Radiation source positioning method, computer device and computer storage medium

Abstract

Embodiments of the present application disclose a radiation source positioning method, a computer device and a computer storage medium. A general radiation source positioning technology based on an efficient neural network is proposed. For a light-weight electromagnetic spectrum sensing terminal, the energy load is limited, the computing power is limited, the computing chip carried by the terminal has limited operators, and the real-time requirement of the positioning algorithm. Therefore, research on a general target positioning technology based on an efficient neural network is needed. In the embodiments, network parameters are optimally selected to realize a sensing field covering a full image, so that the computing cost is reduced, the output delay is reduced, the computing efficiency of the neural network model is improved, and a higher degree of general operator is used, so that the algorithm can be deployed to a wider computing platform, and practical deployment is possible.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of image processing, in particular to a radiation source positioning method, a computer device and a computer storage medium. BACKGROUND

[0002] Traditional target position positioning methods need to estimate the angle of the radiation source, the time difference of arrival at each platform, the signal strength and other information first, and then solve nonlinear equations, so it is also called two-step positioning.

[0003] However, in urban environments, due to the widespread existence of building obstructions, the radiation source often cannot be located within the line-of-sight range of the electromagnetic sensing device, reducing the positioning efficiency of the target position. Moreover, the traditional two-step positioning method has cumulative and amplification of positioning errors, and usually needs to solve nonlinear equations, which will amplify the positioning error, resulting in low positioning accuracy and low positioning efficiency. SUMMARY

[0004] Embodiments of the present application provide a radiation source positioning method, a computer device and a computer storage medium, which realize a perception field covering the entire image by adjusting the depth network hyperparameters, thereby obtaining higher passive positioning accuracy.

[0005] The first aspect of the embodiments of the present application provides a radiation source positioning method, the method comprising:

[0006] Obtain a simulation dataset of a city geographical environment, the simulation dataset comprising a global path loss map of a simulation radiation source, a simulation grid image of the city geographical environment and a preset spatial coordinate of the simulation radiation source;

[0007] Obtain an initial neural network, and use the initial neural network to perform spatial random sampling on the global path loss map to obtain an electromagnetic information image of the city geographical environment;

[0008] Input the simulation grid image and the electromagnetic information image into the initial neural network, so that the initial neural network predicts the position of the radiation source in the city geographical environment according to the simulation grid image and the electromagnetic information image, and obtains the predicted position coordinates of the radiation source output by the initial neural network;

[0009] Adjust the model parameters of the initial neural network according to the predicted position coordinates and the preset spatial coordinates, and stop the model training when the model training of the initial neural network reaches a convergence condition, to obtain a target neural network, the target neural network being used to predict the position of the radiation source in the city geographical environment.

[0010] The second aspect of the embodiments of the present application provides a computer device, the computer device comprising:

[0011] The first obtaining unit is configured to obtain a simulation dataset of a city geographical environment, the simulation dataset comprising a global path loss map of a simulation radiation source, a simulation grid image of the city geographical environment, and preset spatial coordinates of the simulation radiation source.

[0012] The second obtaining unit is configured to obtain an initial neural network, and perform spatial random sampling on the global path loss map by using the initial neural network to obtain an electromagnetic information image of the city geographical environment.

[0013] The model training unit is configured to input the simulation grid image and the electromagnetic information image into the initial neural network, so that the initial neural network predicts a position of a radiation source in the city geographical environment according to the simulation grid image and the electromagnetic information image, and obtains predicted position coordinates of the radiation source output by the initial neural network.

[0014] The model training unit is further configured to adjust model parameters of the initial neural network according to the predicted position coordinates and the preset spatial coordinates, until the model training of the initial neural network reaches a convergence condition, and then stop the model training to obtain a target neural network, the target neural network being configured to predict a position of a radiation source in a city geographical environment.

[0015] The third aspect of the embodiment of the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method of the first aspect when executing the computer program.

[0016] The fourth aspect of the embodiment of the present application provides a computer storage medium, the computer storage medium stores instructions, and the instructions make a computer execute the method of the first aspect when the instructions are executed on the computer.

