Intelligent Target Identification Method and System that Integrates Multi-Source Information

Abstract

The present invention relates to a target intelligent identification domain, specifically to a method and system for intelligent identification of strike targets by fusing multi-source information. The method obtains distribution information of an obstructed area by constructing a real-time three-dimensional model of an urban area, obtains characteristic information of a strike target by a data fusion method, predicts the obstructed area that the strike target enters by spatial comparative analysis; predicts the predicted moving trajectory of the strike target in the obstructed area by a dynamic filtering algorithm; obtains the real-time position of the strike target in the obstructed area by the predicted moving trajectory and characteristic information of the strike target; obtains detection data when the strike target leaves the obstructed area and obtains an identification result by a feature matching algorithm, and when the identification result is that the strike target cannot be confirmed, the target identification system triggers a correction mechanism. This solves the problem that after the strike target enters a concealed area, it is difficult to continuously track and accurately lock the target, resulting in a decrease in strike accuracy or loss of the target.

Description

Technical Field

[0001] The present invention relates to the field of intelligent target identification, and in particular to an intelligent identification method and system for attack targets integrating multi-source information. Background Art

[0002] In urban battlefields or operations, strike systems often face situations where targets quickly cross streets, enter gaps between buildings, or camouflage in complex traffic. Especially when some targets are vehicles, vehicles can enter some concealed areas at any time due to their flexibility, such as tunnels, under bridges, or narrow streets between high-rise buildings to avoid tracking and strikes.

[0003] Traditional target recognition systems have difficulty in continuously tracking and accurately locking targets after they enter concealed areas, so they cannot accurately locate the target's position and movement path, which can easily cause temporary loss of the target or target recognition errors, resulting in reduced strike accuracy or loss of the target. Therefore, there is an urgent need for an intelligent target recognition method that integrates multi-source information. Even if the target enters a complex shielded area in an urban environment, the strike system can still accurately track and re-identify the target, reducing the risk of losing the target, while improving the accuracy and real-time performance of the strike. Summary of the invention

[0004] (1) Technical issues to be resolved

[0005] The purpose of the present invention is to provide an intelligent identification method and system for strike targets that integrate multi-source information, so as to solve the problem that after the strike target enters a hidden area, it is difficult to continuously track and accurately lock the target. Therefore, the position and movement path of the strike target cannot be accurately located, which can easily cause temporary loss of the target or target identification errors, resulting in reduced strike accuracy or loss of the target.

[0006] (2) Technical solution

[0007] To achieve the above object, on the one hand, the present invention provides a method for intelligently identifying strike targets by fusing multi-source information, the method comprising:

[0008] S1. Acquire initial urban area information in a target recognition system, acquire real-time urban area information through sensors of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and acquire distribution information of the occluded area according to the real-time three-dimensional model of the urban area.

[0009] S2. Acquire detection data of the strike target and record it as first data, and obtain feature information of the strike target through a data fusion method; perform spatial comparative analysis on the feature information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters.

[0010] S3. Obtain the motion data of the strike target and the data of the occluded area, and predict the predicted movement trajectory of the strike target in the occluded area through a dynamic filtering algorithm; obtain the real-time position of the strike target in the occluded area through the predicted movement trajectory and the characteristic information of the strike target.

[0011] S4. When the real-time position of the strike target leaves the occlusion area, the target recognition system obtains the detection data of the strike target when it leaves the occlusion area and records it as the second data, and obtains the recognition result through the feature matching algorithm of the second data. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism.

[0012] Furthermore, the method of acquiring the motion data of the strike target and the data of the occluded area, and predicting the predicted movement trajectory of the strike target in the occluded area by a dynamic filtering algorithm; and obtaining the real-time position of the strike target in the occluded area by the predicted movement trajectory and the characteristic information of the strike target includes:

[0013] The first data includes optical data, infrared data and radar data. The first data is used to construct a multimodal feature fusion model of the strike target through a multimodal data fusion algorithm, and the feature information of the strike target is obtained according to the multimodal feature fusion model of the strike target.

