CN-122018843-A - Radar display control picture analysis method and device

CN122018843ACN 122018843 ACN122018843 ACN 122018843ACN-122018843-A

Abstract

The application discloses a radar display control picture analysis method and a device, which relate to the technical field of radar display control picture analysis, and the method comprises the steps of collecting a radar display control terminal display image containing multiple areas, dividing and processing according to preset rules, the processed images, task prompt words and query text are input into a visual language model, so that the model can synthesize multi-region information for analysis, and visual and language information is effectively fused. The effect of breaking through the limitation of traditional single-mode data processing during radar target identification is achieved, and multi-source information integration can be achieved. Meanwhile, interactive operation is executed on the basis of the human-computer interaction interface on the basis of the analysis result, so that flexible interaction between an operator and the model is facilitated, and the information communication and feedback efficiency is improved. Therefore, the accuracy and the efficiency of radar target identification can be improved, the interpretability of model output is enhanced, an operator is helped to better understand and verify judgment logic, a complex air threat situation is effectively treated, and the defects of the existing intelligent method are overcome.

Inventors

ZHANG JING
YANG LEI
LIU HAIFENG
SHI LI
GUO YANGYANG

Assignees

白杨时代(北京)科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. A radar display control picture analysis method, characterized in that the method comprises: Collecting a display image of a radar display control terminal, wherein the display image comprises a PPI display area, an R-D image display area, an A display waveform area and a track list area; Performing segmentation processing on the display image to obtain four independent sub-images corresponding to the PPI display area, the R-D image display area, the A display waveform area and the track list area respectively; processing the four independent sub-images according to a preset rule to obtain an image to be analyzed; Inputting the image to be analyzed, a preset task prompt word and a query text into a visual language model constructed in advance for analysis, and inputting the image to be analyzed into the visual language model according to a preset input sequence; and executing interactive operation on the analysis result output by the visual language model based on the man-machine interaction interface.
2. The method of claim 1, wherein the acquiring the display image of the radar display control terminal comprises: the method comprises the steps of collecting display images of a radar display control terminal according to a preset collecting time sequence and a preset collecting mode, wherein the preset collecting time sequence is real-time collecting synchronous with a radar scanning period or timing collecting at a fixed time interval, and the preset collecting mode is non-invasive collecting or system integrated collecting.
3. The method of claim 1, wherein the processing the four independent sub-images according to a preset rule to obtain an image to be analyzed comprises: Performing quality detection on each sub-image, judging whether the sub-images have abnormal conditions of black screen, picture freezing or data missing, and triggering re-acquisition or alarm operation on unqualified sub-images with abnormal conditions; And carrying out normalization processing on the qualified sub-image subjected to quality detection, adjusting the resolution of the grid-closing image to the input size required by the visual language model, and retaining the original color information of the qualified sub-image.
4. The method of claim 1, wherein before the image to be analyzed, the preset task prompt word, and the query text are input into a pre-constructed visual language model for analysis, the method further comprises: And constructing a task prompt word, wherein the task prompt word comprises role definition information, task description information, image interpretation information, domain knowledge base information and output format information, and the domain knowledge base information is structured information comprising radar target characteristic parameters, radar signal processing principles and radar tactical interpretation rules.
5. The method of claim 4, wherein before the image to be analyzed, the preset task prompt word and the query text are input into a pre-constructed visual language model for analysis, the method further comprises: and constructing a query text, wherein the query text comprises a radar scanning period number and a text corresponding to the space situation analysis problem.
6. The method according to claim 1, wherein the inputting the image to be analyzed, the preset task prompt word and the query text into a pre-constructed visual language model for analysis processing comprises: Combining the image to be analyzed with the task prompt words and the query text to form an inference request according to the sub-image sequence corresponding to the PPI display area, the R-D image display area, the A display waveform area and the track list area, and inputting the visual language model; and the visual language model sequentially executes the processing steps of visual coding, multi-modal fusion and autoregressive generation, then outputs a structural analysis result, and executes validity verification on the structural analysis result.
7. The method of claim 6, wherein said performing validity verification on said structured analysis result comprises: and verifying field integrity, numerical rationality and logic consistency of the structural analysis result, triggering reasonement or performing degradation processing on the structural analysis result which fails verification, generating a simplified version analysis result for extracting effective fields in the structural analysis result by the degradation processing, and storing the structural analysis result which fails verification into a situation database.
8. The method of claim 1, wherein performing an interactive operation on the analysis result output by the visual language model based on the human-machine interaction interface comprises: The human-computer interaction interface is provided with a situation analysis panel and a decision suggestion panel, wherein the situation analysis panel is used for displaying a blank condition summary, identification results of all targets and threat levels, and the decision suggestion panel is used for displaying operation suggestions in a priority list form; the man-machine interaction interface also comprises an alarm module, wherein the alarm module is used for actively popup reminding the conditions of newly-appearing threat targets, flight path mutation and abnormal echoes; And supporting the natural language interaction between an operator and the visual language model, and giving questions or inquiring about the judgment result of the visual language model.
9. The method according to claim 1, wherein the method further comprises: And executing track association on the same target by utilizing the analysis results for multiple times, executing fusion processing on the multiple types of judgment results of the same target, updating the confidence coefficient of target type judgment, carrying out state marking on the target according to the updated confidence coefficient, and triggering an abnormal alarm on the target with contradictory track mutation and target type judgment results.
10. A radar display control picture analysis device, characterized in that the device comprises: The acquisition unit is used for acquiring a display image of the radar display control terminal, wherein the display image comprises a PPI display area, an R-D image display area, an A display waveform area and a track list area; The image processing unit is used for executing segmentation processing on the display image to obtain four independent sub-images respectively corresponding to the PPI display area, the R-D image display area, the A display waveform area and the track list area; The image processing unit is also used for processing the four independent sub-images according to a preset rule to obtain an image to be analyzed; the input unit is used for inputting the image to be analyzed, the preset task prompt word and the query text into a visual language model constructed in advance for analysis and processing, and the image to be analyzed is input into the visual language model according to a preset input sequence; And the interaction unit is used for executing interaction operation on the analysis result output by the visual language model based on the man-machine interaction interface.

