Methods and systems for identifying hypersonic vehicle target types in complex battlefield environments

Abstract

This invention relates to the field of aircraft target type identification technology, specifically a method and system for identifying hypersonic aircraft target types in complex battlefield environments. The method includes acquiring image frame sequences and evaluating their validity to obtain an observation window sequence; predicting the instantaneously reachable spatial domain and discretizing it into a set of spatial index units; establishing a correlation between the observation windows and the spatial index units; extracting target fragment images from the correlated windows and obtaining fragment feature data; establishing and updating a type confidence vector based on the fragment feature data and validity markers; and finally determining the target type based on the updated type confidence vector. This invention can effectively integrate fragmented observation information under highly dynamic conditions, improving the reliability of identifying high-value target types.

Description

Technical Field

[0001] This invention relates to the field of aircraft target type identification technology, specifically to a method and system for identifying hypersonic aircraft target types in complex battlefield environments. Background Technology

[0002] Hypersonic vehicles typically rely on imaging seekers to acquire image frame sequences of ground or sea targets during the terminal guidance phase, and automatically complete target identification and type determination in the final stage of flight. This process is generally based on the extraction and analysis of the target's external features.

[0003] In complex battlefield environments such as coastal ports and near-shore waterways, targets are diverse and densely deployed, and obstruction is caused by dock facilities, shoreline features, and large ship structures. At the same time, hypersonic vehicles cannot obtain a continuous and stable observation perspective through large maneuvers during the terminal phase of flight.

[0004] Existing automatic target recognition technologies based on imaging seekers are mostly designed for subsonic or low-dynamic flight platforms. These technologies typically assume that the seeker can acquire target image sequences with gradual changes in perspective and a relatively continuous temporal sequence, relying on the gradual accumulation and optimization of feature information across consecutive image frames. However, in these scenarios, actual observations often consist of brief observation segments separated by numerous invalid image sections, with significant differences in perspective between adjacent segments. This results in discontinuous and fragmented usable observation information, making it difficult to timely and reliably distinguish between various types of mixed targets and associate the identity of the same target across different observation segments.

[0005] Therefore, there is an urgent need for a method and system for identifying the target type of hypersonic vehicles in complex battlefield environments, which can adapt to the actual observation mode of the terminal phase of hypersonic vehicles and improve the stability and reliability of target type identification under conditions of discontinuous and fragmented observation information. Summary of the Invention

[0006] (1) Technical problems to be solved

[0007] The purpose of this invention is to provide a method and system for identifying the target type of hypersonic vehicles in complex battlefield environments. This addresses the problem that hypersonic vehicles, when facing complex battlefield environments during the terminal guidance phase, can only obtain discrete, discontinuous observation windows with significant changes in viewing angle, making it difficult to effectively integrate fragmented image information and thus making it difficult to complete reliable target type identification under limited time and computing resource constraints.

[0008] (2) Technical solution

[0009] To achieve the above objectives, on the one hand, the present invention provides a method for identifying the type of hypersonic aircraft targets in complex battlefield environments, the method comprising:

[0010] Step S1: The hypersonic vehicle acquires an image frame sequence via an imaging seeker during the terminal guidance phase; the image frame sequence is evaluated for image validity to obtain a corresponding validity marker sequence; based on the validity marker sequence, consecutive image frames marked as valid are divided into observation windows, and image frames marked as invalid are divided into invalid segments; the observation windows are arranged in chronological order to obtain an observation window sequence.

[0011] Step S2: During flight, based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle, predict the instantaneously accessible spatial domain that updates over time; discretize the instantaneously accessible spatial domain to obtain a set of spatial index units; based on the attitude data and line-of-sight data at the corresponding moment of each observation window, determine the projection unit of the corresponding imaging area in the spatial index unit; associate the observation window with the corresponding projection unit to obtain the observation record organized by the spatial index unit.

[0012] Step S3: Extract the target fragment image from the observation window associated with the spatial index unit within the imaging area; perform feature extraction on the target fragment image to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; perform multi-window update on the type confidence vector to obtain the updated type confidence vector.

[0013] Step S4: Determine the corresponding target type based on the updated type confidence vector of each spatial index unit; when the updated type confidence vector does not meet the preset judgment condition, the corresponding target type is determined as an undetermined type.

[0014] Furthermore, the method for evaluating the image frame sequence to obtain the corresponding validity marker sequence includes:

[0015] For each image frame, at least one of the following indicators is calculated to obtain a validity index: sharpness index, field of view occupancy index, brightness contrast index, and obstruction index; the validity index is subjected to sliding statistical processing within a preset time window to obtain a stable validity index; based on the comparison result between the stable validity index and a preset validity threshold, a validity label is generated for each image frame to obtain the validity label sequence.

[0016] Furthermore, the method for predicting the instantaneously reachable spatial domain updated over time based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle includes:

[0017] The real-time attitude, velocity, and acceleration are subjected to low-pass filtering with a fixed time constant to obtain filtered state variables. Based on the filtered state variables and the maximum normal overload and maximum tangential overload of the aircraft, the heading change range, pitch change range, and velocity change range of the aircraft within a preset time step are calculated.

[0018] Based on the range of heading changes, pitch changes, velocity changes, and the effective range parameters of the imaging seeker, a spatial envelope region with the current position of the aircraft as the reference is determined; the spatial envelope region is then used as the instantly accessible spatial domain.

[0019] Furthermore, the method for discretizing the instantaneously accessible spatial domain to obtain a set of spatial index cells includes:

[0020] The angular step length is determined based on the angular resolution parameters of the imaging seeker, and the radial step length is determined based on the radial range of the instantaneously accessible spatial domain. Based on the angular step length and the radial step length, the instantaneously accessible spatial domain is divided into a three-dimensional mesh to obtain an initial set of spatial units.

[0021] Based on the boundary set consisting of the reachable boundary determined by the aircraft's maneuverability and the ground feature boundary determined by environmental data, the initial spatial units are trimmed to obtain the spatial index unit set.

[0022] Furthermore, the method for determining the angular step length based on the angular resolution parameters of the imaging seeker, and determining the radial step length based on the radial range of the instantaneously reachable spatial domain, includes:

[0023] The instantaneously accessible spatial domain is divided radially to obtain a set of distance intervals, and the boundaries of the distance intervals are determined based on the resolution of the imaging seeker at different distances.

[0024] The angular deviation length is calculated based on the center distance value of the distance interval and the angular resolution parameters of the imaging seeker; the radial deviation length is calculated based on the range of the distance interval and the preset radial resolution coefficient.

[0025] Furthermore, the method for determining the projection unit of the corresponding imaging region in the spatial index unit based on the attitude data and line-of-sight data at the corresponding time of each observation window includes:

[0026] The orientation vector of the imaging seeker's optical axis in the reference coordinate system is determined based on the attitude data; the field of view geometry is determined based on the line-of-sight data and the field of view parameters of the imaging seeker.

