Multi-frequency radio source modeling method based on hybrid expert model and multi-Gaussian fitting
By using a hybrid expert model and multi-Gaussian fitting method, the low efficiency and information fragmentation problems of multi-frequency data cubes of radio telescopes are solved, accurate three-dimensional modeling and feature analysis of radio sources are achieved, and the detection accuracy and recall rate are improved, which is suitable for multi-frequency data processing in radio astronomy.
Patent Information
- Application Number
- CN202411926954.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2044-12-25
AI Technical Summary
Existing technologies are inefficient and consume a lot of computing resources when processing multi-frequency data cubes of radio telescopes. They ignore the correlation between adjacent frequencies and have difficulty coping with the diverse characteristics of radio sources, resulting in performance degradation in complex scenarios. Traditional methods fail to fully consider the spatial and frequency distribution characteristics of radio sources.
A method based on a hybrid expert model and multi-Gaussian fitting is used to divide the multi-frequency data cube into sub-datasets. Multiple target detection methods are used to form a hybrid expert model. Combined with a deep convolutional neural network and a gated network, the weights are dynamically adjusted to output the three-dimensional coordinate information of the radio source, and the three-dimensional model is reconstructed through multi-Gaussian fitting.
It achieves accurate detection and three-dimensional modeling of multi-frequency radio sources, fully utilizes the correlation between frequencies, provides more comprehensive data information, improves detection accuracy and recall rate, and outputs reliable three-dimensional coordinates and multi-Gaussian parameters of radio sources, which is suitable for target detection and feature analysis of large-field-of-view radio observation data.
Smart Images
Figure CN119862772B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radio astronomy data cube processing, and in particular to a multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting. Background Art
[0002] Radio telescopes cover a wide frequency band, from hundreds of hertz to tens of gigahertz, enabling them to observe a rich array of radio information from celestial objects. This data is typically stored in a three-dimensional data cube, with two dimensions corresponding to spatial location and one corresponding to the observation frequency. The foreground sources (radio sources) in the data cube contain important astrophysical information. Different settings of radio telescope instrument parameters produce data cubes with varying numbers of frequencies. Information about radio sources at different frequencies within a data cube is highly correlated, such as spectral energy distribution or morphological characteristics under frequency variation. However, existing technologies have significant limitations in processing this type of data. Traditional target detection methods only search for targets (radio sources) in two-dimensional images of a single frequency. For multi-frequency data cubes, detection is performed only on a per-frequency basis, resulting in low efficiency and high computational resource consumption. Detection also ignores the correlation between adjacent frequencies, and detection metrics need to be improved. Existing target detection methods exhibit significant limitations when processing complex scenes (such as dense star fields or low signal-to-noise ratio backgrounds) and are easily affected by noise interference or background complexity. Furthermore, traditional target detection methods typically rely on a single detection model, making it difficult to cope with the diverse characteristics of radio sources. This results in significantly degraded performance in complex scenarios such as high density and extreme brightness (too bright or too dim). Traditional two-dimensional Gaussian fitting only fits radio sources in single-frequency two-dimensional images, failing to fully exploit the rich information in multi-frequency data. By splitting the three-dimensional data cube into multiple two-dimensional images for processing, information in the frequency dimension is fragmented, making it impossible to fully consider the overall spatial and frequency distribution characteristics of radio sources during fitting. This makes it difficult to accurately reveal the physical properties of radio sources, such as their spectral distribution or the evolution of their structure with frequency. Summary of the Invention
[0003] To address the shortcomings of existing technologies, the present invention aims to provide a multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting. This method combines multiple target detection methods with a hybrid expert model to effectively detect multi-frequency data cubes acquired by radio telescope arrays. Finally, based on the detection results, a comprehensive three-dimensional model of the radio source is constructed, enabling more relevant analysis of the relationships between frequencies within the source, facilitating subsequent scientific research and analysis.
