Method and apparatus for identifying radar returns based on artificial intelligence

Abstract

This application relates to a method and apparatus for identifying radar echoes based on artificial intelligence. The method includes: acquiring echo data from multiple consecutive scan times of a meteorological radar; preprocessing the data and constructing multidimensional radar echo features that simultaneously include spatial and temporal dimensions; inputting the multidimensional features into an artificial intelligence recognition model that includes spatial structure branches and temporal evolution branches for joint recognition; and fusing and evaluating the recognition results to output a stable result. This method can fully exploit the spatiotemporal evolution information of radar echoes, significantly improving the accuracy of automatic identification of complex echo types such as strong convective boundaries and gust fronts, as well as the stability of operational operations.

Description

Technical Field

[0001] This application relates to the fields of meteorological radar signal processing and artificial intelligence technology, and in particular to a method and apparatus for identifying radar echoes based on artificial intelligence. Background Technology

[0002] Weather radar is a key device for monitoring and issuing early warnings of severe weather. It detects meteorological targets such as precipitation and convective systems by emitting electromagnetic waves and receiving their scattered echoes on atmospheric targets. Radar echo data contains a wealth of meteorological information, and the automatic and accurate identification of its types is crucial for improving the automation level of meteorological monitoring and early warning capabilities.

[0003] With the development of artificial intelligence technology, especially the successful application of deep learning technology in image recognition, research has attempted to introduce deep learning models into the field of radar echo recognition. Existing solutions typically convert radar echo data into two-dimensional images, such as planar position indication maps, and use models like convolutional neural networks for feature extraction and classification, achieving a certain degree of automatic recognition. However, existing solutions still have significant shortcomings in practical applications. First, radar echoes are essentially time-varying electromagnetic signals. Existing methods mostly analyze two-dimensional images at a single moment, ignoring the temporal evolution characteristics of the echo during continuous scanning, such as boundary advancement and intensity changes, resulting in limited accuracy in recognizing rapidly evolving or structurally complex targets. Second, existing methods often rely on fixed rules or thresholds set by human experience, making it difficult to adapt to different radar systems, scanning strategies, and variable weather environments, resulting in poor generalization ability. Third, under complex weather backgrounds and noise interference, existing methods lack recognition stability, leading to high false alarm and false negative rates. Finally, existing solutions often require a high degree of human intervention for result verification or parameter adjustment, and their real-time processing capabilities and operational efficiency are insufficient to meet the needs of rapid early warning for severe weather.

[0004] Therefore, there is an urgent need to provide a recognition method that can jointly model the multi-dimensional features of radar echoes, fully explore the effective information of radar echoes in multiple dimensions such as time and space, and construct an adaptive and robust intelligent recognition method to achieve automatic, accurate and stable recognition of radar echo types and improve operational efficiency. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and apparatus for identifying radar echoes based on artificial intelligence, which can fully utilize the spatiotemporal multidimensional characteristics of radar echoes to achieve automatic, accurate, and stable identification, in response to the above-mentioned technical problems.

[0006] A method for identifying radar echoes based on artificial intelligence, the method comprising:

[0007] Acquire echo data from a weather radar over T consecutive scan times; where T is an integer greater than 1. The echo data from the T consecutive scan times are preprocessed to obtain the preprocessed time-series echo sequence. Based on the preprocessed time-series echo sequence, a multidimensional radar echo feature is constructed; wherein, the multidimensional radar echo feature is composed of multiple feature tensor channels, the feature tensor channels include a spatial dimension feature channel for describing the spatial distribution structure of the echo and a temporal dimension feature channel for describing the change law of the echo during continuous scanning time. The multidimensional radar echo features are input into a pre-set artificial intelligence recognition model. The artificial intelligence recognition model identifies the multidimensional radar echo features and outputs the radar echo type identification result and corresponding confidence level. The artificial intelligence recognition model includes a spatial structure branch for processing the spatial dimension feature channel and a temporal evolution branch for processing the temporal dimension feature channel. The spatial features and temporal features are fused through a fusion layer. The output recognition results are fused and confidence is evaluated to obtain a stable final recognition result.

