Lightweight radiation source positioning method

Abstract

The invention relates to a lightweight radiation source positioning method. The method comprises the following steps: acquiring a mask map containing building distribution information in a target area; drawing a grey-scale map containing the position information of each sampling point in the target area and the received signal intensity of each sampling point as an electromagnetic sampling map; an input image sample obtained after the mask map and the electromagnetic sampling map are fused is put into a lightweight radiation source positioning model, and radiation source position coordinates are obtained through analysis; the lightweight radiation source positioning model comprises a plurality of feature extraction modules used for performing multi-scale feature extraction based on residual connection and an attention mechanism on an input image sample; the feature fusion module is used for converting the multi-scale feature maps output by the different feature extraction modules into set sizes and then carrying out feature fusion; and the space mapping module is used for predicting the position coordinates of the radiation source according to the fusion features generated by the feature fusion module. The position of the radiation source can be effectively and accurately obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of radiation source positioning, and in particular to a lightweight radiation source positioning method. Background Art

[0002] Electromagnetic maps can intuitively present information such as electromagnetic energy distribution and spectrum resource distribution. Ideally, the distribution of electromagnetic signal strength can accurately map the number of radiation sources and their locations, thereby effectively assisting in the precise positioning of radiation sources. The main principles and shortcomings of existing positioning technologies are as follows:

[0003] 1. Time-based (ToA and TDoA) and angle-based (AoA) methods

[0004] ToA: Positioning is performed by measuring the arrival time of the received signal between the measurement terminal and the target radiation source and then converting it into distance. At least three measurement terminals are required to calculate the position of the target radiation source.

[0005] TDoA: Based on the distance difference between each measurement terminal and the target radiation source, the relative position of the target radiation source relative to each measurement terminal is inferred by solving a set of nonlinear hyperbolic equations, thereby performing positioning.

[0006] AoA: It mainly measures the arrival angle between the target radiation source and the measurement terminal. The ray formed with the measurement terminal as the starting point must pass through the target radiation source. The intersection of the two rays is the position of the target radiation source.

[0007] Disadvantages: High reliance on high-precision clocks and antenna arrays, and high hardware costs.

[0008] 2. Ranging-based methods

[0009] Main principle: A certain statistical signal attenuation or time delay model is needed to estimate the distance between the target radiation source and the measurement terminal.

[0010] Disadvantages: The complex urban environment cannot be accurately described by a deterministic physical model.

[0011] 3. Fingerprint-based methods

[0012] Main principle: Offline measurement is required to collect RSS measurements with location tags at multiple points in the target area to build a fingerprint database. The target is located by comparing the RSS measurements with the fingerprints in the database.

[0013] Disadvantages: Building a fingerprint database requires a lot of time and labor costs.

[0014] Therefore, a radiation source positioning method is needed to solve the problems existing in the prior art. Summary of the invention

[0015] The technical problem to be solved by the present invention is to provide a radiation source positioning method, which can effectively and accurately obtain the radiation source position.

[0016] The technical solution adopted by the present invention to solve the technical problem is: to provide a lightweight radiation source positioning method, comprising the following steps:

[0017] Get a mask map containing building distribution information in the target area;

[0018] Draw a grayscale image including the location information of each sampling point in the target area and the received signal strength thereof as an electromagnetic sampling map;

[0019] fusing the mask map and the electromagnetic sampling map to obtain an input image sample;

[0020] The input image sample is analyzed using a lightweight radiation source positioning model to obtain the radiation source position coordinates; the lightweight radiation source positioning model includes:

[0021] A plurality of feature extraction modules, wherein the plurality of feature extraction modules are connected in sequence and used to perform multi-scale feature extraction based on residual connection and attention mechanism on the input image sample;

[0022] A feature fusion module, used to convert the multi-scale feature maps output by the different feature extraction modules into a set size and then perform feature fusion;

[0023] The spatial mapping module is used to predict the position coordinates of the radiation source according to the fusion features generated by the feature fusion module.