[0014] The motion data of the target includes historical movement trajectory, movement speed, and acceleration. The data of the occluded area is the geometric features of the occluded area. The dynamic filtering algorithm uses a particle filtering algorithm to predict the predicted movement trajectory of the target in the occluded area. When the target enters the occluded area, feature extraction is performed on the target to obtain limited feature information and the missing feature information is supplemented by a deep learning model to obtain the estimated feature information of the target.

[0015] The feature information of the strike target before entering the occlusion area is obtained and recorded as historical feature information, and the estimated feature information of the strike target and the historical feature information are calculated by a cosine similarity method to obtain a correlation coefficient. When the correlation coefficient is greater than a preset correlation threshold, the current position of the strike target is obtained and recorded as the real-time position; when the correlation coefficient is less than the preset correlation threshold, the current position predicted according to the predicted movement trajectory is recorded as the real-time position.

[0016] Furthermore, the method further comprises:

[0017] The second data is preprocessed to obtain third data, wherein the data preprocessing includes image enhancement and noise filtering; the feature matching algorithm is to calculate the similarity between the third data and the historical feature information, and when the similarity is greater than a preset similarity threshold, the recognition result is that the strike target is confirmed; when the similarity is less than the preset similarity threshold, the recognition result is that the strike target cannot be confirmed.

[0018] Furthermore, the method further comprises:

[0019] When the strike target is in the shielding area, a sliding time window is set. When the strike target appears completely camouflaged within the sliding time window, the detection frequency of the target recognition system is increased when the strike target reappears.

[0020] Furthermore, the method further comprises:

[0021] The size of the sliding time window is adaptively adjusted according to the motion data of the strike target and the data of the occluded area.

[0022] The completely disguised target recognition system cannot capture the characteristic information of the strike target; the time and position of the reappearance of the strike target are obtained through time series prediction using a hidden Markov model; and the detection frequency of the sensor is adjusted through a nonlinear control algorithm based on the time and position of the reappearance of the strike target.

[0023] Furthermore, the method of triggering the correction mechanism of the target recognition system includes:

[0024] The correction mechanism is feature reconstruction calibration, which is to obtain the dynamic environmental factor changes of the strike target when it enters and leaves the occluded area, adaptively adjust the weight of the detection data according to the dynamic environmental factor changes, re-fuse the data and re-identify the strike target.

[0025] Furthermore, the method of performing spatial comparative analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters includes:

[0026] The characteristic information of the strike target and the distribution information of the occlusion area are mapped to the same coordinate system for spatial comparative analysis; the spatial comparative analysis is to obtain the current motion data of the strike target and obtain the predicted real-time movement trajectory through a dynamic filtering algorithm, and the predicted movement trajectory is analyzed by a spatial distance calculation method to determine whether there is an intersection between the predicted real-time movement trajectory and the occlusion area. If there is an intersection between the predicted real-time movement trajectory and the occlusion area, the occlusion area is recorded as the occlusion area entered by the strike target.

[0027] Based on the same inventive concept, the present invention also provides an intelligent identification system for attack targets that integrates multi-source information, the system comprising: an occlusion area distribution information analysis module, an occlusion area acquisition module, a real-time position acquisition module, and a target identification management module, and each module is sequentially connected in communication;

[0028] The distribution information analysis module of the occluded area is used to obtain the initial urban area information in the target recognition system, obtain the real-time urban area information through the sensor of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and obtain the distribution information of the occluded area according to the real-time three-dimensional model of the urban area.

[0029] The occlusion area acquisition module is used to obtain the detection data of the strike target and record it as the first data, and obtain the characteristic information of the strike target through a data fusion method; the characteristic information of the strike target and the distribution information of the occlusion area are spatially compared and analyzed to predict the occlusion area that the strike target enters.

[0030] A real-time position acquisition module is used to obtain the motion data of the strike target and the data of the occluded area, and predict the predicted movement trajectory of the strike target in the occluded area through a dynamic filtering algorithm; and obtain the real-time position of the strike target in the occluded area through the predicted movement trajectory and the characteristic information of the strike target.

[0031] The target recognition management module is used for, when the real-time position of the strike target leaves the shielding area, the target recognition system obtains the detection data of the strike target when leaving the shielding area and records it as the second data, and obtains the recognition result by the feature matching algorithm of the second data. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism.