[0008] In one embodiment, preprocessing of the echo data from the T consecutive scan times includes: Based on the sliding window background estimation method, calculate the first... Echo background estimates at each scan time :

[0009] Among them, D t-k:t-1 Let represent the set of echo data from the (tk)th scan time to the (t-1)th scan time, where t represents the time sequence index of the current scan time, k represents the sliding window length, and k is an integer greater than 1; The first Raw echo data at each scan time Subtract the estimated echo background value Echo data after suppressing static background was obtained. .

[0010] In one embodiment, constructing the multidimensional radar echo features includes: A multi-channel feature tensor is constructed for the preprocessed temporal echo sequence; the channels of the multi-channel feature tensor include at least two of the following: the original echo intensity channel composed of the preprocessed temporal echo sequence, the inter-frame difference channel composed of the difference between echo data at adjacent scan times, the time stacking channel composed of echo data at multiple scan times stacked along the channel dimension, the gradient channel composed of the spatial gradient of the echo data, and the statistical summary channel composed of the statistical features within the scan window.

[0011] In one embodiment, the spatial structure branch is used to extract spatial structure feature vectors from the spatial dimension feature channels. ; The time evolution branch is used to extract time evolution feature vectors from the time dimension feature channels. ; The fusion layer is used to integrate the spatial structure feature vector. With the time evolution feature vector By fusing the features, a joint feature vector is obtained. .

[0012] In one embodiment, the fusion layer achieves feature fusion through a fusion mapping function as follows:

[0013] in ( ) represents a fusion mapping.

[0014] In one embodiment, the output recognition results are subjected to result fusion and confidence evaluation processing, including: Consistency fusion is performed on the recognition results of the same spatial region obtained at N consecutive scanning times or through M overlapping recognition windows; the consistency fusion adopts the majority voting or confidence weighted average method, where N and M are both integers greater than 1; The confidence level of the fused recognition result is calibrated, and the final recognition result is output based on a preset trigger threshold.

[0015] In one embodiment, the method further includes an online self-calibration step: Collect confirmed radar echo identification samples during business operations to form a feedback sample library; Based on the returned sample library, the artificial intelligence recognition model is periodically retrained or incrementally updated.

[0016] A device for identifying radar echoes based on artificial intelligence, the device comprising: The data acquisition module is used to acquire echo data from the weather radar over T consecutive scan times, where T is an integer greater than 1. The preprocessing module is used to perform signal-level preprocessing on the echo data of the T consecutive scanning times to suppress non-meteorological interference and extract time-series change information, so as to obtain the preprocessed time-series echo sequence. The feature construction module is used to construct multidimensional radar echo features based on the preprocessed time-series echo sequence. The multidimensional radar echo features include at least spatial dimension features and temporal dimension features. The intelligent recognition module is used to input the multi-dimensional radar echo features into the artificial intelligence recognition model to jointly model and recognize the multi-dimensional radar echo features, and output the radar echo type recognition result and corresponding confidence level. The fusion evaluation module is used to perform result fusion and confidence evaluation on the radar echo type identification results to obtain a stable final identification result. The business output module is used to perform business-oriented post-processing on the final recognition result, and generate and output target recognition information suitable for the business system. The self-updating module is used to perform online self-calibration and incremental updates on the artificial intelligence recognition model based on the confirmation samples collected during business operations.

[0017] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps: Acquire echo data from a weather radar over T consecutive scan times; where T is an integer greater than 1. The echo data from the T consecutive scan times are preprocessed to obtain the preprocessed time-series echo sequence. Based on the preprocessed time-series echo sequence, a multidimensional radar echo feature is constructed; wherein, the multidimensional radar echo feature is composed of multiple feature tensor channels, the feature tensor channels include a spatial dimension feature channel for describing the spatial distribution structure of the echo and a temporal dimension feature channel for describing the change law of the echo during continuous scanning time. The multidimensional radar echo features are input into a pre-set artificial intelligence recognition model. The artificial intelligence recognition model identifies the multidimensional radar echo features and outputs the radar echo type identification result and corresponding confidence level. The artificial intelligence recognition model includes a spatial structure branch for processing the spatial dimension feature channel and a temporal evolution branch for processing the temporal dimension feature channel. The spatial features and temporal features are fused through a fusion layer. The output recognition results are fused and confidence is evaluated to obtain a stable final recognition result.