[0017] As can be seen from the above technical solutions, the embodiment of the present application has the following advantages:

[0018] For a lightweight electromagnetic spectrum sensing terminal, the energy load is limited, the computing power is limited, the computing chip carried can only calculate limited operators, and the real-time requirement of the positioning algorithm, the general target positioning technology based on the high-efficiency neural network needs to be researched. The embodiment of the present application optimally selects network parameters, realizes the perception field covering the whole image, thereby reducing the computing cost, reducing the output delay, improving the computing efficiency of the neural network model, and using a higher degree of general operator, so that the algorithm can be deployed to a wider computing platform, and the actual deployment is possible. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1A flowchart of a radiation source positioning method in an embodiment of the present application is shown in FIG. 1.

[0020] Figure 2 An exemplary simulation example diagram of a city geographical environment obtained by using simulation software in an embodiment of the present application is shown in FIG. 2.

[0021] Figure 3 An exemplary simulation grid image of a city geographical environment obtained based on the simulation example diagram shown in FIG. 2 is shown in FIG. 3. Figure 2

[0022] An exemplary simulation grid image of a city geographical environment obtained based on the simulation example diagram shown in FIG. 2 is shown in FIG. 3. Figure 4 Figure 2 Figure 1 An exemplary simulation grid image of a city geographical environment obtained based on the simulation example diagram shown in FIG. 2 is shown in FIG. 3.

[0023] Figure 5 An exemplary structure diagram of a backbone structure of an initial neural network in an embodiment of the present application is shown in FIG. 4.

[0024] Figure 6 An exemplary diagram of positioning effect of a radiation source output by a target neural network based on a test set in an embodiment of the present application is shown in FIG. 5.

[0025] Figure 7 A structure diagram of a computer device in an embodiment of the present application is shown in FIG. 6.

[0026] Figure 8 Another structure diagram of a computer device in an embodiment of the present application is shown in FIG. 7. DETAILED DESCRIPTION

[0027] The embodiment of the present application provides a radiation source positioning method, a computer device and a computer storage medium. By adjusting a deep network hyperparameter, a perception field covering a full image is realized, so that higher passive positioning precision is obtained.

[0028] The radiation source positioning method in the embodiment of the present application uses the propagation characteristics of electromagnetic signals to develop target position direct estimation technology. Although the city environment is complex and changeable, such real-time mapping means can often provide rough prior condition information of the environment, so as to be used for position inference. In the embodiment of the present application, based on an encoder-decoder network structure, a general radiation source positioning technology based on an efficient neural network is proposed. By adjusting a deep network hyperparameter, a perception field covering a full image is realized, so that higher passive positioning precision is obtained.

[0029] The radiation source positioning method in the embodiment of the present application is described below:

[0030] Referring to FIG. 1, one embodiment of the radiation source positioning method in the embodiment of the present application includes the following steps. Figure 1 ​​​

[0031] 101. obtaining a simulation dataset of an urban geographic environment, the simulation dataset comprising a global path loss map of a simulation radiation source, a simulation raster image of the urban geographic environment, and preset spatial coordinates of the simulation radiation source;

[0032] The method of the embodiment can be applied to a computer device, which can be a terminal, a server, or any device with certain computing and data processing capabilities. The computer device can use a simulation software program to simulate an urban geographic environment, which can be any regional location. During the simulation of the urban geographic environment using the simulation software program, a simulation dataset of the urban geographic environment generated by the simulation of the urban geographic environment can be obtained, which includes a global path loss map of a simulation radiation source, a simulation raster image of the urban geographic environment, and preset spatial coordinates of the simulation radiation source.

[0033] In this embodiment, the radiation source to be located refers to an object capable of generating electromagnetic radiation, including but not limited to a communication base station, a black radio station, etc.