[0024] Furthermore, the drawing of a grayscale image including the location information of each sampling point in the target area and the received signal strength thereof as an electromagnetic sampling map includes:

[0025] Convert the received signal strength into a grayscale value between 0 and 1;

[0026] The grayscale value of the pixel corresponding to the sampling point is set to the received signal strength at the corresponding position after conversion, and the grayscale value of the pixel in the remaining area is set to 0, so as to obtain an electromagnetic sampling map.

[0027] Furthermore, in the mask map, the grayscale value of the pixels in the building area is 1, and the grayscale value of the pixels in the non-building area is 0.

[0028] Furthermore, the feature extraction module includes a first convolution block and a second convolution block connected in sequence; after the input image sample is input into the first convolution block, the output features are processed in sequence by the convolution layer and batch normalization layer of the second convolution block to obtain a first feature map, the first feature map is fused with the input image sample through a residual connection to obtain a second feature map, and the second feature map is sequentially passed through an attention mechanism module and an activation function to output the multi-scale feature map.

[0029] Furthermore, the first convolution block includes a convolution layer, a batch normalization layer and an activation function connected in sequence, and the activation functions of the first convolution block and the second convolution block are both ReLU functions.

[0030] Furthermore, different feature extraction modules are connected via a maximum pooling layer.

[0031] Furthermore, the feature fusion module includes several adaptive average pooling layers and a dimensionality transformation layer; the adaptive average pooling layer corresponds to the feature extraction module one by one, and different multi-scale feature maps are input into the corresponding adaptive average pooling layer and then spliced ​​in the channel dimension, and the spliced ​​feature maps are processed by the dimensionality transformation layer and then output as the fused features.

[0032] Furthermore, the spatial mapping layer is a multi-layer perceptron.

[0033] Furthermore, the spatial mapping layer includes an input layer, a first hidden layer, a second hidden layer and an output layer, and the number of neurons in the input layer is the same as the number of features of the fusion feature.

[0034] Furthermore, the loss function of the lightweight radiation source positioning model is:

[0035]

[0036] Where W is the number of input image samples, InputMap i and are the i-th input image sample and the i-th training label, respectively, position (·) and Θ represent the lightweight radiation source positioning model and its model parameters respectively.

[0037] Beneficial Effects

[0038] Due to the adoption of the above technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention only needs to obtain the building distribution information, the location information of the sampling point and the received signal strength of the corresponding position, does not rely on high-precision clocks and antenna arrays, and does not need to build fingerprint data to obtain relatively accurate predicted coordinates, which is easy to implement and deploy and has low cost, so it has a wide range of application scenarios; the present invention uses the location information of the sampling point and the received signal strength of the corresponding position to generate an electromagnetic sampling map, and superimposes it with the mask map containing the building distribution information. The generated input image sample is analyzed and processed by a lightweight radiation source positioning model to obtain the radiation source coordinates. There is no need to abstract a complex physical model, and there are no special requirements for the implementation environment, which is particularly suitable for complex urban environments. The present invention directly uses sparse sampling values ​​for positioning, omits the electromagnetic map recovery step that is prone to errors, effectively reduces error transmission, and improves the reliability and accuracy of the positioning results. The computing resources occupied by the present invention during the operation process are significantly reduced, and whether it is CPU, GPU or memory resources, they can be more reasonably allocated and utilized. The reasoning speed of the present invention has been significantly improved, and the positioning task can be completed in a short time, meeting the application scenarios with high real-time requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of an embodiment of the present invention;

[0040] Figure 2 It is a structural schematic diagram of a lightweight radiation source positioning model according to an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of the structure of a feature extraction module according to an embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of the structure of a feature fusion module according to an embodiment of the present invention;

[0043] Figure 5 It is a schematic diagram of the structure of a space mapping module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0045] The embodiments of the present invention relate to a method for locating a lightweight radiation source, such as Figure 1 As shown, the following steps are included:

[0046] Get a mask map containing building distribution information in the target area;

[0047] Draw a grayscale image containing the location information of each sampling point in the target area as an electromagnetic sampling map;

[0048] The mask map and the electromagnetic sampling map are fused to obtain the input image sample;

[0049] The lightweight radiation source localization model is used to analyze the input image samples to obtain the radiation source position coordinates.