[0018] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Acquire echo data from a weather radar over T consecutive scan times; where T is an integer greater than 1. The echo data from the T consecutive scan times are preprocessed to obtain the preprocessed time-series echo sequence. Based on the preprocessed time-series echo sequence, a multidimensional radar echo feature is constructed; wherein, the multidimensional radar echo feature is composed of multiple feature tensor channels, the feature tensor channels include a spatial dimension feature channel for describing the spatial distribution structure of the echo and a temporal dimension feature channel for describing the change law of the echo during continuous scanning time. The multidimensional radar echo features are input into a pre-set artificial intelligence recognition model. The artificial intelligence recognition model identifies the multidimensional radar echo features and outputs the radar echo type identification result and corresponding confidence level. The artificial intelligence recognition model includes a spatial structure branch for processing the spatial dimension feature channel and a temporal evolution branch for processing the temporal dimension feature channel. The spatial features and temporal features are fused through a fusion layer. The output recognition results are fused and confidence is evaluated to obtain a stable final recognition result.

[0019] The aforementioned method and apparatus for identifying radar echoes based on artificial intelligence (AI) acquires radar echo data from multiple consecutive scanning moments, changing the traditional processing mode that treats echoes as static images and ensuring the integrity of the dynamic evolution information of the echoes. Based on this, a multi-dimensional feature model integrating spatial structure and temporal variation is constructed, making the representation information more comprehensive and directly overcoming the shortcomings of existing methods, such as low recognition rate and insensitivity to rapidly changing targets due to insufficient feature utilization. Furthermore, an AI recognition model employing a dedicated spatial structure branch and a temporal evolution branch can deeply extract the morphological features and evolutionary patterns of the echoes, and achieve adaptive fusion and joint discrimination of the two within the model, thereby significantly improving the ability to distinguish and recognize complex echoes with varied morphologies and time-series dependence. In addition, by performing multi-moment consistency fusion and confidence calibration on the recognition results, random errors in single-inference are effectively suppressed, significantly reducing false alarms and false negatives, ensuring the stability and reliability of the system's output results during long-term operational use. Finally, combined with an online self-calibration mechanism, the system can continuously adapt to environmental changes, achieving a closed loop from algorithmic innovation to a stable and usable operational system. Overall, this solution systematically addresses the challenges of accuracy, stability, and adaptability in radar echo identification through multi-level collaborative innovation, from data representation and model architecture to engineering deployment, providing efficient and reliable automated technical support for meteorological monitoring and early warning operations. Attached Figure Description

[0020] Figure 1 This is an application scenario diagram of an artificial intelligence-based radar echo identification method in one embodiment; Figure 2 This is a flowchart illustrating a method for identifying radar echoes based on artificial intelligence in one embodiment; Figure 3 This is a structural block diagram of a device for identifying radar echoes based on artificial intelligence in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] The method for identifying radar echoes based on artificial intelligence provided in this application can be applied to, for example... Figure 1 In the application environment shown, the meteorological radar station 102 transmits the echo data acquired through continuous scanning to the data processing server 104 via a network. The data processing server 104 executes the method described in this application, performing preprocessing, multi-dimensional feature construction, intelligent recognition, and result fusion on the echo data, and finally outputs the recognition results to a terminal display device or early warning system via the network. The data processing server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers. The terminal display device can be, but is not limited to, various personal computers, workstations, professional meteorological displays, etc.

[0023] In one embodiment, such as Figure 2 As shown, a method for identifying radar echoes based on artificial intelligence is provided, which can be applied to... Figure 1 Taking the server in the example, the following steps are included: Step 202: Obtain echo data from the weather radar at T consecutive scanning times; where T is an integer greater than 1.