[0027] Determine whether each spatial index unit in the set of spatial index units intersects with the view frustum geometry, and identify the intersecting spatial index units as the projection units of the current observation window; associate and store the time information and validity marker of the observation window with the projection units.

[0028] Furthermore, the method for establishing a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window includes:

[0029] Obtain a preset type set; for each type in the preset type set, calculate the feature similarity between the reference feature data and the fragment feature data; determine the observation quality coefficient of the current observation window based on the validity labeling result.

[0030] Calculate the segment weight parameters for each type based on the observation quality coefficient and feature similarity; then, sum the initial confidence parameters for each type based on the segment weight parameters to obtain the type confidence vector.

[0031] Furthermore, the method for performing multi-window updates on the type confidence vector to obtain the updated type confidence vector includes:

[0032] For all observation windows associated with the same spatial index unit, the observation window sequence is obtained by arranging them in the order of observation time; the basic update weight is determined based on the validity marking results of each observation window; and the time decay factor is determined based on the time interval between each observation window and the current time.

[0033] The comprehensive update weight of each observation window is calculated based on the basic update weight and the time decay factor; the segment feature data corresponding to each observation window are weighted and fused according to the comprehensive update weight to obtain fused feature data.

[0034] The type confidence vector is cumulatively updated and normalized based on the fused feature data to obtain the updated type confidence vector.

[0035] Furthermore, the method for weighted fusion of the segment feature data corresponding to each observation window based on the comprehensive update weight to obtain fused feature data includes:

[0036] Based on the aircraft attitude and position data at the corresponding time of each observation window, the observation angle parameters of each observation window are calculated; based on the observation angle parameters of each observation window, the angle difference value between any two observation windows is calculated; observation windows with angle difference values ​​less than a preset angle consistency threshold are assigned to the same angle group to obtain the angle group set.

[0037] For each viewpoint group, the segment feature data is weighted and fused within the group according to the comprehensive update weight of each observation window to obtain the fused feature data within the group.

[0038] In the case of multiple view groups, the fused feature data within each view group is spliced ​​together or independently retained according to the weight of each view group to obtain fused feature data, which includes view identification information.

[0039] Based on the same inventive concept, this invention also provides a hypersonic vehicle target type identification system for complex battlefield environments, the system comprising:

[0040] The observation window division module is used for hypersonic vehicles to acquire image frame sequences through imaging seekers during the terminal guidance phase; to perform image validity evaluation on the image frame sequences to obtain corresponding validity marker sequences; to divide consecutive image frames marked as valid into observation windows and image frames marked as invalid into invalid segments according to the validity marker sequences; and to arrange the observation windows in chronological order to obtain an observation window sequence.

[0041] The spatial indexing and association module is used to predict the instantaneously accessible spatial domain updated over time based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle during flight; discretize the instantaneously accessible spatial domain to obtain a set of spatial indexing units; determine the projection unit of the corresponding imaging area in the spatial indexing unit based on the attitude data and line-of-sight data at the corresponding time of each observation window; and associate the observation window with the corresponding projection unit to obtain the observation record organized by the spatial indexing unit.

[0042] A multi-window feature fusion module is used to extract target fragment images from the observation window associated with the spatial index unit within the imaging region; perform feature extraction on the target fragment images to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; and perform multi-window update on the type confidence vector to obtain an updated type confidence vector.

[0043] The target type decision module is used to determine the corresponding target type based on the updated type confidence vector of each spatial index unit; when the updated type confidence vector does not meet the preset judgment conditions, the corresponding target type is determined as an undetermined type.

[0044] (3) Beneficial effects

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. By constructing an instantly accessible spatial domain bound to flight status and discretizing it into spatial index units, target fragments from observation windows at different times and from different perspectives are associated with unified spatial index units based on geometric relationships. This provides a stable spatial reference framework for non-continuous fragment feature data from multiple observation windows, thereby overcoming the problem of difficulty in continuously tracking and associating target identities caused by observation fragmentation in hypersonic environments.

[0047] 2. Furthermore, by weighting based on validity markers, the type confidence vectors of multiple observation windows associated with the same spatial index unit are fused and updated in a multi-window manner. This automatically weighs the contributions of observation information of different qualities and at different times, making the final updated type confidence vector more reflective of the essential characteristics of the target. Thus, in complex scenarios, it provides a higher confidence basis for determining whether a target is identified as a predetermined strike type, reducing the risk of misfires and missed detections. Attached Figure Description

[0048] Figure 1 This is a flowchart of the hypersonic vehicle target type identification method in a complex battlefield environment according to Embodiment 1 of the present invention;

[0049] Figure 2 This is a block diagram of the hypersonic vehicle target type identification system in a complex battlefield environment according to Embodiment 2 of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Before providing examples, it is necessary to describe the application scenarios of this invention. During the terminal guidance phase, hypersonic vehicles identify and filter target types of surface ships in typical complex battlefield environments such as coastal areas, straits, or ports. The geographical environment includes features such as coastlines, islands, reefs, docks, and near-shore buildings, which cause obstruction and clutter interference.

[0052] Example 1: As Figure 1 As shown in the figure, this embodiment provides a method for identifying the target type of hypersonic vehicles in complex battlefield environments. The method includes:

[0053] Step S1: The hypersonic vehicle acquires an image frame sequence via an imaging seeker during the terminal guidance phase; the image frame sequence is evaluated for image validity to obtain a corresponding validity marker sequence; based on the validity marker sequence, consecutive image frames marked as valid are divided into observation windows, and image frames marked as invalid are divided into invalid segments; the observation windows are arranged in chronological order to obtain an observation window sequence.

[0054] Step S2: During flight, based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle, predict the instantaneously accessible spatial domain that updates over time; discretize the instantaneously accessible spatial domain to obtain a set of spatial index units; based on the attitude data and line-of-sight data at the corresponding moment of each observation window, determine the projection unit of the corresponding imaging area in the spatial index unit; associate the observation window with the corresponding projection unit to obtain the observation record organized by the spatial index unit.

[0055] Step S3: Extract the target fragment image from the observation window associated with the spatial index unit within the imaging area; perform feature extraction on the target fragment image to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; perform multi-window update on the type confidence vector to obtain the updated type confidence vector.

[0056] Step S4: Determine the corresponding target type based on the updated type confidence vector of each spatial index unit; when the updated type confidence vector does not meet the preset judgment condition, the corresponding target type is determined as an undetermined type.

[0057] For example, consider the scenario of a hypersonic vehicle identifying target type in a coastal port area during the terminal guidance phase. The port area contains various types of ships, including destroyers, frigates, supply ships, cargo ships, and fishing boats. Some ships are obscured by port gantry cranes, container yards, and shoreline features. The vehicle enters the terminal guidance phase approximately 42 km from the port, at a speed of approximately Mach 6, and the terminal guidance phase lasts approximately 25 seconds. The imaging seeker operates at a frame rate of 25 frames per second.