[0004] To achieve the above object, the present invention provides the following technical solution, comprising the following steps:
[0005] A multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting is proposed. The multi-frequency data cube is divided into n parts according to a training strategy, and each part is sent to n target detection methods for independent training. The optimal model of each target detection method is combined to form a hybrid expert model. The hybrid expert model is used to train all data to generate the optimal model of the hybrid expert model. Target detection is performed on all radio data, and the three-dimensional coordinate information of the radio source is output. The multi-Gaussian fitting method is used to fit the multi-frequency data cube, and finally the output three-dimensional model is reconstructed. The method specifically includes the following steps:
[0006] Step 1. Construct a radio image annotation dataset: Based on the equipment parameters of the radio telescope array and the observation data, the radio image annotation dataset is generated using the Oskar simulation software. The radio image annotation dataset contains a simulated multi-frequency observation data cube, the spatial position and frequency distribution position of the target within the multi-frequency observation data cube;
[0007] Step 2. Generate the optimal model of the target detection method: Divide the radio image annotation dataset into n sub-datasets, train them using n target detection methods respectively, and obtain the optimal model for each target detection method;
[0008] Step 3. Generate the best model of the hybrid expert model: Use the best models of n target detection methods to train n sub-datasets respectively. During the training process, the hybrid expert model performs convolution and downsampling on the outputs of the best models of the n target detection methods obtained in step 2. Combined with the initialized gating network, dynamic weights are obtained and the output results are fused. Based on the comparison of the validation loss, the best model of the hybrid expert model is obtained and used to detect the multi-frequency observation data cube to obtain the category and location information of the multi-frequency observation data cube. The gating network dynamically adjusts the weight vector of the target detection method to improve the average precision, F1 score and recall rate of the hybrid expert model.
[0009] Step 4. Reconstruct the three-dimensional Gaussian model of the radio source: Based on the category and location information of the multi-frequency observation data cube obtained in step 3, cut the radio source into separate data cubes, construct an initial multi-Gaussian model, minimize the difference between the initial multi-Gaussian model and the data cube through the loss function and minimize function, and solve the optimal parameters of the multi-Gaussian model in the data cube; reconstruct the three-dimensional model of the radio source based on the optimal parameters of the multi-Gaussian model, and perform visualization and residual evaluation.
[0010] Furthermore, in step 1, based on the characteristics that different radio sources have different observable frequency ranges, a deep convolutional neural network model with added frequency labels is established to directly predict the three-dimensional coordinates of the output radio source. The feature extraction backbone network of the deep convolutional neural network is optimized, and the channel attention mechanism is added. The deep convolutional neural network outputs the spatial position and frequency information of the target.
[0011] Furthermore, in step 2, the radio source is divided into sub-datasets with distinct characteristics according to its distribution characteristics, brightness characteristics, and category characteristics. The number n of sub-datasets and target detection methods is determined. According to the characteristics of each sub-dataset, the n sub-datasets are trained using n target detection methods respectively. By repeatedly iteratively optimizing the network parameters, the optimal models of the n target detection methods are generated to obtain different feature information in different sub-datasets.
[0012] Furthermore, in step 3, the hybrid expert model includes a gating network and convolution and downsampling of the outputs of each target detection method to obtain dynamic weights and fusion output results; the gating network includes four fully connected layers and activation functions and dropout functions between layers. The gating network inputs the feature vector of the data cube, dynamically adjusts and outputs the normalized weight of each target detection method, and realizes weighted fusion of the outputs of different target detection methods.
[0013] Furthermore, in step 4, the data cube is processed with 3 sigma to remove outliers and noise, and then normalized to ensure the convergence speed; the multi-Gaussian fitting method traverses different optimization method parameters to solve the optimal multi-Gaussian model parameters.
[0014] In summary, the invention has the following beneficial effects:
[0015] Based on the traditional target detection neural network, the present invention combines multiple target detection methods to form a hybrid expert model, fully utilizing the advantages of different target detection methods to achieve accurate detection of multi-frequency radio sources; the present invention can simultaneously consider the correlation between different frequencies and provide the spatial information and frequency distribution information of the target.