[0050] like Figure 2 As shown in the figure, the lightweight radiation source positioning model includes: Feature Extraction (FE) feature extraction module, which is a structure that combines residual connection and attention mechanism for feature extraction and enhancement; Feature Fusion (FF) feature fusion module, which is used to convert the output features of different levels of FE into output feature maps of fixed size and fuse them while retaining feature distribution information; Spatial Mapping (SM) spatial mapping module, which is used to map high-dimensional and complex feature space to a low-dimensional space that is more suitable for coordinate prediction tasks.

[0051] like Figure 3 As shown in the figure, the FE module consists of two convolution blocks, each of which contains a convolution layer, a batch normalization layer, and a ReLU activation function. In the second convolution block, the input features are convolved by a 1×1 convolution operation to match the output dimension of the batch normalization layer, and then added through a residual connection. The fused feature map passes through a CoordinateAttention (CA) and finally passes through a ReLU activation function. The model connects L FE modules in series, and each FE module is connected by a maximum pooling layer. The input of the first FE module is InputMap, and the input of the second FE module is InputMap. The input of the first FE module is The output of the FE module,

[0052] like Figure 4 As shown in Figure 1, the FF module consists of an adaptive average pooling layer (Adaptive Average Pool, AAP) and a dimensional transformation layer (Dimensional Transformation, DT). The FF module contains L AAPs, The input of the AAP is The output of the FE module, The output results of L AAPs are concatenated in the channel dimension and input into the DT layer, where the dimension is converted into a two-dimensional tensor as the input of the SM module.

[0053] like Figure 5 As shown in the figure, the SM module is a four-layer multilayer perceptron (MLP), including an input layer, two hidden layers, and an output layer. The number of neurons in the input layer is the same as the number of output features of the FF module. The number of neurons in the first hidden layer and the second hidden layer is usually adjusted based on the requirements of the specific task and experiments. The second hidden layer is usually less than the first hidden layer. The output layer is responsible for generating the final radiation source positioning result, and the number of neurons is 2.

[0054] This embodiment uses the following method to establish an image sample data set and conduct a simulation experiment:

[0055] Make size N x ×N y The building mask map BuildingMask is used to set the radiation source parameters, thereby simulating the electromagnetic map based on the received signal strength (RSS) in the urban environment. And save the radiation source coordinate label GroundTruth. Normalize BuildingMask and RadioMap, where the pixels corresponding to buildings in BuildingMask are 1, and the pixels corresponding to non-building areas are 0; the pixels corresponding to buildings in RadioMap are 1, and the pixels corresponding to non-building areas are between (0,1). According to BuildingMask, the simulation measurement terminal performs sparse sampling around the building to obtain the mask SampledMask representing the RSS sampling position, where the pixel corresponding to the sampling position is 1, and the remaining pixels are 0. According to SampledMask, retain the pixels at the sampling position on RadioMap, and set the remaining pixels to 0 to obtain the sampled map SampledMap;

[0056] The dataset Divided into training set and test set, where i is the sample number, M is the total number of samples, GroundTruth = (GT x ,GT y );

[0057] SampledMap i and BuildingMask i After conversion to tensor, it is concatenated in the channel dimension to obtain InputMap as the model input.

[0058] When training the model, the training set SampledMap i and BuildingMask i After conversion to tensor, concatenate in channel dimension to get InputMapi As the model input, the model is optimized by minimizing the Euclidean distance loss function. The Euclidean distance loss function is expressed as:

[0059]

[0060] Among them, W is the number of training samples, InputMap i and are the i-th sample and the i-th radiation source coordinate label, respectively, Map (·) and Θ represent the model and model parameters respectively. The trained model is the lightweight radiation source localization model.