[0034] For example, to construct a simulation dataset on a large scale, first, diverse urban geographic information can be collected from a large-scale open-source world map (such as OpenStreetMap and other map software). OpenStreetMap contains crowd-sourced labeled shape geometry information and height information of urban buildings, which has a key impact on electromagnetic spectrum propagation in the city. Specifically, the simulation process is as follows:

[0035] 1) Import the urban geographic information into WallMan of Feko-Winprop high-precision electromagnetic simulation software, which includes the height and shape of buildings in the urban geographic environment and other information for describing the influence of buildings on electromagnetic spectrum propagation;

[0036] 2) Set the electromagnetic property parameters of the urban buildings in WallMan;

[0037] 3) Set the antenna parameters and position of the radiation source, and the simulation receiving antenna parameters in PropMan of Feko-Winprop, and perform simulation calculation on the signal propagation of the radiation source with a fixed height. Then, spatial and grayscale quantization is performed, and the obtained rasterized image is represented as a global path loss map of the radiation source ~ X i ; wherein the antenna parameters can include frequency response, height, gain, and other parameters related to the antenna;

[0038] 4) Save the rasterized grayscale image representing the geometric shape of the urban buildings from a bird's eye view ~ I iAs the city geographic information. The non-zero gray value area represents the shape and height of the city building, and the zero gray value area represents the open area without building in the city;

[0039] 5) save this round of simulation, the corresponding spatial coordinates of the radiation source

[0040] For example, by using Feko-Winprop simulation software, a simulation example diagram as shown in Figure 2 can be obtained, and based on the simulation example diagram, a simulation grid image as shown in Figure 3 can be obtained and a global path loss diagram of the simulation radiation source as shown in Figure 4 can be obtained

[0041] It should be noted that in this embodiment, OpenStreetMap and Winprop are used as an example in the simulation dataset generation step, and actually, other simulation software can also be used to generate maps and electromagnetic field distribution.

[0042] 102, obtain an initial neural network, and use the initial neural network to perform spatial random sampling on the global path loss diagram to obtain an electromagnetic information image of the city geographic environment;

[0043] The computer device can further deploy an initial neural network, and use the initial neural network to perform spatial random sampling on the global path loss diagram of the simulation radiation source to obtain an electromagnetic information image of the city geographic environment.

[0044] For example, taking the training process of the simulation dataset as an example, the overall framework of the initial neural network adopts an encoder-decoder structure, the first layer input is a simulation grid image of the city geographic environment and the global path loss diagram is spatially randomly sampled to obtain an electromagnetic information image of the city geographic environment

[0045] 103, input the simulation grid image and the electromagnetic information image to the initial neural network, so that the initial neural network predicts the position of the radiation source of the city geographic environment according to the simulation grid image and the electromagnetic information image, and obtains the predicted position coordinates of the radiation source output by the initial neural network;

[0046] This embodiment uses signal strength data with spatial coordinates and city geographic environment data as deep learning training input to directly estimate the spatial coordinates of the radiation source in an end-to-end manner.

[0047] Therefore, after obtaining the electromagnetic information image of the urban geographic environment, the simulated raster image and the electromagnetic information image can be input into the initial neural network. The initial neural network then predicts the location of the radiation source in the urban geographic environment based on the simulated raster image and the electromagnetic information image, and obtains the predicted location coordinates of the radiation source output by the initial neural network.

[0048] It is important to note that the receptive field of the neural network, determined by the convolutional kernel size, dilation coefficient, and sampling scaling ratio, has a significant impact on localization accuracy. Therefore, in this embodiment, a generative model is actually used to solve for the radiation source location, with the goal of obtaining the location from the input simulated raster image of the urban geographic environment. And electromagnetic information images obtained by spatial sampling The predicted location coordinates of the radiation source are obtained from the model output. This is used to estimate the true location η of the radiation source. i .

[0049] In obtaining the predicted location coordinates of the radiation source from the initial neural network output, one possible approach is to obtain the conditional distribution of the predicted location coordinates of the radiation source from the initial neural network output, which represents each predicted location coordinate and its corresponding confidence probability.

[0050] Specifically, since the model training task in this embodiment inherently contains randomness, i.e., there exists a conditional distribution... The idea behind generative modeling is to use neural networks to represent this conditional distribution. For this method, the true radiation source location η i The information is contained in the image-represented simulated raster image. and electromagnetic information images And the interrelationships between them. Therefore, for the model training task in this embodiment to improve positioning accuracy, the key to neural network structure design lies in designing a structure that can better represent... With η i The structure of the interrelationships between them. After representing the urban geographic and electromagnetic environment using rasterized images, their overall information is contained within... and The convolutional neural network (CNN) needs to model and represent the relationships between all grid points in the entire two-dimensional space to effectively characterize the relationships between them. The receptive field of a neural network determines the spatial range within a two-dimensional grid space that the network can represent the relationships between modeled grid points. Experimental results show that a receptive field covering the entire image is crucial for the localization accuracy of the model training task in this embodiment.