[0024] Specifically, T represents the number of consecutive scan moments used for joint analysis. A scan moment refers to the time point at which the radar completes one volume scan or scan sector according to a preset cycle. The echo data includes at least the radar reflectivity factor field, and may also include other physical quantities and scan parameters. The raw echo data set output in this step can be represented as... ,in Indicates the first Echo data acquired at each scan time. By acquiring data from multiple consecutive time points, a foundation is laid for subsequent extraction of temporal evolution features.

[0025] Step 204: Preprocess the echo data from T consecutive scan times to obtain the preprocessed time-series echo sequence.

[0026] Specifically, non-meteorological interference in radar echoes, such as ground clutter, random noise, and isolated spikes, is suppressed to a controllable range, while preserving the structural boundaries and weak echo details of meteorological echoes, avoiding information loss due to excessive smoothing. Preprocessing includes, but is not limited to: range-related gain compensation and intensity normalization, clutter suppression based on morphological filtering or statistical thresholds, inter-frame differencing to highlight temporal variations, and handling of missing and outlier values.

[0027] Step 206: Construct multidimensional radar echo features based on the preprocessed time-series echo sequence; wherein, the multidimensional radar echo features consist of multiple feature tensor channels, including spatial dimension feature channels for describing the spatial distribution structure of the echo and temporal dimension feature channels for describing the change law of the echo during continuous scanning.

[0028] Specifically, the preprocessed time-series echo sequence is converted into a data structure that can simultaneously carry spatial structure and temporal evolution information. First, the data from T consecutive time points are spatiotemporally aligned, for example, by projecting all time-series data onto a unified geographic coordinate grid. Then, centered on the region to be analyzed, a fixed-size local window is cropped from the data at each time point to form a sample. Where X represents a local temporal sample cropped from echo data at spatial location c over T consecutive scan times. This sample X is represented as a multi-channel feature tensor for subsequent feature extraction and recognition processing. P( ) represents a local block cropped at center point c. Finally, for each sample X, a multi-channel feature tensor is constructed, wherein the multi-channel feature tensor contains at least two or more of the following channels: spatial dimension feature channel, temporal dimension feature channel, gradient channel, and statistical summary channel.

[0029] Step 208: Input the multidimensional radar echo features into the pre-set artificial intelligence recognition model, identify the multidimensional radar echo features through the artificial intelligence recognition model, and output the radar echo type identification result and corresponding confidence level; the artificial intelligence recognition model includes the spatial structure branch for processing the spatial dimension feature channel and the temporal evolution branch for processing the temporal dimension feature channel, and fuses the spatial features and temporal features through a fusion layer.

[0030] Specifically, the AI ​​recognition model is a trained multi-branch neural network model, whose core structure includes a spatial structure branch, a temporal evolution branch, and a fusion layer. The spatial structure branch, such as a convolutional neural network (CNN) receiving multi-channel tensors, is mainly used to extract the spatial morphological features of the echo from the spatial dimension feature channels. The temporal evolution branch, such as a long short-term memory network (LSTM), a gated recurrent unit network (GRU), or an attention-based temporal encoder, mainly extracts the evolutionary features of the echo as the scan time changes from the temporal dimension feature channels. The fusion layer fuses the spatial feature vector and the temporal feature vector through a fusion mapping function to obtain the final discriminative features, a joint feature vector. Finally, the joint feature vector is processed by a classification layer to output the probability distribution of the radar echo type, i.e., the recognition result and its corresponding confidence level. Recognition types can include strong convective boundaries, gust fronts, weak echo bands, general precipitation echoes, and non-meteorological echoes.

[0031] Step 210: Perform result fusion and confidence evaluation on the output recognition results to obtain a stable final recognition result.