Description

Radar display control picture analysis method and device Technical Field The application relates to the technical field of radar display control picture analysis, in particular to a radar display control picture analysis method and device. Background In the radar display control screen, a plurality of interfaces such as a plane position indication display (Plan Position Indicator, PPI display), a Range-Doppler Map (R-D Map), and an a-Scope display coexist. The operator needs to observe the interfaces and combine the flight path parameters to realize radar target identification through manual judgment, and the requirements on the professional knowledge and actual combat experience of the operator are extremely high. In recent years, deep learning technology has been applied to the radar target recognition field, and Convolutional Neural Networks (CNNs) have been used in many applications, and mainly aiming at classification recognition works such as synthetic aperture radar images (SYNTHETIC APERTURE RADAR IMAGE, SAR images), high-resolution range profiles (High Resolution Range Profile, HRRP), time-Frequency maps (TF maps) and the like, and certain results have been obtained in specific data sets. However, its limitations are also very prominent. Firstly, the technology can only process single-mode data, and cannot effectively integrate multi-source information, so that complementary advantages among different types of data are difficult to develop, and the technology is worry about when facing complex and changeable actual conditions. Second, the lack of natural language understanding and generating capabilities results in an inability to flexibly interact with the operator in question-and-answer. In an actual operation scene, the information communication and feedback efficiency is greatly reduced, and the rapid and accurate judgment of the target is affected. Moreover, the model output lacks of interpretability, an operator is difficult to know the reasoning basis of the model output, and the judgment logic of the artificial intelligence cannot be understood and verified, so that the credibility and practicability of the model in practical application are reduced. Finally, the general deep learning method fails to fully utilize the expert knowledge in the fields such as radar signal processing principles and target characteristics, and the expert knowledge in the fields is difficult to effectively integrate into the model, so that the generalization capability of the model in practical application and the adaptability to different scenes are insufficient. In summary, the existing intelligent radar target recognition method faces a plurality of limitations in the actual combat environment, and operators still rely on the traditional manual judgment mode to a great extent. However, as the threat situation in the air becomes more complex, the coping ability of this conventional manner is very limited, and a more effective solution is urgently needed. Disclosure of Invention The application provides a radar display control picture analysis method and a device thereof, aiming at the problems, comprising the following contents: in a first aspect, the present application provides a radar display control picture analysis method, which includes: Collecting a display image of a radar display control terminal, wherein the display image comprises a PPI display area, an R-D image display area, an A display waveform area and a track list area; Performing segmentation processing on the display image to obtain four independent sub-images corresponding to the PPI display area, the R-D image display area, the A display waveform area and the track list area respectively; processing the four independent sub-images according to a preset rule to obtain an image to be analyzed; Inputting the image to be analyzed, a preset task prompt word and a query text into a visual language model constructed in advance for analysis, and inputting the image to be analyzed into the visual language model according to a preset input sequence; and executing interactive operation on the analysis result output by the visual language model based on the man-machine interaction interface. Optionally, the acquiring the display image of the radar display control terminal includes: the method comprises the steps of collecting display images of a radar display control terminal according to a preset collecting time sequence and a preset collecting mode, wherein the preset collecting time sequence is real-time collecting synchronous with a radar scanning period or timing collecting at a fixed time interval, and the preset collecting mode is non-invasive collecting or system integrated collecting. Optionally, the processing the four independent sub-images according to a preset rule to obtain an image to be analyzed includes: Performing quality detection on each sub-image, judging whether the sub-images have abnormal conditions of black screen, picture free