[0032] (3) Beneficial effects

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. By fusing multimodal data of optical data, infrared data and radar data, the target recognition system can take advantage of the data of different modalities and dynamically adjust the weight coefficient according to the matching coefficient of each modality through an adaptive weighting algorithm to ensure that the fused target feature information is more accurate.

[0035] 2. When the target enters the occluded area, the dynamic filtering method is used to perform predictive analysis based on motion data and feature information to accurately calculate the movement trajectory and position of the target. In addition, the detection frequency of the sensor is dynamically adjusted in combination with the sliding time window mechanism to ensure that the target can be identified in time when it reappears in the visible area, reducing the possibility of target loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a flowchart of the method for intelligently identifying strike targets by fusing multi-source information according to Embodiment 1 of the present invention;

[0038] Figure 2 This is a schematic diagram of the module composition of the intelligent target identification system that integrates multi-source information according to Example 2 of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Before giving examples, it is necessary to explain the application scenarios conceived by the present invention. The present invention is an intelligent identification method and system for attack targets that integrate multi-source information. In the complex battlefield environment of urban agglomerations, high-speed aircraft attack systems need to identify and track mobile targets, such as escaped vehicles, enemy command vehicles, etc. However, due to the obstruction of buildings such as high-rise buildings and bridges, the attack targets often partially disappear or temporarily disappear completely in the obstructed area, which brings challenges to precision strikes. For example, in urban combat, enemy vehicles try to escape quickly under the obstruction of high-rise buildings and bridges, and use the obstruction of buildings to avoid being locked by high-speed aircraft. Therefore, vehicles constantly shuttle through complex urban roads and overpasses, which brings great difficulties to target locking and continuous tracking. Therefore, it is necessary to be able to accurately identify and track enemy vehicles in complex urban terrain. Even if the vehicle uses a building to dodge for a short time, the system can dynamically track through intelligent prediction to ensure the accuracy and continuity of the attack.

[0041] Example 1: Figure 1 As shown, this embodiment provides a method for intelligently identifying strike targets by integrating multi-source information, the method comprising:

[0042] S1. Acquire the initial urban area information in the target recognition system, acquire the real-time urban area information through the sensor of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and acquire the distribution information of the occluded area according to the real-time three-dimensional model of the urban area; the initial urban area information is the geographic data pre-stored in the target recognition system, including the layout information of urban buildings, roads, rivers, etc. The target recognition system pre-constructs the basic framework model of the urban area by analyzing these geographic data. The sensors include optical sensors and radar sensors. The optical sensors are used to provide high-resolution urban images and capture the appearance of visible objects such as buildings, vehicles, and trees. The radar sensor captures the depth information of the environment by emitting laser pulses and measuring their return time, thereby generating point cloud data of the city for identifying the height, shape, and complex terrain of buildings. The initial urban area information is fused with the data of the optical sensor and the radar sensor to construct a real-time three-dimensional urban model, and the model is updated using SLAM (Simultaneous Localization and Mapping) technology. The geometric features of the occluded areas are analyzed according to the constructed real-time three-dimensional urban model, and the positions of these occluded areas in the urban model are recorded. The occluded areas include behind buildings, under bridges, dense tree areas, etc. Use ray projection technology or volume detection algorithm to determine which buildings or terrain may block the sensor's line of sight and generate distribution information of the blocked area. The distribution information of the blocked area provides a basis for subsequent prediction of the hidden areas that the strike target may enter.

[0043] S2. Obtain the detection data of the strike target and record it as the first data. Use the first data to obtain the characteristic information of the strike target through a data fusion method; perform spatial comparative analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters; the first data includes multi-source information captured by multi-modal sensors such as optical, infrared, and radar. For example, if the strike target is a vehicle, the optical sensor obtains the appearance characteristics of the vehicle (color, shape, size, license plate, etc.), the infrared sensor captures the thermal signal of the vehicle (such as the heat emitted by the engine), and the radar obtains the precise distance information of the speed, acceleration, and position of the vehicle. Use a multi-modal data fusion algorithm (such as weighted fusion and Bayesian fusion) to perform feature fusion on data from different sources and extract the characteristic information of the strike target, including appearance characteristics, speed, thermal signal, position, etc. Map the distribution information of the strike target and the occlusion area to the same coordinate system for spatial analysis.