[0058] After the hypersonic vehicle enters the terminal guidance phase, the imaging seeker begins operation. During the first 20 seconds of terminal guidance flight, the imaging seeker acquires 500 frames of infrared images at a frame rate of 25 frames per second. Image acquisition is stopped in the last 5 seconds to perform target type identification calculations, reserving computation time for terminal decision-making. Due to factors such as aerodynamic heating, attitude jitter, plasma sheath interference, and ground obstruction during hypersonic vehicle flight, not all image frames are usable for target identification. In the effectiveness evaluation of each image frame, various effectiveness indicators are calculated sequentially for frames 0 to 500. Taking frame 87 as an example, the sharpness index is measured as 0.68, the field-of-view ratio as 0.75, the brightness contrast ratio as 0.82, and the obstruction level as 0.15. These four indicators are weighted and summed according to preset effective weights, resulting in a comprehensive effectiveness index of 0.73 for the corresponding image frame. The validity index was averaged over a five-frame time window, and the stable validity index for frame 87 was calculated to be 0.71. This 0.71 is higher than the preset validity threshold of 0.55 (determined based on the minimum image quality required for the imaging seeker to reliably extract target features under typical interference conditions (such as aerodynamic heating and jitter), therefore frame 87 was marked as valid. After evaluating all frames using the same method, a complete validity labeling sequence was obtained, with 276 frames marked as valid and the remaining frames as invalid. Based on the validity labeling sequence, consecutively marked valid image frames were divided into nine observation windows, corresponding to different times within the first 20 seconds. There are eight invalid segments between these nine observation windows. The ten observation windows were arranged chronologically to form an observation window sequence.

[0059] During flight, the aircraft continuously collects real-time attitude, velocity, and acceleration data output by the inertial navigation system. Taking the 8.5s mark of final guidance as an example, the near boundary of the spatial envelope region is approximately 3000m from the aircraft, the far boundary is approximately 38km from the aircraft, the azimuth angle is approximately 12°, and the pitch angle is approximately 8°. This spatial envelope region is the immediately accessible spatial domain at the current moment.

[0060] The imaging seeker has an angular resolution of 0.05°, which determines the angular direction deviation step size. This is used to divide and crop the spatial index unit set, resulting in 31,456 spatial index units. For each observation window, the projection unit of the imaging region within the spatial index unit is determined based on the attitude and line-of-sight data at the corresponding time. Target fragment images are extracted from the observation windows associated with each spatial index unit, and feature extraction is performed to obtain fragment feature data. A preset type set includes six types: destroyer, frigate, supply ship, cargo ship, fishing vessel, and unknown target. A type confidence vector is established based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window. For example, the type confidence vector corresponding to the third observation window of spatial index unit number 17853 is: destroyer 0.30, frigate 0.24, supply ship 0.17, cargo ship 0.13, fishing vessel 0.09, and unknown target 0.07. Multi-window updates were performed on the four observation windows associated with spatial index unit 17853, resulting in the following updated type confidence vectors: destroyer 0.40, frigate 0.24, supply ship 0.15, cargo ship 0.11, fishing boat 0.07, and unknown target 0.03.

[0061] The updated type confidence vector was checked to see if it met the preset judgment criteria. The preset criteria required that the confidence value of the highest confidence type be greater than 0.35, and the difference between its confidence value and that of the second highest confidence type be greater than 0.1. The test showed that the destroyer's confidence value was 0.40, greater than the threshold of 0.35; and the difference between its confidence value and that of the second highest confidence type, the frigate, was 0.16, greater than the threshold of 0.1. Therefore, the target type corresponding to spatial index unit 17853 was determined to be a destroyer. Type identification of 17 main targets within the port area was completed before the end of the terminal guidance phase. Twelve target types were identified, including three destroyers, two frigates, one supply ship, three cargo ships, and three fishing boats; five targets were marked as undetermined due to insufficient observation information.

[0062] The method for evaluating the image frame sequence to obtain the corresponding validity marker sequence includes:

[0063] For each image frame, at least one of the following indicators is calculated to obtain a validity index: sharpness index, field of view occupancy index, brightness contrast index, and obstruction index; the validity index is subjected to sliding statistical processing within a preset time window to obtain a stable validity index; based on the comparison result between the stable validity index and a preset validity threshold, a validity label is generated for each image frame to obtain the validity label sequence.

[0064] For example, taking the 87th frame as an example, the sharpness index is calculated using the image gradient energy method, and the normalized sharpness index is 0.68. The field-of-view proportion index is calculated using the threshold segmentation method, which is 0.75. The brightness difference is calculated and then nonlinearly mapped to obtain the brightness contrast index, which is 0.82. The proportion of abnormal edges detected in the image is used to obtain the obstruction index, which is 0.15. Based on the importance of the image for target recognition, the sharpness index is weighted at 0.3, the field-of-view proportion index at 0.25, the brightness contrast index at 0.25, and the obstruction index at 0.2. The weighted summation yields the comprehensive effectiveness index of the 87th frame as 0.68×0.3+0.75×0.25+0.82×0.25+0.15×0.2≈0.63.

[0065] The preset time window length is 11 frames (the current frame and 5 frames before and after it). The comprehensive validity index sequence within the 87th frame window (frames 82 to 92) is: [0.73, 0.68, 0.72, 0.70, 0.72, 0.63, 0.69, 0.73, 0.71, 0.70, 0.70]. The average value of all indices within this window is calculated to be 0.71, which is used as the stable validity index for the 87th frame. The stable validity index of 0.71 for the 87th frame is greater than the preset validity threshold of 0.55, and is therefore marked as valid. After completing the validity marking for all frame images, a validity marking sequence is obtained.

[0066] The method for predicting the instantaneously reachable spatial domain updated over time based on the real-time attitude, velocity, and acceleration of a hypersonic vehicle includes:

[0067] The real-time attitude, velocity, and acceleration are subjected to low-pass filtering with a fixed time constant to obtain filtered state variables. Based on the filtered state variables and the maximum normal overload and maximum tangential overload of the aircraft, the heading change range, pitch change range, and velocity change range of the aircraft within a preset time step are calculated.

[0068] Based on the range of heading changes, pitch changes, velocity changes, and the effective range parameters of the imaging seeker, a spatial envelope region with the current position of the aircraft as the reference is determined; the spatial envelope region is then used as the instantly accessible spatial domain.