[0061] In practical applications, the sampling data acquisition module can be deployed in the target area to obtain the location information of the sampling point and the received signal strength at the corresponding position, and the actual position of the radiation source can be used as a training label. The sampling data in the target area obtained by the sampling data acquisition module is then represented as a grayscale image. One pixel represents a sampling position or a position to be estimated. The pixel value at the sampling position is represented by different grayscale levels according to the strength of the received signal, and the pixel values ​​of the remaining positions are set to 0. A building mask map is generated based on the environmental image data of the target area. The corresponding pixels of the building in the figure are 1, and the corresponding pixels of the non-building area are 0. The image obtained by superimposing the electromagnetic sampling map and the building mask map is input into the trained model to obtain the prediction result of the radiation source coordinates.

Claims

1. A lightweight radiation source positioning method, It is characterized in that The following steps are involved: Get a mask map containing building distribution information in the target area; Draw a grayscale image including the location information of each sampling point in the target area and the received signal strength thereof as an electromagnetic sampling map; fusing the mask map and the electromagnetic sampling map to obtain an input image sample; Analyzing the input image sample using a lightweight radiation source positioning model to obtain radiation source position coordinates; The lightweight radiation source positioning model includes: A plurality of feature extraction modules, wherein the plurality of feature extraction modules are connected in sequence and used to perform multi-scale feature extraction based on residual connection and attention mechanism on the input image sample; A feature fusion module, used to convert the multi-scale feature maps output by the different feature extraction modules into a set size and then perform feature fusion; The spatial mapping module is used to predict the position coordinates of the radiation source according to the fusion features generated by the feature fusion module.

2. The method according to claim 1, It is characterized in that The drawing of a grayscale image including the location information of each sampling point in the target area and the received signal strength thereof as an electromagnetic sampling map comprises: Convert the received signal strength into a grayscale value between 0 and 1; The grayscale value of the pixel corresponding to the sampling point is set to the received signal strength at the corresponding position after conversion, and the grayscale value of the pixel in the remaining area is set to 0, so as to obtain an electromagnetic sampling map.

3. The method according to claim 1, It is characterized in that The grayscale value of the pixels in the building area in the mask map is 1, and the grayscale value of the pixels in the non-building area is 0.

4. The method according to claim 1, It is characterized in that The feature extraction module includes a first convolution block and a second convolution block connected in sequence; after the input image sample is input into the first convolution block, the output features are processed in sequence by the convolution layer and batch normalization layer of the second convolution block to obtain a first feature map, the first feature map is fused with the input image sample through a residual connection to obtain a second feature map, and the second feature map is sequentially passed through an attention mechanism module and an activation function to output the multi-scale feature map.

5. The method according to claim 4, It is characterized in that The first convolution block includes a convolution layer, a batch normalization layer and an activation function connected in sequence, and the activation functions of the first convolution block and the second convolution block are both ReLU functions.

6. The method according to claim 1, It is characterized in that The different feature extraction modules are connected via a maximum pooling layer.

7. The method according to claim 1, It is characterized in that The feature fusion module includes several adaptive average pooling layers and a dimension transformation layer; the adaptive average pooling layer corresponds to the feature extraction module one by one, and different multi-scale feature maps are input into the corresponding adaptive average pooling layer and then spliced ​​in the channel dimension, and the spliced ​​feature maps are processed by the dimension transformation layer and then output as the fused feature.

8. The method according to claim 1, It is characterized in that The spatial mapping layer is a multi-layer perceptron.

9. The method according to claim 8, It is characterized in that The spatial mapping layer includes an input layer, a first hidden layer, a second hidden layer and an output layer, and the number of neurons in the input layer is the same as the number of features of the fusion feature.

10. The method according to claim 1, It is characterized in that The loss function of the lightweight radiation source positioning model is: Where W is the number of input image samples, InputMap i and are the i-th input image sample and the i-th training label, respectively, position (·) and Θ represent the lightweight radiation source positioning model and its model parameters respectively.

Images