[0044] S3. Obtain the motion data of the target and the data of the occluded area, and predict the predicted moving trajectory of the target in the occluded area through a dynamic filtering algorithm; obtain the real-time position of the target in the occluded area through the predicted moving trajectory and the characteristic information of the target; the motion data of the target comes from the continuous detection of the sensor, including position, speed, and acceleration, and the data of the occluded area includes the geometric features of the occluded area (such as street direction, building height, etc.), and the possible moving trajectory of the target in the occluded area is predicted through trajectory prediction algorithms such as particle filtering. Match the predicted result of the motion trajectory of the target with the characteristic information of the target to determine the real-time position of the target in the occluded area. Because the occluded area may cause the vehicle to change its driving strategy (such as sudden acceleration, steering or camouflage), then combining the characteristic information matching of the target can better help the target recognition system to accurately identify it.

[0045] S4. When the real-time position of the strike target leaves the occluded area, the target recognition system obtains the detection data when the strike target leaves the occluded area and records it as the second data. The second data is used to obtain the recognition result through the feature matching algorithm. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism. The feature matching algorithm can quickly compare the target in the detection data with the target features in the known database, and is suitable for scenarios with high real-time requirements such as high-speed aircraft target recognition. In addition, it can support the fusion of data from different sensors and use multiple features for matching, which helps to improve the recognition accuracy in complex environments (such as when the target is occluded or the target shape changes). However, in some cases, matching failure or misidentification may occur, and the correction mechanism needs to be triggered to improve the accuracy of target recognition.

[0046] It should be noted that the method of obtaining the motion data of the strike target and the data of the occluded area, and predicting the predicted movement trajectory of the strike target in the occluded area by a dynamic filtering algorithm; and obtaining the real-time position of the strike target in the occluded area by the predicted movement trajectory and the characteristic information of the strike target includes:

[0047] The first data includes optical data, infrared data and radar data. The first data is used to construct a multimodal feature fusion model of the strike target through a multimodal data fusion algorithm, and the feature information of the strike target is obtained according to the multimodal feature fusion model of the strike target; the multimodal feature fusion model can comprehensively utilize multiple sensor data to extract key feature information of the strike target. Use a multimodal data fusion algorithm (such as a convolutional neural network CNN or a self-attention mechanism Transformer) to extract and fuse features of various types of data, build a multimodal feature fusion model, and obtain the feature information of the strike target.

[0048] The motion data of the strike target includes historical moving trajectory, moving speed, and acceleration. The data of the occlusion area is the geometric features of the occlusion area. The dynamic filtering algorithm uses a particle filtering algorithm to predict the predicted moving trajectory of the strike target in the occlusion area. When the strike target enters the occlusion area, the strike target is feature extracted to obtain limited feature information and the missing feature information is supplemented by a deep learning model to obtain the estimated feature information of the strike target. The target's motion trajectory is calculated by the position data of the strike target collected historically, and the moving speed and acceleration of the strike target are obtained by radar data, so as to predict the future movement of the strike target. According to the three-dimensional model of the urban area, the occlusion area and its geometric features, including the height, width, and position of the building, are obtained by occlusion calculation or ray projection technology. Particle filtering can handle the uncertainty of the strike target and provide more accurate trajectory prediction in a complex environment. By sampling the motion state of the strike target and generating a large number of possible trajectory samples, and combining the target's motion data and the geometric features of the occlusion area, the weight of each sample is updated, and finally the most likely moving trajectory is obtained. When the target enters the occlusion area, due to the obstruction of the sensor's line of sight, only limited target feature information can be obtained. At this time, these limited features are completed through deep learning models to obtain the estimated feature information of the target. Use deep learning models (such as time series models based on RNN or Transformer) to process limited sensor data, and infer the complete features of the target through context information and historical feature information. Estimate the features that are currently unavailable (such as shape, speed, etc.), so as to complete the incomplete feature information and generate the estimated feature information of the target. After the target enters the occluded area, the system's ability to recognize the target's features is improved through information completion, thereby avoiding misidentification or tracking failure due to feature loss.