[0069] For example, at the 8.5-second mark of final guidance, the original state variables output by the inertial navigation system are: heading angle -135.7°, pitch angle -18.3°, roll angle 2.1°, flight speed 1960 m / s, normal acceleration -12.5 m / s², and tangential acceleration -3.8 m / s². A first-order low-pass filter is applied to these state variables, with a time constant of 0.2 s, a sampling period of 0.04 s, and a filter coefficient of 0.818. The filtered state variables are: pitch angle -18.1°, roll angle 1.9°, speed 1965 m / s, normal acceleration -11.8 m / s², and tangential acceleration -3.5 m / s².

[0070] Taking the gravitational acceleration g = 9.8 m / s², the current normal overload is approximately -11.8 / 9.8 ≈ -1.20. The preset time step is 0.5 seconds. Based on the aircraft's maximum normal overload of ±5g and the current velocity, the maximum rate of change of the yaw angle within 0.5 seconds is calculated to be approximately ±4.6° / s, and the maximum rate of change of the pitch angle is approximately ±3.6° / s. Therefore, the range of yaw angle change is approximately ±(4.6 × 0.5) = ±2.3°, and the range of pitch angle change is approximately ±(3.6 × 0.5) = ±1.8°. Simultaneously, based on the maximum tangential overload and the current tangential acceleration, the velocity change range within 0.5 seconds is estimated to be approximately -10 m / s to +5 m / s. Combining these ranges with the effective range parameters of the imaging seeker (3 km to 38 km), the spatial envelope region is determined. Using the filtered heading angle of -135.7° as a baseline, expanding the heading range horizontally (±2.3°) yields an azimuth boundary of approximately -138.0° to -133.4°. Using the filtered pitch angle of -18.1° as a baseline, expanding the pitch range vertically (±1.8°) yields a pitch boundary of approximately -19.9° to -16.3°. The radial distance boundary is directly determined by the seeker's effective range parameters, ranging from 3km to 38km. This forms a frustum-shaped spatial envelope region with the aircraft's current position as its apex, covering the aforementioned ranges in azimuth, pitch, and radial directions. This region is the predicted instantaneous reachable space domain at the current moment. This space domain will be updated in real-time according to the aircraft's flight status.

[0071] The method for discretizing the instantaneously accessible spatial domain to obtain a set of spatial index cells includes:

[0072] The angular step length is determined based on the angular resolution parameters of the imaging seeker, and the radial step length is determined based on the radial range of the instantaneously accessible spatial domain. Based on the angular step length and the radial step length, the instantaneously accessible spatial domain is divided into a three-dimensional mesh to obtain an initial set of spatial units.

[0073] Based on the boundary set consisting of the reachable boundary determined by the aircraft's maneuverability and the ground feature boundary determined by environmental data, the initial spatial units are trimmed to obtain the spatial index unit set.

[0074] For example, the instantaneous reachable space domain (azimuth approximately -138.0° to -133.4°, elevation approximately -19.9° to -16.3°, radial distance 3km to 38km) predicted at the 8.5-second mark of terminal guidance is discretized. The angular resolution parameter of the imaging seeker is 0.0125°. The angular direction deviation distances corresponding to the 3-8km, 8-15km, 15-25km, and 25-38km ranges are 0.05°, 0.0625°, 0.075°, and 0.0875°, respectively, and the radial deviation distances are 165m, 345m, 600m, and 945m, respectively. The number of grid points in the azimuth direction is approximately |133.4-138.0|÷0.05≈92; the number of grid points in the pitch direction is approximately |16.3-19.9|÷0.05≈72; and the number of grid points in the radial direction is approximately ((8000-3000)÷165+(15000-8000)÷345+(25000-15000)÷600+(38000-25000)÷945) rounded down and summed ≈81. Therefore, the total number of initial spatial units is approximately 92×72×81=536,544.

[0075] The initial spatial units are trimmed based on the boundary set, eliminating units from which targets cannot be observed. The boundary set consists of reachable boundaries and terrain boundaries. Reachable boundaries are determined by the aircraft's maneuverability, eliminating spatial units from which the target cannot be brought into the field of view at the current moment through maneuvering. Terrain boundaries are determined by environmental data. Pre-stored battlefield environment digital elevation model (DEM) data is loaded, and the geographic coordinates and elevation of the center point of each initial spatial unit are calculated to determine whether it is located on land (based on geographic information) or whether it is obstructed by terrain along the line connecting the aircraft and the unit (based on line-of-sight algorithm). If the unit center is located on land or obstructed by terrain, it is eliminated. There is some overlap between the two types of boundary trimming, resulting in 31,456 units. Each spatial index unit is identified by a three-dimensional index number, with the numbering format being a combination of azimuth, elevation, and radial indices. For example, number 17853 corresponds to azimuth index 178, elevation index 53, and radial index calculated according to the numbering rules.

[0076] The method for determining the angular step length based on the angular resolution parameters of the imaging seeker, and determining the radial step length based on the radial range of the instantaneously reachable spatial domain, includes:

[0077] The instantaneously accessible spatial domain is divided radially to obtain a set of distance intervals, and the boundaries of the distance intervals are determined based on the resolution of the imaging seeker at different distances.

[0078] The angular deviation length is calculated based on the center distance value of the distance interval and the angular resolution parameters of the imaging seeker; the radial deviation length is calculated based on the range of the distance interval and the preset radial resolution coefficient.

[0079] For example, distance ranges are divided based on the ground resolution capability of the imaging seeker at different distances. The angular resolution of the imaging seeker is 0.0125° (approximately 0.000218 radians), and the ground resolution is approximately distance × 0.000218. Target recognition requires that the minimum projected size of the target in the image covers at least 3 pixels. For example, if the minimum target feature size to be recognized is 7 meters (such as the width of a fishing boat), then the ground resolution corresponding to each pixel should not be greater than 7 / 3 ≈ 2.33 meters. The maximum distance that meets this resolution requirement is calculated to be 2.33 / 0.000218 ≈ 10688 meters. Based on this, and considering the distance range of the mission's focus, the radial range of the immediately accessible spatial domain (3km to 38km) is divided into four intervals: the first interval (3-8km, ground resolution 0.65-1.74m, meeting high resolution requirements), the second interval (8-15km, ground resolution 1.74-3.27m, partially meeting requirements), the third interval (15-25km, ground resolution 3.27-5.45m, lower resolution), and the fourth interval (25-38km, ground resolution 5.45-8.28m, low resolution).

[0080] Calculate the angular step size for each distance interval. The selection of the angular step size needs to balance the accuracy of the spatial index and the total number of cells. In the short-distance interval, it is desirable to maintain a high angular resolution, so a smaller angular step size multiple is used; in the long-distance interval, to accommodate the total number of control units, the angular resolution can be appropriately relaxed, and a larger multiple can be used. Set the angular step size multiple to 4 times for the first interval, 5 times for the second interval, 6 times for the third interval, and 7 times for the fourth interval. Then the angular step size for the first interval is 0.0125° × 4 = 0.05°, for the second interval it is 0.0625°, for the third interval it is 0.075°, and for the fourth interval it is 0.0875°.