[0051] 104. adjusting model parameters of the initial neural network according to the predicted position coordinates and the preset spatial coordinates until the model training of the initial neural network reaches a convergence condition, and obtaining a target neural network, the target neural network being used for predicting the position of the radiation source in the urban geographic environment;

[0052] When the model outputs the predicted position coordinates of the radiation source in the urban geographic environment, a loss function can be constructed according to the predicted position coordinates and the preset spatial coordinates of the simulation radiation source set in the simulation process of the urban geographic environment, and the model parameters of the initial neural network are adjusted according to the loss function. When the loss function meets the convergence condition, it is determined that the model training of the initial neural network reaches the convergence condition, at which time the model training can be stopped, and the target neural network obtained by the initial neural network after the model training is obtained. The target neural network can be used to predict the position of the radiation source in the urban geographic environment.

[0053] For example, for the current parameters, and considering the computer computing power constraint, a neural network backbone structure can be selected, and the structure is as shown in the table. Figure 5 As shown in the figure, the 10th layer 2D-Softmax performs spatial dimension exponential normalization on the features, and the 11th layer, which is the last layer, adopts the 2D-CenterMass layer to output the predicted positioning coordinates of the radiation source. The 11th layer 2D-CenterMass performs linear combination on the results of the 10th layer and the grid coordinates, so as to obtain the positioning coordinates of the radiation source and output the predicted position coordinates of the radiation source. In the network, the spatial down-sampling features are calculated by the step convolution, the perception field is increased by the dilated convolution, and the spatial up-sampling features are calculated by the combination module of the nearest neighbor interpolation and the convolution, wherein the convolution kernel size, the dilated coefficient and the sampling scaling ratio are shown in the table.

[0054] In terms of model training, the average square error of the predicted position coordinates of the radiation source and the real position coordinates is used as the loss function, the optimizer selects Adam, the parameters β1=0.9, β2=0.999, and the learning rate is set to 1e -3 .

[0055] In an optional implementation of the embodiment, after obtaining the target neural network through model training, the pre-collected spatial spectrum data set of the urban geographic environment and the actually collected grid image of the urban geographic environment can be obtained. The spatial spectrum data set includes a plurality of spatial spectrum data samples, and each spatial spectrum data sample includes the spatial coordinates of the sample of the urban geographic environment and the electromagnetic spectrum signal receiving intensity of the urban geographic environment.

[0056] Then, the spatial data and the intensity data of the spatial spectrum data set and the actually collected grid image can be quantized respectively to obtain the rasterized spatial spectrum data set corresponding to the spatial spectrum data set and the urban geographic environment quantization image corresponding to the actually collected grid image.

[0057] The rasterized spatial spectrum dataset and the quantized map of the urban geographic environment are input into the target neural network so that the target neural network outputs the predicted location of the radiation source based on the rasterized spatial spectrum dataset and the quantized map of the urban geographic environment.

[0058] One possible implementation of quantizing spatial data in a spatial spectrum dataset is to select a top view of a first preset size from the spatial spectrum dataset, and then divide the top view into multiple grids using a second preset size as the unit size, thereby obtaining a rasterized spatial spectrum dataset represented as a rasterized image. The second preset size is smaller than the first preset size.

[0059] One possible implementation for quantizing intensity data in a spatial spectrum dataset is to treat the grayscale data of an image as the intensity data of the image, and to quantize the grayscale data of the spatial spectrum dataset using a preset number of bits to obtain a rasterized spatial spectrum dataset.

[0060] For example, suppose spatial-spectral data of the urban geographic environment is collected at a fixed altitude, and a single spatial-spectral data sample is set as x. ij =[x ij ,y ij ,a ij ], tagged as η i Where x ij and y ij Let a and b represent the spatial coordinates of the j-th sample collected under urban geographic environment i, respectively. ij η represents the received signal strength collected by the electromagnetic spectrum load. i This represents the true spatial coordinates of the radiation source corresponding to the spatial-spectral data set. Assuming that the data collected from the same radiation source location within the same urban geographical environment contains L samples, the corresponding spatial-spectral data can be represented as X. i =[x i1 ,x i2 ,...,x iL The associated urban geographic environment can be represented as I using a raster image. i .