[0049] The characteristic information of the strike target before entering the shielding area is obtained and recorded as historical characteristic information. The estimated characteristic information of the strike target and the historical characteristic information are calculated by the cosine similarity method to obtain a correlation coefficient. When the correlation coefficient is greater than a preset correlation threshold, the current position of the strike target is obtained and recorded as the real-time position; when the correlation coefficient is less than the preset correlation threshold, the current position predicted according to the predicted movement trajectory is recorded as the real-time position. The correlation coefficient of the two is obtained by the cosine similarity calculation method, which is used to determine whether the current characteristic information matches the characteristic information of the strike target. The estimated characteristic information of the strike target and the historical characteristic information are converted into a characteristic vector, and the characteristic vector of the strike target is compared with the historical characteristic vector stored in the system by the cosine similarity algorithm to calculate the correlation coefficient between the two. The value range of cosine similarity is 0 to 1, and the closer the value is to 1, the more similar the two are. Cosine similarity provides a quantitative basis for determining whether the target is correctly identified. When the similarity is high, the current position of the target can be confirmed. The correlation coefficient calculated according to the cosine similarity is used to determine the current position of the strike target. If the correlation coefficient is greater than the preset threshold, the current position of the target is obtained and recorded as the real-time position; if the correlation coefficient is less than the threshold, the current position of the target is determined based on the trajectory predicted by the particle filter. If the cosine similarity is greater than the preset threshold, it means that the system recognizes the target with high accuracy, and the current position of the target is directly obtained as the real-time position. If the cosine similarity is low, the system believes that the target may be in an obscured or partially lost state, and uses the particle filter to predict the trajectory to estimate the target's position, and this position is used as the real-time position. Through this step, the system can continue to accurately estimate the target's position in the obscured area or when the signal is incomplete.

[0050] It should be noted that the method further includes:

[0051] The second data is subjected to data preprocessing to obtain the third data, wherein the data preprocessing includes image enhancement and noise filtering; the feature matching algorithm is to calculate the similarity between the third data and the historical feature information, and when the similarity is greater than a preset similarity threshold, the recognition result is to confirm the strike target; when the similarity is less than the preset similarity threshold, the recognition result is that the strike target cannot be confirmed. Various image processing techniques such as histogram equalization, contrast adjustment, and sharpening processing are used to improve the visibility of the image under poor lighting conditions or blurred images. Enhancement processing can make the target features clearer and improve the accuracy of system recognition. Denoising algorithms (such as Gaussian filtering, median filtering, etc.) are used to remove random noise in the image, reduce interference factors in the data, and maintain the true strike target features in the image. By enhancing image quality and removing noise, the recognizability of target features is improved. The third data generated by preprocessing is extracted using a feature matching algorithm (such as SIFT, SURF, or ORB) to extract the key feature points of the target, and then matched with the historical feature information to obtain a recognition result of whether the target is consistent with the historical data.

[0052] It should be noted that the method further includes:

[0053] When the strike target is in the occlusion area, a sliding time window is set. When the strike target appears completely camouflaged within the sliding time window, the detection frequency of the target recognition system is increased when the strike target reappears. When the strike target enters the occlusion area, the system will set a sliding time window to monitor the target's activities in the occlusion area. The sliding time window refers to a time range set by the system after the target enters the occlusion area for continuous monitoring of the target's activities in the occlusion area. The target camouflage judgment is based on the characteristic information of the strike target, the geometric characteristics of the occlusion area, and the historical movement pattern, to determine whether the target may disappear from the field of vision within the sliding time window due to entering the shadow of the building or being stationary. The sliding time window ensures that the system continuously monitors the dynamics of the strike target in the occlusion area to prevent the target from being invisible for a long time, especially when it may reappear, the detection strategy of the system can be dynamically adjusted. If the target appears completely camouflaged within the sliding time window, the system will increase the detection frequency to ensure that the target can be captured in time when it reappears in the field of vision. The detection frequency adjustment adjusts the detection frequency when the strike target approaches the end of the sliding time window, ensuring that the target can be recaptured and identified in time even if it disappears briefly in the occluded area, thereby reducing the risk of target loss.