[0081] With a preset radial resolution coefficient of 0.03, the radial distance step length = interval center distance × 0.03. Therefore, the center distance of the first interval is (3+8) / 2 = 5.5 km, and the radial distance step length is 5.5 × 0.03 = 165 m; the center distance of the second interval is 11.5 km, and the radial distance step length is 345 m; the center distance of the third interval is 20 km, and the radial distance step length is 600 m; the center distance of the fourth interval is 31.5 km, and the radial distance step length is 945 m. The corresponding angular distance step lengths for each distance interval are (0.05°, 0.0625°, 0.075°, 0.0875°) and radial distance step lengths are (165 m, 345 m, 600 m, 945 m).

[0082] The method for determining the projection unit of the corresponding imaging region in the spatial index unit based on the attitude data and line-of-sight data at the corresponding time of each observation window includes:

[0083] The orientation vector of the optical axis of the imaging seeker in the reference coordinate system is determined based on the attitude data; the field of view geometry is determined based on the line of sight data and the field of view angle parameters of the imaging seeker.

[0084] Determine whether each spatial index unit in the set of spatial index units intersects with the view frustum geometry, and identify the intersecting spatial index units as the projection units of the current observation window; associate and store the time information and validity marker of the observation window with the projection units.

[0085] For example, taking the fifth observation window as an example, the center time is approximately 8.5 seconds into the final guidance phase. Based on the aircraft's attitude data at this moment (heading -135.7°, pitch -18.1°, roll 1.9°) and the seeker's mounting angle, the pointing vector of the imaging seeker's optical axis in the northeast-northeast coordinate system is determined through coordinate transformation. Based on the pointing vector, the missile's current position, and the imaging seeker's field of view parameters (horizontal 8°, vertical 6°), a field of view cone is constructed with the missile as the vertex, its axis along the optical axis, a horizontal angle of 8°, a vertical angle of 6°, a near-field section at 3km, and a far-field section at 38km.

[0086] The set of spatial index cells is traversed. For one spatial index cell (e.g., whose theoretical center position in the spacecraft coordinate system has an azimuth of -136.0°, an elevation of -17.8°, and a radial distance of 12km), the cell's index is converted to azimuth and elevation angles relative to the spacecraft. The azimuth of -136.0° falls within the range of the frustum azimuth (optical axis azimuth -135.7° ± 4°, i.e., -139.7° to -131.7°), the elevation of -17.8° falls within the range of the frustum elevation (optical axis elevation -18.1° ± 3°, i.e., -21.1° to -15.1°), and the radial distance of 12km lies between the near and far sections of the frustum (3-38km). Therefore, it is determined to intersect with the current frustum and is identified as a projection cell for the fifth observation window.

[0087] The observation time and validity marker of the fifth observation window are associated and stored with the identifier (e.g., index number) of the spatial index unit. This method is used to traverse all spatial index units, determining that the fifth observation window is associated with a total of 873 spatial index units. All associated records are organized by spatial index unit, and the same spatial index unit can be associated with multiple observation records from different observation windows.

[0088] The method for establishing a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window includes:

[0089] Obtain a preset type set; for each type in the preset type set, calculate the feature similarity between the reference feature data and the fragment feature data; determine the observation quality coefficient of the current observation window based on the validity labeling result.

[0090] Calculate the segment weight parameters for each type based on the observation quality coefficient and feature similarity; then, sum the initial confidence parameters for each type based on the segment weight parameters to obtain the type confidence vector.

[0091] For example, based on mission requirements and target area intelligence, the preset type set includes six ship types: destroyer, frigate, supply ship, cargo ship, fishing vessel, and unknown target. Each type corresponds to a set of reference feature data, stored in the airborne feature database. For each type in the preset type set, the feature similarity between the reference feature data and the fragment feature data is calculated. Taking the fragment feature data of the third observation window of spatial index unit number 17853 as an example, this fragment feature data includes: target length projection value of 115 meters, width projection value of 22 meters, length-to-width ratio of 5.23, outline compactness of 0.87, relative thermal radiation intensity of the funnel area of ​​1.85, relative thermal radiation intensity of the stern of 1.23, and relative thermal radiation intensity of the midships of 1.05. The similarity to the destroyer reference features is calculated, including length similarity: the segment length of 115 meters is compared with the reference range of 145 to 155 meters. The deviation is the absolute value of the difference between the segment feature value and the reference feature value (taking the median of the corresponding reference range), (145+155)÷2-115=35 meters. The normalized similarity is calculated using the Gaussian mapping function (formula: normalized similarity = e(-(deviation / scale parameter)^2)). The scale parameter is preset based on the importance of feature recognition and prior knowledge. In this embodiment, the length scale parameter is 72. The normalized length similarity is e(-(35÷72)^2)≈0.78. Width similarity: the segment width is 22m, the reference range is 18 to 22m, the width scale parameter is 20m, and the normalized width similarity is approximately 1.0. Aspect Ratio Similarity: Segment aspect ratio 5.23, reference range 6.8 to 7.5, scale parameter 2.92, normalized aspect ratio similarity approximately 0.65. Contour Compactness Similarity: Segment value 0.87, reference range 0.85 to 0.92, scale parameter 0.066, normalized contour compactness similarity approximately 0.95. Chimney Area Thermal Radiation Similarity: Segment value 1.85, reference range 1.7 to 2.1, scale parameter 0.173, normalized chimney area thermal radiation similarity approximately 0.92. Tail Thermal Radiation Similarity: Segment value 1.23, reference range 1.1 to 1.4, scale parameter 0.074, normalized tail thermal radiation similarity approximately 0.93.

[0092] The similarity scores of each component were weighted and summed. Based on prior knowledge of the contribution of each feature to distinguishing ship types, the weights of each component similarity were determined as follows: length 0.2, width 0.15, aspect ratio 0.2, profile compactness 0.15, heat radiation from the funnel area 0.15, and heat radiation from the stern 0.15. The overall feature similarity with destroyers was approximately 0.78×0.2 + 1.0×0.15 + 0.65×0.2 + 0.95×0.15 + 0.92×0.15 + 0.93×0.15 ≈ 0.86. The feature similarity with frigates was calculated similarly: the length deviation was small, but the width was larger, and the aspect ratio deviation was significant, resulting in an overall similarity of 0.65. The feature similarity with supply ships was: the length was significantly smaller, but the width was similar, resulting in an overall similarity of 0.41. Similarity to cargo ships: Wide size range but mismatched thermal radiation characteristics, overall similarity is 0.28. Similarity to fishing vessels: Significant size difference, overall similarity is 0.12. The feature similarity for unknown targets is fixed at 0.05, representing the lowest similarity benchmark.