[0061] Afterwards, in order to target X i and I i Processed uniformly using a raster image method, it can handle X... i The spatial and intensity data were pre-processed with quantization, and then further processed to obtain rasterized spatial spectral data represented as rasterized images. When quantizing spatial data, the urban geographic environment within a 256m × 256m square area in the top-view can be selected, with 1m × 1m as the unit distance, generating 256 × 256 spatial resolution spatial-spectral data. For intensity (grayscale) quantization, 8-bit quantization can be used. Similarly, the actual raster image I of the urban geographic environment data can also be processed. i Similarly, by performing quantization processing on spatial and intensity data, a quantized map of the urban geographic environment corresponding to the actual raster image can be obtained.

[0062] In this process, multiple intensity data points located at the same grid point after spatial quantization can be averaged. Therefore, after the above preliminary preprocessing, a real-sampled spatial-spectral dataset represented in the form of a raster image can be obtained. Where N is the number of spatial-spectral data pairs in the dataset.

[0063] In addition, during storage, as a standardized dataset, image data with a spatial resolution of 256×256 can be extracted and stored within a 256m×256m range corresponding to the real map.

[0064] For example, under the aforementioned current parameter settings, using the aforementioned obtained sampled spatial-spectrum dataset The dataset is used as a test set for model testing. By inputting this dataset into the target neural network after model training, the localization accuracy of the target neural network for the radiation source can be obtained, such as... Figure 6 As shown in the figure, the black area represents city buildings, the red-marked center represents the actual location of the radiation source, and the blue-marked center represents the location predicted by the model based on the spatial sampling points marked in green. Experiments on the test set demonstrate that the positioning accuracy of the method in this embodiment is an average of 4.03 pixels, meaning that within a 256m × 256m urban geographical area, the simulated positioning error is 4.03 meters.

[0065] Therefore, this embodiment proposes a general radiation source localization technology based on efficient neural networks. For lightweight electromagnetic spectrum sensing terminals, limited energy payload, limited computing power, limited operators that can be computed by the onboard computing chip, and the real-time requirements of the localization algorithm necessitate research into a general target localization technology based on efficient neural networks. This embodiment achieves a full-map sensing field by optimally selecting network parameters, thereby reducing computational overhead, lowering output latency, and improving the computational efficiency of the neural network model without excessively reducing localization accuracy. Furthermore, it employs more general-purpose operators, enabling the algorithm to be deployed on a wider range of computing platforms, thus providing possibilities for practical deployment.

[0066] The radiation source positioning method in the embodiments of the present application is described above, and the computer device in the embodiments of the present application is described below. Please refer to Figure 7 One embodiment of the computer device in the embodiments of the present application includes:

[0067] The first obtaining unit is configured to obtain a simulation data set of a city geographical environment, wherein the simulation data set includes a global path loss map of a simulation radiation source, a simulation grid image of the city geographical environment, and a preset spatial coordinate of the simulation radiation source.

[0068] The second obtaining unit is configured to obtain an initial neural network, and perform spatial random sampling on the global path loss map by using the initial neural network to obtain an electromagnetic information image of the city geographical environment.

[0069] The model training unit is configured to input the simulation grid image and the electromagnetic information image into the initial neural network, so that the initial neural network predicts the position of the radiation source in the city geographical environment according to the simulation grid image and the electromagnetic information image, and obtains the predicted position coordinate of the radiation source output by the initial neural network.

[0070] The model training unit is further configured to adjust the model parameters of the initial neural network according to the predicted position coordinate and the preset spatial coordinate, and stop the model training when the model training of the initial neural network reaches a convergence condition, to obtain a target neural network, wherein the target neural network is used to predict the position of the radiation source in the city geographical environment.

[0071] In a preferred embodiment of the present embodiment, the method further includes:

[0072] The third obtaining unit is configured to obtain a pre-acquired spatial spectrum data set of the city geographical environment, and obtain a real-acquired grid image of the city geographical environment, wherein the spatial spectrum data set includes a plurality of spatial spectrum data samples, and each spatial spectrum data sample includes a spatial coordinate of a sample of the city geographical environment and an electromagnetic spectrum signal receiving intensity of the city geographical environment.

