A high signal-to-noise ratio downsampling method and system for a spaceborne photoelectric event stream
Patent Information
- Application Number
- CN202610939185.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-26
- Publication Date
- 2026-09-22
AI Technical Summary
[0009]为此，本发明实施例提供一种面向星载光电事件流的高信噪比降采样方法及系统，以解决在星载计算能力与下行遥测带宽极度受限的条件下，如何从海量异步事件流中高信噪比地提取目标运动状态信息并实现极低带宽精简传输的技术问题
[0030]本发明实施例接收空间传感器生成的宽视场原始异步事件流；在硬件底层或边缘节点执行时空相关性初筛，剔除孤立的本底噪声事件，将符合预设时空分布规律的局部事件聚集为疑似目标簇；将疑似目标簇输入动态局部波门，引入多模型状态观测器提取目标的运动学抽象特征；抛弃局部波门内的原始物理像素坐标，生成且仅下发包含目标角位置、多阶运动导数及协方差置信度的精简结构体。本发明通过将海量离散光子坐标抽象为低维高阶运动学矢量，在星载计算能力与下行遥测带宽极度受限条件下，实现了高信噪比的数据降采样与极低带宽精简传输，有效解决了星地数据传输瓶颈问题。
Smart Images
Figure CN122802657A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the fields of spaceborne data processing and space-to-ground communication technology, specifically to a high signal-to-noise ratio downsampling method and system for spaceborne optoelectronic event streams. Background Technology
[0002] Asynchronous vision sensors (also known as event cameras or dynamic vision sensors, DVS) are biomimetic vision sensors inspired by biological vision systems. Unlike traditional frame cameras that acquire entire images at fixed time intervals, event cameras respond independently to changes in light intensity at the pixel level. When the logarithmic change in light intensity at a pixel exceeds a preset threshold, that pixel outputs an event containing information such as pixel coordinates, a timestamp, and the polarity of the change. This asynchronous, sparse output mechanism gives event cameras a series of advantages that traditional sensors cannot match: microsecond-level temporal resolution (response latency as low as 10 μs), a high dynamic range exceeding 120 dB, extremely low power consumption (approximately 10 mW), and low data redundancy. Furthermore, event cameras do not suffer from motion blur, making them particularly suitable for capturing high-speed moving targets.
[0003] In recent years, event cameras have been widely researched and applied in ground-based fields such as robotics, autonomous driving, and drones. Simultaneously, their application scenarios are rapidly expanding from ground systems to space platforms. In space missions such as space situational awareness, space debris monitoring, high-dynamic star sensors, and on-orbit target tracking, event cameras, with their high temporal resolution and high dynamic range, are considered to effectively solve the bottleneck problems of traditional frame-based sensors, such as motion blur and insufficient dynamic range under high dynamic conditions. Existing research has applied event cameras to scenarios such as space debris monitoring, tracking of non-cooperative targets in space, and detection of weak targets in deep space. Organizations such as the European Space Agency (ESA) have also launched research programs to use dynamic visual sensors for high-speed space imaging. It can be said that space visual perception based on event cameras is becoming an important development direction in the field of aerospace remote sensing and situational awareness.
[0004] However, deploying event cameras on spaceborne platforms still faces significant technical challenges. First, event cameras generate massive asynchronous event streams—with event output rates reaching millions of events per second. Although the data volume of a single event is far less than that of a complete image frame, the accumulated number of events during continuous observation is still enormous. Second, the on-orbit computing power of satellite platforms is extremely limited. Onboard processors are severely constrained by power consumption, heat dissipation, size, and the space radiation environment, making it difficult to support real-time, high-intensity on-orbit processing of massive event streams. Third, satellite downlink telemetry communication bandwidth is extremely limited. The capacity of the most advanced current satellite-to-ground data transmission channels is far from sufficient to support the complete downlink transmission of raw event stream data. Taking a spaceborne CCD camera as an example, its post-imaging data transmission rate can reach several Gbps, while the actual usable downlink bandwidth is only a small fraction of that. Similarly, planetary and Earth observation missions are also subject to severe bandwidth, latency, and operational constraints. How to efficiently extract valuable information from massive event streams and transmit it with extremely low bandwidth under conditions of extremely limited onboard resources has become a core bottleneck restricting the space application of event cameras.
[0005] To address the above problems, existing technologies mainly explore the following directions:
[0006] Firstly, there are traditional spaceborne data compression schemes for frame-based sensors. These schemes primarily target image frame data output from CCD / CMOS image sensors, employing spatial correlation-based compression algorithms such as JPEG and JPEG-LS, or combining them with interspectral compression methods like PCA to compress multispectral images on-board. However, these methods are essentially still image frame compression processing, and their processing objects and data paradigms are completely incompatible with the asynchronous and sparse characteristics of event streams, making them unsuitable for direct application to event camera data.