[0016] This paper uses a multi-Gaussian fitting method to fit the three-dimensional targets detected by the hybrid expert model, finding the optimal multi-Gaussian parameters. These multi-Gaussian parameters combine spatial information and multi-frequency correlations to directly construct a three-dimensional model of the radio source. Compared to traditional two-dimensional spatial models and one-dimensional frequency models, the proposed three-dimensional model is more direct and comprehensive, more consistent with physical interpretation, and has smaller fitting residuals, providing more comprehensive data information for subsequent astronomical research.
[0017] The present invention performs real-time end-to-end radio source detection and modeling analysis on multi-frequency radio data cubes, outputting reliable three-dimensional coordinates of radio sources and important astronomical parameters of multi-Gaussian parameters, which are used for three-dimensional modeling and feature analysis of multi-frequency radio sources in radio astronomy data cubes. The method is particularly suitable for radio source target detection, frequency correlation extraction, and accurate multi-Gaussian fitting modeling in large-field-of-view radio observation data. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 Flowchart of the present invention.
[0019] Figure 2 This figure shows the mean absolute error (MAE) analysis results of the multi-Gaussian fit for a multi-frequency radio source with 96 frequencies. The horizontal axis shows the serial number of the multi-frequency radio source, with a total of 480 radio sources. The vertical axis shows the MAE of the multi-Gaussian fit for each radio source. DETAILED DESCRIPTION
[0020] The present invention will be described in further detail below with reference to the accompanying drawings.
[0021] It should be noted that, for the sake of convenience, the directions described below are consistent with the directions of the drawings themselves, but do not limit the structure of the present invention.
[0022] like Figure 1 As shown, the present invention discloses a multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting. On the basis of a single target detection method, multiple target detection methods are combined to form a hybrid expert model, which effectively detects the multi-frequency data cube obtained by the radio telescope array. Finally, based on the detection results, a comprehensive three-dimensional model of the three-dimensional radio source is established. The relationship between the frequencies in the radio source can be analyzed more relatively, thereby facilitating subsequent scientific research and analysis, such as the evolution of the radio source spectrum distribution or the radio source structure with frequency. Specifically, the method includes the following steps:
[0023] Step 1. Construct a radio image annotation dataset: Based on the radio telescope array's instrument parameters and observational data, the Oskar simulation software was used to generate a radio image annotation dataset. This dataset consists of a simulated multi-frequency observation data cube containing the spatial and frequency distribution of targets within the multi-frequency observation data cube. Based on the radio array's observation principles, observation parameter settings, and model, the Oskar simulation software was first used to import the GLEAM (GaLactic and Extragalactic All-sky MWA) catalog. A sky model was then constructed based on the RA (right ascension), DEC (declination), and Ref_freq (reference frequency) variables. The starting frequency and frequency increment for the radio source observation were set, generating data cubes containing the target's spatial location and frequency information. These data cubes constitute the annotation dataset. Radio sources observed in radio observations are only observable within specific frequency ranges, and the observable frequency ranges vary between different radio sources. To address this problem, a deep convolutional neural network model was developed. This model, which incorporates frequency labels in addition to the original input labels consisting solely of x and y coordinates, enables the deep convolutional neural network to directly predict the three-dimensional coordinates of the output radio source. To better capture the characteristic information of radio data at different frequencies and the correlation between these frequencies, the feature extraction backbone of the deep convolutional neural network was optimized as follows: First, the number of input channels of the first-layer convolution kernel was set to automatically match the number of frequencies (i.e., the number of channels) of the input data cube. Next, the number of output channels was adjusted based on the original input-output scaling factor. To avoid non-convergence during training due to excessive network parameters, a threshold for the number of convolution kernel input channels was set. A channel-attention mechanism was added to the original feature extraction backbone network to extract the frequency distribution of the data and the frequency information of each radio source. Specifically, the input feature map dimensions were adjusted to C×H×W, where C is the number of frequencies and H and W are the height and width, respectively. First, a global average pooling operation is performed on the input feature map of the deep convolutional neural network, performing average pooling in the spatial dimension to obtain a global average pooled feature vector of dimension C. This global average pooled feature vector is then input into the first fully connected layer to produce an intermediate feature vector, which also has a dimension of C. Subsequently, the intermediate feature vector is input into the second and third fully connected layers, enabling the deep convolutional network to learn to automatically match the complex relationship between position and frequency in the input data cube. Finally, two weight vectors of dimension C are obtained through the second and third fully connected layers, representing the frequency importance weight vector and the frequency suppression weight vector, respectively.The frequency importance weights are multiplied by the input feature map to obtain a weighted feature map. This weighted feature map is then fed into a sigmoid function to map the frequency suppression weight vector to a probability value within the range [0, 1]. Next, the frequency suppression weight vector is multiplied by the weighted feature map to obtain the final channel attention feature map, which is used to further extract feature information. Ultimately, the deep convolutional neural network output includes the spatial location coordinates and frequency information of the target.