[0073] The quantifying unit is configured to quantize the spatial spectrum data set and the real-acquired grid image in spatial data and intensity data respectively, to obtain a rasterized spatial spectrum data set corresponding to the spatial spectrum data set and a city geographical environment quantization image corresponding to the real-acquired grid image.

[0074] The testing unit is configured to input the rasterized spatial spectrum data set and the city geographical environment quantization image into the target neural network, so that the target neural network outputs a predicted radiation source position according to the rasterized spatial spectrum data set and the city geographical environment quantization image.

[0075] In a preferred implementation of the embodiment, the quantization unit is specifically configured to:

[0076] selecting a top view of a first preset size from the spatial spectrum dataset, and dividing the top view into a plurality of grids with a second preset size as a unit size to obtain a gridded spatial spectrum dataset represented by a gridded image;

[0077] wherein the second preset size is smaller than the first preset size.

[0078] In a preferred implementation of the embodiment, the quantization unit is specifically configured to:

[0079] treating the gray scale data of the image as intensity data of the image, and performing preset-bit quantization on the gray scale data of the spatial spectrum dataset to obtain the gridded spatial spectrum dataset.

[0080] In a preferred implementation of the embodiment, the model training unit is specifically configured to:

[0081] obtaining a conditional distribution of the predicted position coordinates of the radiation source of the initial neural network output, the conditional distribution being used to represent each of the predicted position coordinates and its corresponding confidence probability.

[0082] In the embodiment, the operations performed by each unit in the computer device are similar to those described in the foregoing Figure 1 embodiments and will not be described here again.

[0083] The computer device in the embodiments of the present application will be described below. Please refer to Figure 8 , one embodiment of the computer device in the embodiments of the present application includes:

[0084] The computer device 800 can include one or more central processing units (CPUs) 801 and a memory 805, and the memory 805 stores one or more application programs or data.

[0085] The memory 805 can be volatile storage or persistent storage. The programs stored in the memory 805 can include one or more modules, and each module can include a series of instruction operations in the computer device. Furthermore, the central processing unit 801 can be configured to communicate with the memory 805 and execute the series of instruction operations in the memory 805 on the computer device 800.

[0086] The computer device 800 can also include one or more power supplies 802, one or more wired or wireless network interfaces 803, one or more input / output interfaces 804, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0087] The central processing unit 801 can execute the operations performed by the computer device in the embodiments described above, which will not be described here in detail. Figure 1 The central processing unit 801 can execute the operations performed by the computer device in the embodiments described above, which will not be described here in detail.

[0088] The embodiments of the present application also provide a computer storage medium, one embodiment of which includes: instructions stored in the computer storage medium, which, when executed on a computer, cause the computer to perform the operations performed by the computer device in the embodiments described above. Figure 1 The central processing unit 801 can execute the operations performed by the computer device in the embodiments described above, which will not be described here in detail.

[0089] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, device and unit described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here in detail.

[0090] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division, and actual implementation can have another division manner. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0091] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. can be located in one place or distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0092] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0093] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, read-only memory), a random access memory (RAM, random access memory), a magnetic disk or an optical disk, and various media that can store program codes.

Claims

1. A method of positioning a radiation source, characterized by, The method comprises: obtaining a simulation data set of a city geographical environment, the simulation data set comprising a global path loss map of a simulation radiation source, a simulation grid image of the city geographical environment, and a preset spatial coordinate of the simulation radiation source; obtaining an initial neural network, using the initial neural network to perform spatial random sampling on the global path loss map to obtain an electromagnetic information image of the city geographical environment; inputting the simulation grid image and the electromagnetic information image into the initial neural network, so that the initial neural network predicts the position of the radiation source in the city geographical environment according to the simulation grid image and the electromagnetic information image, and obtains the predicted position coordinate of the radiation source output by the initial neural network; adjusting the model parameters of the initial neural network according to the predicted position coordinate and the preset spatial coordinate until the model training of the initial neural network reaches a convergence condition, and stopping the model training to obtain a target neural network, the target neural network being used to predict the position of the radiation source in the city geographical environment; The method further comprises: obtaining a pre-collected spatial spectrum data set of the city geographical environment, and obtaining a real-collected grid image of the city geographical environment, the spatial spectrum data set comprising a plurality of spatial spectrum data samples, each spatial spectrum data sample comprising a sample spatial coordinate of the city geographical environment and an electromagnetic spectrum signal receiving intensity of the city geographical environment; quantizing the spatial data and the intensity data of the spatial spectrum data set and the real-collected grid image respectively to obtain a rasterized spatial spectrum data set corresponding to the spatial spectrum data set and a city geographical environment quantization image corresponding to the real-collected grid image; inputting the rasterized spatial spectrum data set and the city geographical environment quantization image into the target neural network, so that the target neural network outputs a predicted radiation source position according to the rasterized spatial spectrum data set and the city geographical environment quantization image.