[0054] It should be noted that the method further includes:

[0055] The size of the sliding time window is adaptively adjusted according to the motion data of the strike target and the data of the occluded area; the adaptive adjustment of the sliding time window can flexibly respond to the actual scene and the dynamic situation of the strike target, ensuring that the system finds the best balance between resource efficiency and the ability to capture the strike target. For example, if the target moves faster or the occluded area is smaller, the time window may be shorter, otherwise it may be longer.

[0056] The complete camouflage means that the target recognition system cannot capture the characteristic information of the strike target; the time and position of the reappearance of the strike target are obtained by time series prediction through the hidden Markov model; the detection frequency of the sensor is adjusted by the nonlinear control algorithm according to the time and position of the reappearance of the strike target. When the strike target enters the occlusion area, if the target recognition system cannot capture any characteristic information of the target, the target is determined to be in a complete camouflage state. At this time, the sliding time window continues to count, and when the window ends, the system uses the hidden Markov model to perform time series prediction to infer the time and position of the possible reappearance of the strike target. After the strike target enters the occlusion area, if the target features cannot detect data in the multimodal sensor, the system will determine that the target has entered a complete camouflage state. This means that the target may be behind a building, in dense obstacles or stationary, and the sensor cannot obtain any valid data. The system uses the hidden Markov model to make predictions based on the historical motion data of the strike target. The model can infer the time and position of the next reappearance of the target based on the state transition probability. For example, based on the motion pattern and environmental constraints before the target enters the occlusion area, the possible reappearance time of the target is inferred. In the fully camouflaged state, the system performs timing prediction, which helps to improve the accuracy of the target's reappearance time and position, and can roughly infer the target's movement even without any direct observation. The system uses a nonlinear control algorithm to dynamically adjust the sensor's detection frequency. In particular, when it is predicted that the strike target is about to reappear from the occluded area, the system will increase the sensor's detection frequency to ensure that the target can be captured in time. The system uses a nonlinear control algorithm (such as PID control or adaptive control algorithm) to adjust the sensor's detection frequency. The nonlinear control algorithm can flexibly adjust the sensor's sampling frequency based on the target's motion state, predicted trajectory, and occluded area characteristics. For example, when the target is predicted to leave the occluded area, the system will gradually increase the detection frequency to ensure that the target is captured. As the target's trajectory prediction is continuously updated, the sensor's detection frequency will also be adjusted in real time according to the target's proximity. If the target is about to enter the visible area, the detection frequency will increase significantly; if the strike target is predicted to still be in the occluded area, the detection frequency may remain or decrease. By dynamically adjusting the detection frequency through a nonlinear control algorithm, the system can ensure that the target is captured at a high frequency at critical moments, while reducing unnecessary frequent detection when the target moves slowly or is in an occluded state, thereby ensuring that the target recognition system is in efficient operation.

[0057] It should be noted that the method for triggering the correction mechanism of the target recognition system includes:

[0058] The correction mechanism is feature reconstruction calibration, which is to obtain the dynamic environmental factor changes of the strike target when it enters and leaves the occlusion area, and adaptively adjust the weight of the detection data according to the dynamic environmental factor changes to re-fuse the data and re-identify the strike target. When the strike target enters and leaves the occlusion area, the system monitors and records the changes of the surrounding dynamic environmental factors. These changes may include factors that affect the performance of the sensor, such as light, weather, wind speed, temperature, etc. According to the collected dynamic environmental factors, the system will adaptively adjust the weight of the detection data. For example, if the light suddenly decreases, affecting the performance of the optical sensor, the system can reduce the weight of the sensor data; on the contrary, if the data of the infrared sensor is more advantageous, its weight is increased. Using the adjusted weights, the system will re-fuse the detection data from different sensors. This step is to integrate data from various sources into more reliable target feature information through weighted averaging, weighted decision trees, etc. After feature reconstruction calibration, the system will output an updated target recognition result. If the re-fused feature information allows the target recognition system to confirm the strike target, subsequent operations will be performed, such as tracking, launching a strike, etc.