[0007] Secondly, there are methods for denoising and refining event streams. Existing research has proposed denoising methods for event camera data streams based on data density and centroid, as well as denoising time-plane representation methods based on spatiotemporal neighborhood correlation. These methods focus on filtering out noisy events in the event stream, but they fail to address the problem of further extracting high-level semantic features (such as target motion state) from the denoised event stream, and they do not address the issue of a simplified data transmission scheme for spaceborne downlink bandwidth constraints.
[0008] Thirdly, target detection and tracking methods based on event cameras. Existing research has proposed spatial point target tracking methods based on asynchronous event streams, spatial non-cooperative target feature tracking methods fused with biomimetic dynamic vision, and spatial target detection and tracking methods based on event cameras. Although these methods achieve the detection and tracking of specific targets, their processing flow usually relies on ground-based terminals or platforms with strong computing power, and the output results are still intermediate data containing a large amount of pixel coordinate information, failing to fundamentally solve the problems of limited spaceborne computing resources and downlink bandwidth bottlenecks. For weak spatial targets with low signal-to-clutter ratios and sparse and unstable event triggering, existing event-based methods are more prone to high false alarm rates and missed detections. Summary of the Invention
[0009] To address this, embodiments of the present invention provide a high signal-to-noise ratio downsampling method and system for spaceborne photoelectric event streams, in order to solve the technical problem of how to extract target motion state information from massive asynchronous event streams with high signal-to-noise ratio and achieve simplified transmission with extremely low bandwidth under conditions of extremely limited spaceborne computing power and downlink telemetry bandwidth.
[0010] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:
[0011] According to a first aspect of the present invention, a high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams is provided, the method comprising:
[0012] Receives a wide-field-of-view raw asynchronous event stream generated by a spatial sensor;
[0013] Perform initial spatiotemporal correlation screening at the hardware bottom layer or edge node to eliminate isolated background noise events and only cluster local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters;
[0014] The suspected target cluster is input into a dynamic local gate, and a multi-model state observer is introduced to extract the kinematic abstract features of the target. The kinematic abstract features include at least the target angular position and multiple kinematic derivatives.
[0015] Discard the original physical pixel coordinates within the local gate, and generate and send only a simplified structure containing the target angular position, multi-order motion derivatives, and covariance confidence.
[0016] Furthermore, the initial screening of spatiotemporal correlation is a generalization to find the clustering characteristics of discrete events in a multidimensional space jointly formed by the time and space dimensions. The subsequent calculation mechanism is only triggered when the signal has a clear physical motion correlation.
[0017] Furthermore, the multi-model state observer includes an extended Kalman filter and / or an interactive multi-model filter to strip away massive coordinate jumps and roll noise and extract stable and realistic kinematic features.
[0018] Furthermore, the size of the dynamic local gate is adaptively adjusted according to the target's motion state. When the target is in a high dynamic motion state, the gate range is expanded, and when the target is in a low dynamic motion state, the gate range is reduced.
[0019] Furthermore, the multi-order kinematic derivatives include velocity and acceleration terms, and the covariance confidence is used as a measure of uncertainty to characterize the abstract kinematic features.
[0020] Furthermore, the simplified structure transforms the original event stream data, which originally required broadband downlink communication, into low-bandwidth telemetry data packets at the KBs level.
[0021] Furthermore, the initial screening of spatiotemporal correlation is performed at the hardware level. Based on the timestamp and pixel address of the event, the spatial neighborhood and temporal neighborhood are jointly judged, and the noise event is removed in real time by hardware parallel computing.
[0022] Furthermore, after generating the simplified structure, the method further includes: packaging the simplified structure according to a preset communication data packet encapsulation format, wherein the encapsulation format is required to include a position field, a multi-order motion derivative field, and a covariance confidence matrix field.
[0023] Furthermore, the spatial sensor is an asynchronous visual sensor, and each event in the original asynchronous event stream contains at least pixel coordinates, timestamps, and polarity change information.
[0024] According to a second aspect of the present invention, a high signal-to-noise ratio downsampling system for spaceborne photoelectric event streams is provided, the system comprising:
[0025] The event stream receiving module is used to receive the wide-field-of-view raw asynchronous event stream generated by the space sensor;
[0026] The spatiotemporal preliminary screening module is deployed at the hardware bottom layer or edge node to perform spatiotemporal correlation preliminary screening, eliminate isolated background noise events, and only aggregate local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters.
[0027] The kinematic abstraction module is used to input the suspected target cluster into a dynamic local gate and introduce a multi-model state observer to extract the kinematic abstract features of the target. The kinematic abstract features include at least the target angular position and multiple kinematic derivatives.
[0028] The data simplification and delivery module is used to discard the original physical pixel coordinates within the local gate and generate and deliver only a simplified structure containing the target angular position, multi-order motion derivatives, and covariance confidence.