[0024] Step 2. Generate the optimal model for each target detection method: Divide the radio image annotated dataset into n sub-datasets, train each of these sub-datasets using n target detection methods, and obtain the optimal model for each target detection method. Based on the distribution, brightness, and category characteristics of the radio source simulated previously, the datasets are divided into "dense star field" dataset, "sparse star field" dataset, "bright source dense" dataset, "dark source dense" dataset, "category 1" dataset, and "category 2" dataset. Since the computational resources required for the subsequent mixture of experts model are linearly proportional to the number of single target detection methods, the number of datasets and target detection methods is selected based on practical needs. The number of sub-datasets and target detection methods, n, is determined based on the availability of computational resources. These n sub-datasets are then fed into each of the n target detection methods for training. The single target detection methods used include YOLOX and YOLOv5. By matching each sub-dataset with a specific target detection method, the strengths of each target detection method are fully utilized. For dense star fields, target detection methods must accurately distinguish radio sources against complex backgrounds; for sparse star fields, target detection methods must more sensitively capture the characteristics of individual radio sources. Similarly, for bright source data, target detection methods need to avoid oversaturation and accurately identify high-brightness radio sources. For faint source data, however, target detection methods require higher sensitivity to capture faint signals. During training, single target detection methods iteratively optimize network parameters, ultimately generating corresponding optimal target detection method model files. These target detection method model files contain the optimal parameter weights for detecting radio source targets. This allows each target detection method to focus on its strengths in feature extraction and target detection during training, maximizing the potential for extracting radio source feature information from each sub-dataset and providing high-quality target detection method models for subsequent hybrid expert model training. During this process, each sub-dataset has a corresponding training model and method, and each model is trained independently, with each target detection method generating the optimal target detection method model.
[0025] Step 3. Generate the best model of the hybrid expert model: n sub-datasets are trained separately using the best models of n target detection methods. During the training process, the hybrid expert model performs convolution and downsampling on the outputs of the best models of the n target detection methods obtained in step 2. The outputs are combined with the initialized gating network to obtain dynamic weights and fuse the output results. Based on the comparison of the validation loss, the best model of the hybrid expert model is obtained and used to detect the multi-frequency observation data cube to obtain the category and location information of the multi-frequency observation data cube. The gating network dynamically adjusts the weight vector of the target detection method to improve the average precision, F1 score and recall rate of the hybrid expert model.