[0032] Specifically, since single-step model inference may fluctuate due to noise interference, an ensemble strategy is employed to improve output stability. First, a consistent fusion strategy is used. When multiple overlapping windows exist in the same region, or when repeated identification occurs at consecutive time points, their spatiotemporal continuity is utilized for fusion. For example, multi-window voting and confidence-weighted averaging are used to smooth the output at consecutive time points. Then, the confidence level of the fused result is calibrated and judged based on a preset trigger threshold θ. For example, the final recognition result is only output when the final confidence level is higher than θ; results with confidence levels in the uncertain range are marked as requiring manual review or delayed confirmation. This effectively filters out occasional misjudgments, ensuring the continuity and reliability of the output results and meeting operational requirements.

[0033] The aforementioned method and apparatus for identifying radar echoes based on artificial intelligence fundamentally changes the traditional processing mode of treating echoes as static images by acquiring radar echo data from multiple consecutive scanning moments, ensuring the integrity of the dynamic evolution information of the echoes. Based on this, a multi-dimensional feature integrating spatial structure and temporal variation is constructed, making the representation information more comprehensive and directly overcoming the shortcomings of existing methods, such as low recognition rate and insensitivity to rapidly changing targets due to insufficient feature utilization. Furthermore, an artificial intelligence recognition model employing a dedicated spatial structure branch and a temporal evolution branch can deeply extract the morphological features and evolutionary patterns of the echoes, and achieve adaptive fusion and joint discrimination of the two within the model, thereby significantly improving the ability to distinguish and recognize complex echoes with varied morphologies and time-series dependence. In addition, by performing multi-moment consistency fusion and confidence calibration on the recognition results, random errors in single-inference are effectively suppressed, significantly reducing false alarms and false negatives, ensuring the stability and reliability of the system's output results during long-term operational use. Finally, combined with an online self-calibration mechanism, the system can continuously adapt to environmental changes, achieving a closed loop from algorithmic innovation to a stable and usable operational system. Overall, this solution systematically addresses the challenges of accuracy, stability, and adaptability in radar echo identification through multi-level collaborative innovation, from data representation and model architecture to engineering deployment, providing efficient and reliable automated technical support for meteorological monitoring and early warning operations.

[0034] In one embodiment, preprocessing of echo data from T consecutive scan times includes: Based on the sliding window background estimation method, calculate the first... Echo background estimates at each scan time :

[0035] Among them, D t-k:t-1 Let represent the set of echo data from the (tk)th scan time to the (t-1)th scan time, where t represents the time series index of the current scan time, and k represents the sliding window length, where k is an integer greater than 1; then the Raw echo data at each scan time Subtract the estimated echo background value Echo data after suppressing static background was obtained. .

[0036] In practice, by calculating the sliding window value as a background estimate, static interference such as long-term stable ground clutter can be effectively filtered out, while meteorological echo signals are preserved. Compared with the simple thresholding method, this method is more adaptable to non-stationary backgrounds and can more accurately highlight effective echo targets that change over time, providing a cleaner data foundation for subsequent time-series feature extraction.

[0037] In one embodiment, constructing multidimensional radar echo features includes constructing a multi-channel feature tensor for the preprocessed time-series echo sequence; the channels of the multi-channel feature tensor include at least two of the following: an original echo intensity channel composed of the preprocessed time-series echo sequence, an inter-frame difference channel composed of the difference between echo data at adjacent scanning times, a time stacking channel composed of echo data at multiple scanning times stacked along the channel dimension, a gradient channel composed of the spatial gradient of the echo data, and a statistical summary channel composed of the statistical features within the scanning window.

[0038] In practical implementation, information such as the "spatial structure + temporal evolution + intensity gradient / boundary morphology" of radar echoes is explicitly expressed as multi-channel input to avoid information loss caused by a single image. For each sample X, the multi-channel feature tensor contains two or more of the following channels: Original echo intensity channel ; Inter-frame differential channels ; Time-stacked channels (stack t=1,…,T according to channel dimensions); Boundary / Gradient Channel ; Statistical summary channels (maximum value, mean, variance, quantiles, etc. within the window); Final output sample This forms a multi-channel tensor input. Here, H and W are the local window sizes, and C is the number of channels.