2. The method of claim 1, wherein, Quantizing the spatial data of the spatial spectrum data set comprises: selecting a first preset size of an overhead view in the spatial spectrum data set, and dividing the overhead view into a plurality of grids in the overhead view with a second preset size as a unit size to obtain a rasterized spatial spectrum data set represented by a rasterized image; wherein the second preset size is smaller than the first preset size.

3. The method of claim 1, wherein, Quantizing the intensity data of the spatial spectrum data set comprises: regarding the gray data of an image as the intensity data of the image, and quantizing the gray data of the spatial spectrum data set by a preset number of bits to obtain the rasterized spatial spectrum data set.

4. The method of claim 1, wherein, The method further comprises: obtaining a conditional distribution of the predicted position coordinate of the radiation source output by the initial neural network, the conditional distribution being used to represent each predicted position coordinate and its corresponding confidence probability.

5. A computer device, comprising: The computer device comprises: The first obtaining unit is configured to obtain a simulation dataset of a city geographical environment, the simulation dataset comprising a global path loss map of a simulation radiation source, a simulation grid image of the city geographical environment, and preset spatial coordinates of the simulation radiation source. The second obtaining unit is configured to obtain an initial neural network, and perform spatial random sampling on the global path loss map by using the initial neural network to obtain an electromagnetic information image of the city geographical environment. The model training unit is configured to input the simulation grid image and the electromagnetic information image into the initial neural network, so that the initial neural network predicts a position of a radiation source of the city geographical environment according to the simulation grid image and the electromagnetic information image, and obtains predicted position coordinates of the radiation source output by the initial neural network. The model training unit is further configured to adjust model parameters of the initial neural network according to the predicted position coordinates and the preset spatial coordinates, until the model training of the initial neural network reaches a convergence condition, and then stop the model training to obtain a target neural network, the target neural network being configured to predict a position of a radiation source in a city geographical environment. The computer device further comprises: The third obtaining unit is configured to obtain a pre-collected spatial spectrum dataset of the city geographical environment, and obtain a real-collected grid image of the city geographical environment, the spatial spectrum dataset comprising a plurality of spatial spectrum data samples, each spatial spectrum data sample comprising spatial coordinates of a sample of the city geographical environment and electromagnetic spectrum signal receiving intensity of the city geographical environment. The quantization unit is configured to quantize spatial data and intensity data of the spatial spectrum dataset and the real-collected grid image respectively to obtain a rasterized spatial spectrum dataset corresponding to the spatial spectrum dataset and a city geographical environment quantization image corresponding to the real-collected grid image. The test unit is configured to input the rasterized spatial spectrum dataset and the city geographical environment quantization image into the target neural network, so that the target neural network outputs a predicted radiation source position according to the rasterized spatial spectrum dataset and the city geographical environment quantization image.

6. The computer device of claim 5, wherein, The model training unit is specifically configured to: obtain a conditional distribution of the predicted position coordinates of the radiation source output by the initial neural network, the conditional distribution being configured to represent each predicted position coordinate and a corresponding confidence probability thereof. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor implements the method in any one of claims 1 to 4 when executing the computer program.

8. A computer storage medium, characterized in that The computer storage medium stores instructions, and the instructions cause the computer to execute the method in any one of claims 1 to 4 when executed on the computer. The computer storage medium stores instructions, and the instructions cause the computer to execute the method in any one of claims 1 to 4 when executed on the computer.

Images