[0026] The gating network consists of four fully connected layers, with activation functions (ReLU / LeakyReLU) and dropout functions added between layers to improve generalization. The gating network input is the feature vector of the data cube. After multiple fully connected layers and dynamic adjustment through nonlinear transformations, the gating network outputs the normalized weights of each expert, achieving weighted fusion of the outputs of different target detection methods. The hybrid expert model contains an initialized gating network and convolution and downsampling of the outputs of each expert, i.e., the target detection method. Finally, the dynamic weight calculation and fusion output results are obtained. The n target detection model files and n sub-datasets from step 2 are loaded into the hybrid expert model at once for training. During training, each sub-dataset generates specific feature outputs through the corresponding target detection method (such as YOLOX, YOLOv5, etc.). The gating network (GatingNetwork) combines the feature information of the sub-dataset through the fully connected layer to generate a set of weights, with one weight corresponding to each sub-dataset. During this process, the gating network dynamically assigns weights to different experts (object detection methods). Specifically, for different feature distributions in the input data cube, such as object scale, shape complexity, or background noise level, certain object detection methods may perform better on specific metrics (such as detection accuracy, recall, or inference time). Based on the characteristics of the input data cube and the performance of each expert (object detection method) on these metrics, the gating network assigns higher weights to experts (object detection methods) that are more appropriate for the data characteristics, thereby achieving a weighted fusion of the outputs of different object detection methods. Through repeated iterative optimization, the hybrid expert model can gradually improve the detection capabilities of different object detection methods at different frequencies in different data cubes. After training, the hybrid expert model generates an optimal model file, which is used to detect the category and location information of the data cube. The gating network dynamically adjusts the vectors of expert (object detection method) weights. The size of the weight reflects the degree of adaptation of each expert (object detection method) to the current input, thereby controlling the contribution of each expert (object detection method) output. This not only detects the two-dimensional spatial position of targets in radio data (represented by detection boxes), but also analyzes and outputs frequency counts, revealing the correlations between different frequencies. Furthermore, this improves detection metrics. Specifically, the hybrid expert model significantly improves average precision (AP), F1 score, and recall compared to single target detection methods. Table 1 compares detection metrics between the single target detection method and the hybrid expert model method when score_threhold = 0.5.
[0027] Table 1
[0028] score_threhold=0.5 AP F1 P R yolox 62.14％ 0.72 96.48％ 56.84％ yolov5 62.52％ 0.73 98.57％ 58.00％ Mixture of Experts Model (MOE) 65.90％ 0.74 97.25％ 59.62％
[0029] Step 4. Reconstruct the three-dimensional Gaussian model of the radio source: Based on the category and location information of the multi-frequency observation data cube obtained in step 3, cut the radio source into separate data cubes, construct an initial multi-Gaussian model, minimize the difference between the initial multi-Gaussian model and the data cube through the loss function and minimize function, and solve the optimal parameters of the multi-Gaussian model in the data cube; reconstruct the three-dimensional model of the radio source based on the optimal parameters of the multi-Gaussian model, and perform visualization and residual evaluation.
[0030] Based on the two-dimensional coordinate position and frequency of the radio source detected by the hybrid expert model, the radio source is cut into separate data cubes, and the data cubes are processed with 3sigma to eliminate the influence of outliers and noise, and then normalized to ensure that the subsequent optimization method converges quickly. Then, an initial multi-Gaussian model is constructed from the data cube. The initial mean and covariance parameters are derived through the statistical characteristics of the data cube. During the fitting process, the minimize function in the scipy library will use the loss function and the built-in parameter optimization method provided by the minimize function to minimize the difference between the model and the data cube, thereby solving the optimal parameters of the multi-Gaussian model in the data cube. In addition, the multi-Gaussian fitting method will also traverse different optimization method parameters to solve the optimal multi-Gaussian model parameters. Then, based on the multi-Gaussian model parameters (multi-Gaussian mean and multi-Gaussian covariance), the three-dimensional model of the radio source is reconstructed, and visualization and residual evaluation are performed. Mean absolute error (MAE) evaluation, Figure 2 The experimental results show that the mean absolute error (MAE) of the multi-Gaussian fitting results of 469 out of 480 radio sources is within 0.01, and the evaluation results are reliable.
[0031] The present invention combines a loss function, initial multi-Gaussian parameters, and data with a library function to calculate the multi-Gaussian parameters with the lowest loss. This state is then determined to be the optimal multi-Gaussian parameters. Traditional two-dimensional fitting and one-dimensional frequency fitting fragment the three-dimensional radio sources in the data cube and fail to consider them as a whole. The multi-Gaussian fitting method of the present invention fits the spatial distribution of radio sources more comprehensively and directly than two-dimensional fitting and one-dimensional frequency fitting. The resulting multi-Gaussian model parameters are also of great significance for subsequent astronomical physical analysis.