[0039] In one embodiment, the spatial structure branch is used to extract spatial structure feature vectors from the spatial dimension feature channels. The time evolution branch is used to extract time evolution feature vectors from the time dimension feature channels. The fusion layer is used to combine spatial structure feature vectors. With the time evolution feature vector By fusing the features, a joint feature vector is obtained. .

[0040] In practical implementation, the model branch used to extract the spatial structural features of radar echo distribution can be replaced by artificial intelligence models with different structural forms, such as different numbers of layers, different convolutional structures, or other neural network structures capable of extracting spatial features, rather than being limited to a specific implementation. The model branch used to extract the temporal evolution features of radar echo changes between consecutive scanning moments can employ different temporal modeling methods, such as models based on recurrent structures, attention mechanisms, or temporal convolution. The fusion of spatial and temporal features can be achieved through various methods such as concatenation, weighted summation, attention-weighted fusion, or gated fusion, enabling joint discrimination of multi-dimensional features.

[0041] In one embodiment, the fusion layer achieves feature fusion through a fusion mapping function as follows:

[0042] in ( ) represents a fusion mapping.

[0043] In practice, This approach involves concatenating two feature vectors along their respective feature dimensions. This split-then-merge structure allows the spatial and temporal branches to focus on learning their respective strengths, avoiding interference between features. By applying nonlinear transformations and dimensionality reduction to the concatenated features through a fusion layer, the complex interactions and collaborative discrimination patterns between spatial and temporal features can be learned, generating a more discriminative joint feature representation—a key to achieving accurate classification.

[0044] In one embodiment, the output recognition results are subjected to result fusion and confidence evaluation processing, including consistency fusion of recognition results obtained from the same spatial region at N consecutive scanning times or through M overlapping recognition windows; consistency fusion adopts majority voting or confidence weighted average method, where N and M are both integers greater than 1; the confidence of the fused recognition results is calibrated, and whether to output the final recognition result is determined according to a preset trigger threshold.

[0045] In practical implementation, a continuous consistency judgment is adopted, such as requiring three consecutive identifications of the same category, which can effectively filter out occasional misidentifications caused by instantaneous noise or model fluctuations. Multi-window voting utilizes information from spatially adjacent regions for cross-validation, improving the robustness of identifying irregularly shaped or vaguely defined targets. Combined with confidence threshold calibration, a double-insurance mechanism is constructed, ensuring that the final output to the operational system is highly reliable and stable, greatly reducing the probability of false alarms and meeting the stringent reliability requirements of meteorological early warning operations.

[0046] In one embodiment, an online self-calibration step is also included: collecting confirmed radar echo identification samples during business operations to form a backflow sample library; and periodically retraining or incrementally updating the artificial intelligence identification model based on the backflow sample library.

[0047] In practical implementation, the system can gradually collect valid samples confirmed by business personnel or corrected samples into the backlog sample library during long-term operation. Within a preset period, or when model performance monitoring indicators fluctuate, the deployed artificial intelligence recognition model can be incrementally updated or its parameters fine-tuned based on new samples. In this way, the system can adapt to seasonal climate changes, differences in data characteristics from newly added observation stations, and slow changes in data distribution over long-term operation, thereby helping to maintain the continuous availability and stability of recognition performance.

[0048] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0049] In one embodiment, such as Figure 3 As shown, an artificial intelligence-based radar echo recognition device is provided, comprising: a data acquisition module 302, a preprocessing module 304, a feature construction module 306, an intelligent recognition module 308, a fusion evaluation module 310, a service output module 312, and a self-updating module 314, wherein: The data acquisition module 302 is used to acquire echo data of the weather radar at T consecutive scanning times, where T is an integer greater than 1.

[0050] The preprocessing module 304 is used to perform signal-level preprocessing on the echo data of the T consecutive scanning times to suppress non-meteorological interference and extract time-series change information to obtain the preprocessed time-series echo sequence.

[0051] The feature construction module 306 is used to construct multidimensional radar echo features based on the preprocessed time-series echo sequence. The multidimensional radar echo features include at least spatial dimension features and temporal dimension features.