[0093] The observation quality coefficient is determined based on the relationship between the stability validity index and the validity threshold. When the stability validity index is greater than or equal to the validity threshold of 0.55, the observation quality coefficient is calculated as follows: subtract the validity threshold from the stability validity index, divide by 1 minus the validity threshold, multiply by 0.5, and finally add 0.5. For example, if the validity of the third observation window is marked as valid, the corresponding stability validity index is 0.71, which is greater than the preset validity threshold of 0.55. Then, the observation quality coefficient of the third observation window is calculated using the following formula: 0.5 + 0.5 × (stability validity index - preset validity threshold) / (1 - preset validity threshold). The product of the observation quality coefficient and the feature similarity of each type is defined as the segment weight parameter for each type. The calculated segment weight parameters are: destroyer: 0.86 × 0.68 = 0.5848; frigate: 0.65 × 0.68 = 0.4420; supply ship: 0.41 × 0.68 = 0.2788; cargo ship: 0.28 × 0.68 = 0.1904; fishing boat: 0.12 × 0.68 = 0.0816; and unknown target: 0.05 × 0.68 = 0.0340. The sum of the segment weight parameters for the six types is 0.5848 + 0.4420 + 0.2788 + 0.1904 + 0.0816 + 0.0340 = 1.6116. After normalization, the weight parameters of each type of fragment are approximately: destroyer 0.363, frigate 0.274, supply ship 0.173, cargo ship 0.118, fishing boat 0.051, and unknown target 0.021.

[0094] The initial confidence parameters for each type are weighted and summed based on the fragment weight parameters. The initial confidence parameters are uniformly distributed, with each of the six types having a value of 0.167. The confidence of each type is equal to the normalized fragment weight parameter multiplied by the update factor, plus the initial confidence parameter multiplied by the retention factor. The update factor is adjusted according to the degree of dependence on observation information at each mission stage; in this embodiment, it is set to 0.7, and the retention factor is 1-0.7=0.3. The confidence of the destroyer type is 0.363×0.7+0.167×0.3≈0.30. The remaining types are calculated in the same way. After calculation, the values ​​are normalized again, resulting in the following type confidence vectors: destroyer 0.30, frigate 0.24, supply ship 0.17, cargo ship 0.13, fishing boat 0.09, and unknown target 0.07.

[0095] The method for performing multi-window updates on the type confidence vector to obtain the updated type confidence vector includes:

[0096] For all observation windows associated with the same spatial index unit, the observation window sequence is obtained by arranging them in the order of observation time; the basic update weight is determined based on the validity marking results of each observation window; and the time decay factor is determined based on the time interval between each observation window and the current time.

[0097] The comprehensive update weight of each observation window is calculated based on the basic update weight and the time decay factor; the segment feature data corresponding to each observation window are weighted and fused according to the comprehensive update weight to obtain fused feature data.

[0098] The type confidence vector is cumulatively updated and normalized based on the fused feature data to obtain the updated type confidence vector.

[0099] For example, spatial index cell number 17853 is covered by four observation windows: the third, fifth, seventh, and ninth. These four observation windows are arranged chronologically as follows: third window, fifth window, seventh window, and ninth window. The center times of each window are: 4.0 seconds after the end of guidance in the third window, 8.5 seconds after the end of guidance in the fifth window, 14.2 seconds after the end of guidance in the seventh window, and 21.0 seconds after the end of guidance in the ninth window. Each observation window has an independent type confidence vector, and all four observation windows are marked as valid. For observation windows marked as valid, the base update weight is the corresponding observation quality coefficient; for invalid observation windows, the base update weight is 0. In this example, the observation quality coefficient of the third window is 0.68, and the base update weight is 0.68; the observation quality coefficient of the fifth window is 0.73, and the base update weight is 0.73; the observation quality coefficient of the seventh window is 0.71, and the base update weight is 0.71; and the observation quality coefficient of the ninth window is 0.75, and the base update weight is 0.75. The higher the weight of the basic update, the better the quality.

[0100] The time decay factor reflects the degree of decay of observed information over time; the earlier the observation window, the more severe the decay. Assuming the current time is the 22.0 second of final guidance, the time decay factor is calculated using the natural constant e as the base and the negative time interval divided by the decay time constant (set to 10 s) as the exponent. The calculated decay factors are: 18.0 s for the third window (approximately 0.165); 13.5 s for the fifth window (approximately 0.259); 7.8 s for the seventh window (approximately 0.458); and 1.0 s for the ninth window (approximately 0.905). The overall update weight equals the base update weight multiplied by the time decay factor. The overall update weight for the third window is approximately 0.112, for the fifth window approximately 0.189, for the seventh window approximately 0.325, and for the ninth window approximately 0.679. The sum of the overall update weights is 1.305.

[0101] The fragment feature data corresponding to each observation window are weighted and fused according to the comprehensive update weights, resulting in fused feature data with a width projection value of 22.7 meters, an aspect ratio of 5.13, a contour compactness of 0.86, a relative thermal radiation intensity of 1.83 for the chimney area, and a relative thermal radiation intensity of 1.22 for the tail. The similarity to each type of reference feature is recalculated based on the fused feature data, using the same method as when establishing the confidence vector for a single observation window type. The similarity is 0.83 with destroyers, 0.62 with frigates, 0.38 with supply ships, 0.25 with cargo ships, 0.10 with fishing boats, and 0.05 with unknown targets. Cumulative updates use a weighted summation method. The cumulative confidence for each type is equal to the sum of the confidence for each observation window type multiplied by the corresponding comprehensive update weight. The type confidence vectors for the four observation windows are as follows: Third window: [Destroyer 0.30, Frigate 0.24, Supply ship 0.17, Cargo ship 0.13, Fishing boat 0.09, Unknown target 0.07]; Fifth window: [Destroyer 0.35, Frigate 0.26, Supply ship 0.16, Cargo ship 0.12, Fishing boat 0.08, Unknown target 0.03]; Seventh window: [Destroyer 0.39, Frigate 0.25, Supply ship 0.15, Cargo ship 0.11, Fishing boat 0.07, Unknown target 0.03]; Ninth window: [Destroyer 0.44, Frigate 0.23, Supply ship 0.14, Cargo ship 0.10, Fishing boat 0.06, Unknown target 0.03]. The overall update weights for each window are: 0.112 for the third window, 0.189 for the fifth window, 0.325 for the seventh window, and 0.679 for the ninth window.

[0102] The cumulative confidence scores are: destroyer 0.30×0.112+0.35×0.189+0.39×0.325+0.44×0.679=0.52526, frigate 0.31344, supply ship 0.19309, cargo ship 0.14089, fishing boat 0.08869, and unknown target 0.04363. The sum of the cumulative confidence scores is 1.305. Normalization (dividing by the sum of the cumulative confidence scores of 1.305) yields the following updated type confidence vectors: destroyer 0.40, frigate 0.24, supply ship 0.15, cargo ship 0.11, fishing boat 0.07, and unknown target 0.03.