[0032] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting, characterized in that: The multi-frequency data cube is divided into n parts according to the training strategy, and each part is sent to n target detection methods for independent training. The optimal model of each target detection method is combined to form a hybrid expert model. The hybrid expert model is used to train all data to generate the optimal model of the hybrid expert model. Target detection is performed on all radio data, and the three-dimensional coordinate information of the radio source is output. The multi-Gaussian fitting method is used to fit the multi-frequency data cube, and finally the output three-dimensional model is reconstructed. The specific steps include: Step 1. Construct a radio image annotation dataset: Based on the equipment parameters of the radio telescope array and the observation data, the radio image annotation dataset is generated using the Oskar simulation software. The radio image annotation dataset contains a simulated multi-frequency observation data cube, the spatial position and frequency distribution position of the target within the multi-frequency observation data cube; Step 2. Generate the optimal model of the target detection method: Divide the radio image annotation dataset into n sub-datasets, train them using n target detection methods respectively, and obtain the optimal model for each target detection method; Step 3. Generate the best model of the hybrid expert model: Use the best models of n target detection methods to train n sub-datasets respectively. During the training process, the hybrid expert model performs convolution and downsampling on the outputs of the best models of the n target detection methods obtained in step 2. Combined with the initialized gating network, dynamic weights are obtained and the output results are fused. Based on the comparison of the validation loss, the best model of the hybrid expert model is obtained and used to detect the multi-frequency observation data cube to obtain the category and location information of the multi-frequency observation data cube. The gating network dynamically adjusts the weight vector of the target detection method to improve the average precision, F1 score and recall rate of the hybrid expert model. Step 4. Reconstruct the three-dimensional Gaussian model of the radio source: Based on the category and location information of the multi-frequency observation data cube obtained in step 3, cut the radio source into separate data cubes, construct an initial multi-Gaussian model, minimize the difference between the initial multi-Gaussian model and the data cube through the loss function and minimize function, and solve the optimal parameters of the multi-Gaussian model in the data cube; reconstruct the three-dimensional model of the radio source based on the optimal parameters of the multi-Gaussian model, and perform visualization and residual evaluation.
2. The multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting according to claim 1, characterized in that: In step 1, based on the characteristic that the observable frequency ranges of different radio sources are different, a deep convolutional neural network model with added frequency labels is established to directly predict the three-dimensional coordinates of the output radio source. The feature extraction backbone network of the deep convolutional neural network is optimized, and a channel attention mechanism is added. The deep convolutional neural network outputs the spatial position and frequency information of the target.
3. The multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting according to claim 1, characterized in that: In step 2, the radio source is divided into sub-datasets with distinct characteristics according to its distribution characteristics, brightness characteristics, and category characteristics, and the number n of sub-datasets and target detection methods is determined. According to the characteristics of each sub-dataset, the n sub-datasets are trained using n target detection methods respectively. By repeatedly iteratively optimizing the network parameters, the optimal models of the n target detection methods are generated to obtain different feature information in different sub-datasets.
4. The multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting according to claim 1, characterized in that: In step 3, the hybrid expert model includes a gating network and convolution and downsampling of the outputs of each target detection method to obtain dynamic weights and fusion output results; The gating network consists of four fully connected layers and activation and dropout functions between layers. The gating network inputs the feature vector of the data cube, dynamically adjusts and outputs the normalized weight of each target detection method, and realizes the weighted fusion of the outputs of different target detection methods.
5. The multi-frequency radio source modeling method based on a hybrid expert model and multi-Gaussian fitting according to claim 1, characterized in that: In step 4, the data cube is processed with 3 sigma to remove outliers and noise, and then normalized to ensure convergence speed; The multi-Gaussian fitting method traverses different optimization method parameters to solve the optimal multi-Gaussian model parameters.