[0059] It should be noted that the method of performing spatial comparative analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters includes:

[0060] The characteristic information of the strike target and the distribution information of the occlusion area are mapped to the same coordinate system for spatial comparison analysis; the spatial comparison analysis is to obtain the current motion data of the strike target and obtain the predicted real-time moving trajectory through the dynamic filtering algorithm, and the predicted moving trajectory is analyzed by the spatial distance calculation method to see whether there is an intersection between the predicted real-time moving trajectory and the occlusion area. If there is an intersection between the predicted real-time moving trajectory and the occlusion area, the occlusion area is recorded as the occlusion area entered by the strike target. The motion data of the strike target and the geometric distribution information of the occlusion area are mapped to the unified coordinate system of the three-dimensional model of the city to ensure that all data in the subsequent analysis are performed in the same spatial reference frame. The geometric shape information of the occlusion area is obtained, such as the three-dimensional contour and position of buildings and obstacles. The relationship between the predicted moving trajectory of the strike target and the geometric boundary of the occlusion area is calculated by the spatial distance calculation method (such as the ray and polygon intersection detection algorithm or the closest point distance algorithm). Specifically, through geometric analysis, it is determined whether the trajectory of the strike target will intersect with the boundary of the occlusion area. The detection can be performed step by step by calculating the motion vector of the trajectory at different time points and the edge of the occlusion area. If, during the trajectory prediction process, it is found that the predicted trajectory of the target intersects with the boundary of the occlusion area, it can be determined that the target is about to enter this occlusion area.

[0061] Embodiment 2: Based on the same inventive concept, this embodiment also provides an intelligent target identification system integrating multi-source information, including: a distribution information analysis module of an occlusion area, an occlusion area acquisition module, a real-time position acquisition module, and a target identification management module, wherein each module is sequentially connected in communication;

[0062] The occlusion area distribution information analysis module is used to obtain the initial urban area information in the target recognition system, obtain the real-time urban area information through the sensor of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and obtain the distribution information of the occlusion area according to the real-time three-dimensional model of the urban area;

[0063] The occlusion area acquisition module is used to acquire the detection data of the strike target and record it as the first data, obtain the characteristic information of the strike target through the data fusion method of the first data; perform spatial comparison analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters;

[0064] A real-time position acquisition module is used to acquire the motion data of the strike target and the data of the occluded area, and predict the predicted movement trajectory of the strike target in the occluded area through a dynamic filtering algorithm; and obtain the real-time position of the strike target in the occluded area through the predicted movement trajectory and the characteristic information of the strike target;

[0065] The target recognition management module is used for, when the real-time position of the strike target leaves the shielding area, the target recognition system obtains the detection data of the strike target when leaving the shielding area and records it as the second data, and obtains the recognition result by the feature matching algorithm of the second data. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism.

[0066] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0067] Finally, it should be noted that: Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent identification method for attack targets integrating multi-source information, characterized in that: The method comprises: Acquire initial urban area information in the target recognition system, acquire real-time urban area information through a sensor of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and acquire distribution information of the occluded area according to the real-time three-dimensional model of the urban area; Acquire detection data of the strike target and record it as first data, obtain feature information of the strike target by using a data fusion method for the first data; perform spatial comparison analysis on the feature information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters; Acquire the motion data of the strike target and the data of the occluded area, and predict the predicted movement trajectory of the strike target in the occluded area through a dynamic filtering algorithm; obtain the real-time position of the strike target in the occluded area through the predicted movement trajectory and the characteristic information of the strike target; When the real-time position of the strike target is out of the shielding area, the target recognition system obtains the detection data of the strike target when it leaves the shielding area and records it as the second data, and obtains the recognition result by the feature matching algorithm. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism; The method of obtaining the motion data of the strike target and the data of the occluded area, and predicting the predicted movement trajectory of the strike target in the occluded area by a dynamic filtering algorithm; and obtaining the real-time position of the strike target in the occluded area by the predicted movement trajectory and the characteristic information of the strike target includes: The first data includes optical data, infrared data and radar data, and the first data is used to construct a multimodal feature fusion model of the strike target through a multimodal data fusion algorithm, and characteristic information of the strike target is obtained according to the multimodal feature fusion model of the strike target; The motion data of the strike target includes historical movement trajectory, movement speed, and acceleration; the data of the occluded area is the geometric features of the occluded area; the dynamic filtering algorithm uses a particle filtering algorithm to predict the predicted movement trajectory of the strike target in the occluded area; when the strike target enters the occluded area, feature extraction is performed on the strike target to obtain limited feature information and the missing feature information is supplemented by a deep learning model to obtain the estimated feature information of the strike target; Acquire characteristic information of the strike target before it enters the shielded area and record it as historical characteristic information, calculate the correlation coefficient between the estimated characteristic information of the strike target and the historical characteristic information by a cosine similarity calculation method, and when the correlation coefficient is greater than a preset correlation threshold, acquire the current position of the strike target and record it as the real-time position; when the correlation coefficient is less than the preset correlation threshold, acquire the current position predicted according to the predicted movement trajectory and record it as the real-time position; When the strike target is in the shielded area, a sliding time window is set, and when the strike target appears completely camouflaged within the sliding time window, the detection frequency of the target recognition system is increased when the strike target reappears; The size of the sliding time window is adaptively adjusted according to the motion data of the strike target and the data of the occluded area; The completely disguised target recognition system cannot capture the characteristic information of the strike target; the time and position of the reappearance of the strike target are obtained through time series prediction using a hidden Markov model; and the detection frequency of the sensor is adjusted through a nonlinear control algorithm based on the time and position of the reappearance of the strike target.