[0103] The method for weighted fusing of segment feature data corresponding to each observation window based on the comprehensive update weight to obtain fused feature data includes:

[0104] Based on the aircraft attitude and position data at the corresponding time of each observation window, the observation angle parameters of each observation window are calculated; based on the observation angle parameters of each observation window, the angle difference value between any two observation windows is calculated; observation windows with angle difference values ​​less than a preset angle consistency threshold are assigned to the same angle group to obtain the angle group set.

[0105] For each viewpoint group, the segment feature data is weighted and fused within the group according to the comprehensive update weight of each observation window to obtain the fused feature data within the group.

[0106] In the case of multiple view groups, the fused feature data within each view group is spliced ​​together or independently retained according to the weight of each view group to obtain fused feature data, which includes view identification information.

[0107] For example, spatial index cell number 17853 is covered by four observation windows: the third, fifth, seventh, and ninth. The overall update weights for each window are: 0.112 for the third window, 0.189 for the fifth window, 0.325 for the seventh window, and 0.679 for the ninth window. The observation perspective parameters include azimuth and pitch angles, characterizing the observation geometry of the aircraft relative to the target. The target is located in the port berth area, with the bow pointing 35° east of north. Calculate the azimuth observation angles relative to the target's bow for each window. The azimuth observation angle for the third window is +11.2° (starboard stern direction), and the pitch angle is -16.2°; the azimuth observation angle for the fifth window is +8.5°, and the pitch angle is -21.3°; the azimuth observation angle for the seventh window is +2.3°, and the pitch angle is -28.6°; and the azimuth observation angle for the ninth window is -11.3° (port stern direction), and the pitch angle is -35.8°. Calculate the difference in viewing angles between any two observation windows. Considering both the azimuth and pitch angle differences, take the square root of the sum of the squares of the absolute values ​​of the azimuth and pitch angle differences. For example, the combined visual angle difference between the third and fifth observation windows is approximately 5.77°, between the third and seventh observation windows is approximately 15.26°, between the third and ninth observation windows is approximately 29.84°, between the fifth and seventh observation windows is approximately 9.58°, between the fifth and ninth observation windows is approximately 24.54°, and between the seventh and ninth observation windows is approximately 15.39°.

[0108] Observation windows with viewing angle differences less than a preset viewing angle consistency threshold are grouped into the same viewing angle group. The preset viewing angle consistency threshold is set at 10° based on the sensitivity of target features to changes in viewing angle. The third and fifth observation windows have viewing angle differences less than the threshold and are therefore grouped into the same viewing angle group. The fifth and seventh observation windows also have viewing angle differences less than the threshold, and the seventh observation window is also grouped into this group. The third and seventh observation windows have a viewing angle difference of 15.26°, but because they both belong to the same viewing angle group as the fifth observation window, they remain in the same group. The ninth observation window has a viewing angle difference greater than 10° with the other three observation windows and is grouped into a separate viewing angle group. Viewing angle group one includes the third, fifth, and seventh observation windows, representing the starboard aft viewing angle. Viewing angle group two includes only the ninth observation window, representing the port aft viewing angle.

[0109] Taking the width projection value feature as an example, view group one includes the third, fifth, and seventh windows. Its intra-group comprehensive update weight sum is: 0.112 + 0.189 + 0.325 = 0.626. The width of the third window is 22.0m, the fifth window is 22.5m, and the seventh window is 22.3m. The intra-group weighted fused width is (0.112 × 22.0 + 0.189 × 22.5 + 0.325 × 22.3) ÷ 0.626 ≈ 22.31m. View group two only includes the ninth window. The intra-group weight sum is the comprehensive update weight of this window, 0.679. The intra-group fused feature data is the original fragment feature data of this window, for example, a width value of 23.1 meters. The weight sums of the two view groups are 0.626 for view group one and 0.679 for view group two. The total weight sum is 1.305. For cases requiring the generation of a single fusion vector, the fusion features within a group can be weighted twice according to group weights. For example, the final fusion result for the width value is (0.626×22.31+0.679×23.1)÷1.305≈22.7 meters. The final generated fusion feature data will include the fusion values ​​of various features calculated using this method, namely, a width of 22.7 meters, an aspect ratio of 5.13, a profile compactness of 0.86, a relative thermal radiation intensity of 1.83 for the chimney area, and a relative thermal radiation intensity of 1.22 for the tail. These fusion values ​​can be distinguished by data identifiers as results of multi-view synthesis.

[0110] Example 2: Based on the same inventive concept, such as Figure 2 As shown, this embodiment also provides a hypersonic vehicle target type identification system in complex battlefield environments, the system comprising:

[0111] The observation window division module is used for hypersonic vehicles to acquire image frame sequences through imaging seekers during the terminal guidance phase; to perform image validity evaluation on the image frame sequences to obtain corresponding validity marker sequences; to divide consecutive image frames marked as valid into observation windows and image frames marked as invalid into invalid segments according to the validity marker sequences; and to arrange the observation windows in chronological order to obtain an observation window sequence.

[0112] The spatial indexing and association module is used to predict the instantaneously accessible spatial domain updated over time based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle during flight; discretize the instantaneously accessible spatial domain to obtain a set of spatial indexing units; determine the projection unit of the corresponding imaging area in the spatial indexing unit based on the attitude data and line-of-sight data at the corresponding time of each observation window; and associate the observation window with the corresponding projection unit to obtain the observation record organized by the spatial indexing unit.

[0113] A multi-window feature fusion module is used to extract target fragment images from the observation window associated with the spatial index unit within the imaging region; perform feature extraction on the target fragment images to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; and perform multi-window update on the type confidence vector to obtain an updated type confidence vector.

[0114] The target type decision module is used to determine the corresponding target type based on the updated type confidence vector of each spatial index unit; when the updated type confidence vector does not meet the preset judgment conditions, the corresponding target type is determined as an undetermined type.

[0115] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0116] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying the type of hypersonic aircraft targets in complex battlefield environments, characterized in that, The method includes: In the terminal guidance phase, the hypersonic vehicle acquires an image frame sequence via an imaging seeker; the image frame sequence is evaluated for image validity to obtain a corresponding validity marker sequence; based on the validity marker sequence, consecutive image frames marked as valid are divided into observation windows, and image frames marked as invalid are divided into invalid segments; the observation windows are arranged in chronological order to obtain an observation window sequence. During flight, based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle, the instantaneous reachable spatial domain updated over time is predicted; the instantaneous reachable spatial domain is discretized to obtain a set of spatial index units; based on the attitude data and line-of-sight data at the corresponding moment of each observation window, the projection unit of the corresponding imaging area in the spatial index unit is determined; the observation window is associated with the corresponding projection unit to obtain the observation record organized by the spatial index unit. From the observation window associated with the spatial index unit, extract the target fragment image within the imaging region; perform feature extraction on the target fragment image to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; update the type confidence vector through multiple windows to obtain the updated type confidence vector; Based on the updated type confidence vector of each spatial index unit, the corresponding target type is determined; when the updated type confidence vector does not meet the preset judgment condition, the corresponding target type is determined as an undetermined type.

2. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 1, characterized in that, The method for evaluating the image frame sequence to obtain the corresponding validity marker sequence includes: For each image frame, at least one of the following indicators is calculated to obtain a validity index: sharpness index, field of view occupancy index, brightness contrast index, and obstruction index; the validity index is subjected to sliding statistical processing within a preset time window to obtain a stable validity index; based on the comparison result between the stable validity index and a preset validity threshold, a validity label is generated for each image frame to obtain the validity label sequence.

3. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 2, characterized in that, The method for predicting the instantaneously reachable spatial domain updated over time based on the real-time attitude, velocity, and acceleration of a hypersonic vehicle includes: The real-time attitude, velocity, and acceleration are subjected to low-pass filtering with a fixed time constant to obtain filtered state variables. Based on the filtered state variables and the maximum normal overload and maximum tangential overload of the aircraft, the heading change range, pitch change range, and velocity change range of the aircraft within a preset time step are calculated. Based on the range of heading changes, pitch changes, velocity changes, and the effective range parameters of the imaging seeker, a spatial envelope region with the current position of the aircraft as the reference is determined; the spatial envelope region is then used as the instantly accessible spatial domain.

4. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 3, characterized in that, The method for discretizing the instantaneously accessible spatial domain to obtain a set of spatial index cells includes: The angular step length is determined based on the angular resolution parameters of the imaging seeker, and the radial step length is determined based on the radial range of the instantaneously accessible spatial domain. Based on the angular step length and the radial step length, the instantaneously accessible spatial domain is divided into a three-dimensional grid to obtain an initial set of spatial units. Based on the boundary set consisting of the reachable boundary determined by the aircraft's maneuverability and the ground feature boundary determined by environmental data, the initial spatial units are trimmed to obtain the spatial index unit set.

5. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 4, characterized in that, The method for determining the angular step length based on the angular resolution parameters of the imaging seeker, and determining the radial step length based on the radial range of the instantaneously reachable spatial domain, includes: The instantaneously accessible spatial domain is divided along the radial direction to obtain a set of distance intervals, and the boundaries of the distance intervals are determined according to the resolution of the imaging seeker at different distances. The angular deviation length is calculated based on the center distance value of the distance interval and the angular resolution parameters of the imaging seeker; the radial deviation length is calculated based on the range of the distance interval and the preset radial resolution coefficient.

6. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 5, characterized in that, The method for determining the projection unit of the corresponding imaging region in the spatial index unit based on the attitude data and line-of-sight data at the corresponding time of each observation window includes: The orientation vector of the optical axis of the imaging seeker in the reference coordinate system is determined based on the attitude data; the field of view geometry is determined based on the line of sight data and the field of view angle parameters of the imaging seeker. Determine whether each spatial index unit in the set of spatial index units intersects with the view frustum geometry, and identify the intersecting spatial index units as the projection units of the current observation window; associate and store the time information and validity marker of the observation window with the projection units.

7. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 6, characterized in that, The method for establishing a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window includes: Obtain a preset type set; for each type in the preset type set, calculate the feature similarity between the reference feature data and the fragment feature data; determine the observation quality coefficient of the current observation window based on the validity labeling result; Calculate the segment weight parameters for each type based on the observation quality coefficient and feature similarity; then, sum the initial confidence parameters for each type based on the segment weight parameters to obtain the type confidence vector.

8. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 7, characterized in that, The method for performing multi-window updates on the type confidence vector to obtain the updated type confidence vector includes: For all observation windows associated with the same spatial index unit, arrange them in order of observation time to obtain the observation window sequence; determine the basic update weight based on the validity marking results of each observation window; determine the time decay factor based on the time interval between each observation window and the current time. The comprehensive update weight of each observation window is calculated based on the basic update weight and the time decay factor; the segment feature data corresponding to each observation window are weighted and fused according to the comprehensive update weight to obtain fused feature data. The type confidence vector is cumulatively updated and normalized based on the fused feature data to obtain the updated type confidence vector.

9. The method for identifying hypersonic vehicle target types in complex battlefield environments according to claim 8, characterized in that, The method for weighted fusing of segment feature data corresponding to each observation window based on the comprehensive update weight to obtain fused feature data includes: Based on the aircraft attitude and position data at the corresponding time of each observation window, the observation angle parameters of each observation window are calculated; based on the observation angle parameters of each observation window, the angle difference value between any two observation windows is calculated; observation windows with angle difference values ​​less than a preset angle consistency threshold are divided into the same angle group to obtain the angle group set. For each viewpoint group, the segment feature data is weighted and fused within the group according to the comprehensive update weight of each observation window to obtain the fused feature data within the group. In the case of multiple view groups, the fused feature data within each view group is spliced ​​together or independently retained according to the weight of each view group to obtain fused feature data, which includes view identification information.

10. A hypersonic vehicle target type identification system for complex battlefield environments, characterized in that, The system includes: The observation window division module is used for hypersonic vehicles to acquire image frame sequences via an imaging seeker during the terminal guidance phase; to perform image validity evaluation on the image frame sequences to obtain corresponding validity marker sequences; to divide consecutive image frames marked as valid into observation windows and image frames marked as invalid into invalid segments according to the validity marker sequences; and to arrange the observation windows in chronological order to obtain an observation window sequence. The spatial indexing and association module is used to predict the instantaneously accessible spatial domain updated over time based on the real-time attitude, velocity, and acceleration of the hypersonic vehicle during flight; discretize the instantaneously accessible spatial domain to obtain a set of spatial indexing units; determine the projection unit of the corresponding imaging area in the spatial indexing unit based on the attitude data and line-of-sight data at the corresponding time of each observation window; and associate the observation window with the corresponding projection unit to obtain the observation record organized by the spatial indexing unit. A multi-window feature fusion module is used to extract target fragment images from the observation window associated with the spatial index unit within the imaging region; perform feature extraction on the target fragment images to obtain fragment feature data, the fragment feature data including at least one of contour features, brightness distribution features, and thermal radiation distribution features; establish a type confidence vector based on the fragment feature data corresponding to the spatial index unit and the validity label of the observation window; and perform multi-window update on the type confidence vector to obtain an updated type confidence vector. The target type decision module is used to determine the corresponding target type based on the updated type confidence vector of each spatial index unit; when the updated type confidence vector does not meet the preset judgment conditions, the corresponding target type is determined as an undetermined type.

Images