remote sensing images; The contrast of each remote sensing image is enhanced to obtain a set of remote sensing images, wherein the set of remote sensing images includes at least M enhanced remote sensing images. The remote sensing image set is converted into a grayscale image set, wherein the grayscale image set includes at least M grayscale images; Based on the M grayscale images, a threshold set is determined, wherein the threshold set includes at least M separation thresholds; The M grayscale images are segmented using the threshold set to obtain a set of shadow images; Obtain the M shadow maps from the set of shadow maps; The method further includes, after inputting the target remote sensing image into the target detection model for detection processing and outputting the height information of the target building, the method further includes: Based on the height information of the target building, the construction schedule of the target building is determined; Determine whether the construction progress has reached the target progress; If the construction progress reaches the target progress, it means that the target object has no repayment risk. The target object is the object that applied for a loan from a financial institution, and the loan is used to construct the target building. If the construction progress does not reach the target progress, it indicates that the target entity faces repayment risk.

2. The method according to claim 1, characterized in that, The target neural network model is trained using the training set and the sample labels to obtain the target detection model, which includes: Based on the training set and the sample labels, determine the loss function; The target neural network model is trained using the loss function to obtain the target detection model.

3. The method according to claim 2, characterized in that, The target detection model is obtained by training the target neural network model using the loss function, including: Determine the initial parameters of the target neural network model and the number of iterations for learning and training the target neural network model; The initial parameters are updated using the loss function, and the target number of times the initial parameters are updated using the loss function is obtained. When the target number is the same as the iteration number, obtain the target parameters obtained by updating the initial parameters using the loss function; The target detection model is obtained by replacing the initial parameters of the target neural network model with the target parameters.

4. The method according to claim 1, characterized in that, Obtaining the M target elevation maps includes: Identify M target areas, where each target area is the area where the building is located; Based on the M target regions, a radar image set is obtained, wherein the radar image set includes at least M radar images; Based on the radar image set, M first elevation maps and M second elevation maps are determined, wherein the first elevation map includes at least the elevation information of the ground and the height information of the target object, the target object being an object on the ground, and the second elevation map includes at least the elevation information of the ground. Based on the M first elevation maps and the M second elevation maps, an elevation map set is obtained; Obtain the M target elevation maps from the set of elevation maps.

5. The method according to claim 1, characterized in that, Before training the target neural network model using the training set and the sample labels to obtain the target detection model, the method further includes: A first sub-module of the first neural network model is determined, wherein the first sub-module is at least one of the following: an encoder module and a decoder module; The first submodule is replaced with the second submodule to obtain the target neural network model, wherein the second submodule is a submodule with residual connections.

6. The method according to claim 1, characterized in that, Based on the dataset, the training set obtained for model training includes: The dataset is divided into chunks to obtain a chunked dataset; The training set is obtained from the data set after it has been segmented according to a preset ratio.

7. A device for detecting the height of a building, characterized in that, include: The first acquisition unit is used to acquire a remote sensing image of the target building, wherein the target building is a building whose height is to be detected; The first processing unit is used to input the target remote sensing image into a target detection model for detection processing and output the height information of the target building. The target detection model is a model generated by learning and training a target neural network model using M remote sensing images, M shadow maps, and M target elevation maps. The target neural network model is a neural network model with residual connections. The remote sensing images are remote sensing images corresponding to the building. The shadow maps are shadow maps of the building. The target elevation maps are used to represent the height information of the building. M is a positive integer. The target detection model is obtained through the following methods: a second acquisition unit, used to acquire M remote sensing images, M shadow maps, and M target elevation maps; a first determination unit, used to obtain a dataset based on the M remote sensing images and M shadow maps; a second determination unit, used to obtain a training set for model training based on the dataset; a third determination unit, used to use the M target elevation maps as sample labels; and a first training unit, used to train the target neural network model using the training set and sample labels to obtain the target detection model. The second acquisition unit includes: a first acquisition module for acquiring M remote sensing images; a first processing module for enhancing the contrast of each remote sensing image to obtain a set of remote sensing images, wherein the set of remote sensing images includes at least M enhanced remote sensing images; a first conversion module for converting the set of remote sensing images into a set of grayscale images, wherein the set of grayscale images includes at least M grayscale images; a second determination module for determining a threshold set based on the M grayscale images, wherein the threshold set includes at least M separation thresholds; a second processing module for segmenting the M grayscale images using the threshold set to obtain a set of shadow images; and a second acquisition module for acquiring M shadow images from the set of shadow images. The device further includes: a fifth determining unit, used to determine the construction progress of the target building based on the target building's height information after inputting the target remote sensing image into the target detection model for detection processing and outputting the target building's height information; a first judging unit, used to judge whether the construction progress has reached the target progress; a sixth determining unit, used to indicate that the target object has no repayment risk if the construction progress has reached the target progress, wherein the target object is an object applying for a loan from a financial institution, and the loan is used to construct the target building; and a seventh determining unit, used to indicate that the target object has repayment risk if the construction progress has not reached the target progress.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program, wherein the program executes the building height detection method according to any one of claims 1 to 6.

9. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the building height detection method according to any one of claims 1 to 6.

Images