2. The method for intelligently identifying strike targets by integrating multi-source information according to claim 1 is characterized in that: The method further comprises: The second data is preprocessed to obtain third data, wherein the data preprocessing includes image enhancement and noise filtering; the feature matching algorithm is to calculate the similarity between the third data and the historical feature information, and when the similarity is greater than a preset similarity threshold, the recognition result is that the strike target is confirmed; when the similarity is less than the preset similarity threshold, the recognition result is that the strike target cannot be confirmed.

3. The method for intelligently identifying strike targets by integrating multi-source information according to claim 1 is characterized in that: The method for triggering the correction mechanism of the target recognition system includes: The correction mechanism is feature reconstruction calibration, which is to obtain the dynamic environmental factor changes of the strike target when it enters and leaves the occluded area, adaptively adjust the weight of the detection data according to the dynamic environmental factor changes, re-fuse the data and re-identify the strike target.

4. The method for intelligently identifying strike targets by integrating multi-source information according to claim 1 is characterized in that: The method of performing spatial comparative analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area into which the strike target enters comprises: The characteristic information of the strike target and the distribution information of the occlusion area are mapped to the same coordinate system for spatial comparative analysis; the spatial comparative analysis is to obtain the current motion data of the strike target and obtain the predicted real-time movement trajectory through a dynamic filtering algorithm, and the predicted movement trajectory is analyzed by a spatial distance calculation method to determine whether there is an intersection between the predicted real-time movement trajectory and the occlusion area. If there is an intersection between the predicted real-time movement trajectory and the occlusion area, the occlusion area is recorded as the occlusion area entered by the strike target.

5. An intelligent target identification system integrating multi-source information, characterized in that: The system includes: an occlusion area distribution information analysis module, an occlusion area acquisition module, a real-time position acquisition module, and a target recognition management module, and each module is sequentially connected in communication; The occlusion area distribution information analysis module is used to obtain the initial urban area information in the target recognition system, obtain the real-time urban area information through the sensor of the target recognition system, construct a real-time three-dimensional model of the urban area with the initial urban area information and the real-time urban area information, and obtain the distribution information of the occlusion area according to the real-time three-dimensional model of the urban area; The occlusion area acquisition module is used to acquire the detection data of the strike target and record it as the first data, obtain the characteristic information of the strike target through the data fusion method of the first data; perform spatial comparison analysis on the characteristic information of the strike target and the distribution information of the occlusion area to predict the occlusion area that the strike target enters; A real-time position acquisition module is used to acquire the motion data of the strike target and the data of the occluded area, and predict the predicted movement trajectory of the strike target in the occluded area through a dynamic filtering algorithm; and obtain the real-time position of the strike target in the occluded area through the predicted movement trajectory and the characteristic information of the strike target; The target recognition management module is used for, when the real-time position of the strike target leaves the shielding area, the target recognition system obtains the detection data of the strike target when leaving the shielding area and records it as the second data, and obtains the recognition result by the feature matching algorithm of the second data. When the recognition result is that the strike target cannot be confirmed, the target recognition system triggers the correction mechanism.

Images