[0052] The intelligent recognition module 308 is used to input the multi-dimensional radar echo features into the artificial intelligence recognition model to jointly model and recognize the multi-dimensional radar echo features, and output the radar echo type recognition result and corresponding confidence level.

[0053] The fusion evaluation module 310 is used to perform result fusion and confidence evaluation on the radar echo type identification results to obtain a stable final identification result.

[0054] The business output module 312 is used to perform business-oriented post-processing on the final recognition result, and generate and output target recognition information suitable for the business system.

[0055] The self-updating module 314 is used to perform online self-calibration and incremental updates on the artificial intelligence recognition model based on the confirmation samples collected during business operations.

[0056] In one embodiment, the preprocessing module 304 is further configured to preprocess the echo data over T consecutive scan times, including: Based on the sliding window background estimation method, calculate the first... Echo background estimates at each scan time :

[0057] Among them, D t-k:t-1 Let represent the set of echo data from the (tk)th scan time to the (t-1)th scan time, where t represents the time sequence index of the current scan time, k represents the sliding window length, and k is an integer greater than 1; The first Raw echo data at each scan time Subtract the estimated echo background value Echo data after suppressing static background was obtained. .

[0058] In one embodiment, the feature construction module 306 is further configured to construct multidimensional radar echo features, including: A multi-channel feature tensor is constructed for the preprocessed temporal echo sequence. The channels of the multi-channel feature tensor include at least two of the following: the original echo intensity channel composed of the preprocessed temporal echo sequence, the inter-frame difference channel composed of the difference between echo data at adjacent scan times, the time stacking channel composed of the echo data at multiple scan times stacked along the channel dimension, the gradient channel composed of the spatial gradient of the echo data, and the statistical summary channel composed of the statistical features within the scan window.

[0059] In one embodiment, the feature construction module 306 is further configured for spatial structure branching to extract spatial structure feature vectors from spatial dimension feature channels. ; The temporal evolution branch is used to extract temporal evolution feature vectors from the temporal dimension feature channels. ; The fusion layer is used to combine spatial structure feature vectors With time evolution eigenvectors By fusing the features, a joint feature vector is obtained. .

[0060] In one embodiment, the intelligent recognition module 308 is further configured to achieve feature fusion through a fusion mapping function in the fusion layer:

[0061] in ( ) represents a fusion mapping.

[0062] In one embodiment, the fusion evaluation module 310 is further configured to perform result fusion and confidence evaluation processing on the output recognition results, including: Consistency fusion is performed on the recognition results of the same spatial region obtained at N consecutive scanning times or through M overlapping recognition windows; consistency fusion adopts majority voting or confidence weighted average method, where N and M are both integers greater than 1; The confidence level of the fused recognition result is calibrated, and the final recognition result is output based on a preset trigger threshold.

[0063] In one embodiment, the self-updating module 314 is also used for the online self-calibration step: Collect confirmed radar echo identification samples during business operations to form a feedback sample library; Based on the returned sample library, the artificial intelligence recognition model is periodically retrained or incrementally updated.

[0064] Specific limitations regarding the device for identifying radar echoes based on artificial intelligence can be found in the limitations of the method for identifying radar echoes based on artificial intelligence mentioned above, and will not be repeated here. Each module in the aforementioned device for identifying radar echoes based on artificial intelligence can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0065] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores historical radar echo data, a backflow sample library, and model parameters. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements an artificial intelligence-based method for identifying radar echoes.

[0066] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0067] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described above.

[0068] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0069] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identifying radar echoes based on artificial intelligence, characterized in that, The method includes: Acquire echo data from a weather radar over T consecutive scan times; where T is an integer greater than 1. The echo data from the T consecutive scan times are preprocessed to obtain the preprocessed time-series echo sequence. Based on the preprocessed time-series echo sequence, a multidimensional radar echo feature is constructed; wherein, the multidimensional radar echo feature is composed of multiple feature tensor channels, the feature tensor channels include a spatial dimension feature channel for describing the spatial distribution structure of the echo and a temporal dimension feature channel for describing the change law of the echo during continuous scanning time. The multidimensional radar echo features are input into a pre-set artificial intelligence recognition model. The artificial intelligence recognition model identifies the multidimensional radar echo features and outputs the radar echo type identification result and corresponding confidence level. The artificial intelligence recognition model includes a spatial structure branch for processing the spatial dimension feature channel and a temporal evolution branch for processing the temporal dimension feature channel. The spatial features and temporal features are fused through a fusion layer. The output recognition results are fused and confidence is evaluated to obtain a stable final recognition result.

2. The method according to claim 1, characterized in that, Preprocessing of the echo data from the T consecutive scan times includes: Based on the sliding window background estimation method, calculate the first... Echo background estimates at each scan time : ； Among them, D t-k:t-1 Let represent the set of echo data from the (tk)th scan time to the (t-1)th scan time, where t represents the time sequence index of the current scan time, k represents the sliding window length, and k is an integer greater than 1; The first Raw echo data at each scan time Subtract the estimated echo background value Echo data after suppressing static background was obtained. .

3. The method according to claim 1, characterized in that, The construction of multi-dimensional radar echo features includes: A multi-channel feature tensor is constructed for the preprocessed temporal echo sequence; the channels of the multi-channel feature tensor include at least two of the following: the original echo intensity channel composed of the preprocessed temporal echo sequence, the inter-frame difference channel composed of the difference between echo data at adjacent scan times, the time stacking channel composed of echo data at multiple scan times stacked along the channel dimension, the gradient channel composed of the spatial gradient of the echo data, and the statistical summary channel composed of the statistical features within the scan window.

4. The method according to claim 1, characterized in that, The spatial structure branch is used to extract spatial structure feature vectors from the spatial dimension feature channels. ; The time evolution branch is used to extract time evolution feature vectors from the time dimension feature channels. ; The fusion layer is used to integrate the spatial structure feature vector. With the time evolution feature vector By fusing the features, a joint feature vector is obtained. .

5. The method according to claim 4, characterized in that, The fusion layer achieves feature fusion through a fusion mapping function as follows: in ( ) represents a fusion mapping.

6. The method according to claim 1, characterized in that, The output recognition results are processed through result fusion and confidence evaluation, including: Consistency fusion is performed on the recognition results of the same spatial region obtained at N consecutive scanning times or through M overlapping recognition windows; the consistency fusion adopts the majority voting or confidence weighted average method, where N and M are both integers greater than 1; The confidence level of the fused recognition result is calibrated, and the final recognition result is output based on a preset trigger threshold.

7. The method according to claim 1, characterized in that, The method also includes an online self-calibration step: Collect confirmed radar echo identification samples during business operations to form a feedback sample library; Based on the returned sample library, the artificial intelligence recognition model is periodically retrained or incrementally updated.

8. A device for identifying radar echoes based on artificial intelligence, characterized in that, The device includes: The data acquisition module is used to acquire echo data from the weather radar over T consecutive scan times, where T is an integer greater than 1. The preprocessing module is used to perform signal-level preprocessing on the echo data of the T consecutive scanning times to suppress non-meteorological interference and extract time-series change information, so as to obtain the preprocessed time-series echo sequence. The feature construction module is used to construct multidimensional radar echo features based on the preprocessed time-series echo sequence. The multidimensional radar echo features include at least spatial dimension features and temporal dimension features. The intelligent recognition module is used to input the multi-dimensional radar echo features into the artificial intelligence recognition model to jointly model and recognize the multi-dimensional radar echo features, and output the radar echo type recognition result and corresponding confidence level. The fusion evaluation module is used to perform result fusion and confidence evaluation on the radar echo type identification results to obtain a stable final identification result. The business output module is used to perform business-oriented post-processing on the final recognition result, and generate and output target recognition information suitable for the business system. The self-updating module is used to perform online self-calibration and incremental updates on the artificial intelligence recognition model based on the confirmation samples collected during business operations.

Images