[0029] The embodiments of the present invention have the following advantages:
[0030] This invention receives a wide-field-of-view raw asynchronous event stream generated by a space sensor; performs initial spatiotemporal correlation screening at the hardware layer or edge nodes to eliminate isolated background noise events, and aggregates local events conforming to a preset spatiotemporal distribution pattern into suspected target clusters; inputs the suspected target clusters into a dynamic local gate, introduces a multi-model state observer to extract the kinematic abstract features of the targets; discards the original physical pixel coordinates within the local gate, and generates and sends only a simplified structure containing the target angular position, multi-order kinematic derivatives, and covariance confidence. This invention, by abstracting massive discrete photon coordinates into low-dimensional high-order kinematic vectors, achieves high signal-to-noise ratio data downsampling and extremely low bandwidth simplified transmission under conditions of extremely limited spaceborne computing power and downlink telemetry bandwidth, effectively solving the bottleneck problem of satellite-to-ground data transmission. Attached Figure Description
[0031] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.
[0032] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.
[0033] Figure 1 This is a schematic diagram of the logic structure of a high signal-to-noise ratio downsampling system for spaceborne photoelectric event streams provided in an embodiment of the present invention;
[0034] Figure 2 This is a flowchart illustrating a high signal-to-noise ratio downsampling method for spaceborne optoelectronic event streams provided in an embodiment of the present invention. Detailed Implementation
[0035] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.
[0036] refer to Figure 1 This invention discloses a high signal-to-noise ratio downsampling system for spaceborne photoelectric event streams. The system includes: an event stream receiving module 1; a spatiotemporal initial screening module 2; a kinematic abstraction module 3; and a data simplification and distribution module 4.
[0037] Event stream receiving module 1 is used to receive the wide field-of-view raw asynchronous event stream generated by the spatial sensor (i.e., asynchronous vision sensor) and complete data acquisition and buffering through a high-speed serial interface.
[0038] The spatiotemporal preliminary screening module 2 is deployed at the hardware bottom layer (i.e., the FPGA logic layer) to perform the above-mentioned spatiotemporal correlation preliminary screening. It uses hardware parallel computing to eliminate isolated background noise events and only gathers local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters.
[0039] The kinematic abstraction module 3 is used to input the suspected target cluster into the dynamic local gate and introduce a multi-model state observer (an interactive multi-model filter and extended Kalman filter fusion architecture) to extract the kinematic abstract features of the target. The kinematic abstract features include at least the target's angular position, velocity, and acceleration.
[0040] The data simplification and distribution module 4 is used to discard the original physical pixel coordinates of all events within the local gate, generate and distribute only a simplified structure containing the target angular position, multi-order motion derivatives and covariance confidence, and complete satellite-to-ground downlink transmission with extremely low bandwidth at the KBs level.
[0041] The aforementioned modules can be integrated into the same onboard signal processing hardware platform (such as an FPGA+DSP heterogeneous computing architecture). The spatiotemporal initial screening module runs on the programmable logic section of the FPGA to achieve hardware-level parallel acceleration, while the kinematic abstraction module runs on the soft-core processor of the DSP or FPGA to perform complex floating-point matrix operations.
[0042] Corresponding to the high signal-to-noise ratio (SNR) downsampling system for spaceborne photoelectric event streams disclosed above, this invention also discloses a high SNR downsampling method for spaceborne photoelectric event streams. The following details a high SNR downsampling method for spaceborne photoelectric event streams disclosed in this invention, in conjunction with the high SNR downsampling system for spaceborne photoelectric event streams described above.
[0043] refer to Figure 2This invention discloses a high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams, which can be deployed on a spaceborne digital signal processor (DSP) or field-programmable gate array (FPGA) platform. The method mainly includes four core steps: receiving and buffering the original asynchronous event stream, initial screening of spatiotemporal correlations at the hardware level, kinematic feature extraction based on dynamic local gates and multi-model state observers, and generation and distribution of simplified structures.
[0044] I. Receiving the original asynchronous event stream
[0045] In this embodiment, the spatial sensor employs an asynchronous vision sensor (also known as a dynamic vision sensor, DVS), which responds independently to changes in light intensity at the pixel level. When the logarithmic change in light intensity received by a pixel within the sensor's field of view exceeds a preset threshold, that pixel outputs an event. Each event contains at least the following information: the pixel's two-dimensional coordinates (x, y) in the sensor array, the timestamp t of the event (with an accuracy down to the microsecond level), and the polarity p of the light intensity change (indicating brightening or darkening). The sensor continuously outputs the aforementioned asynchronous event stream, with an event output rate reaching millions per second.
[0046] The onboard data acquisition front end receives the aforementioned raw asynchronous event stream via a high-speed serial interface (such as MIPI CSI-2 or LVDS) and buffers it in the FPGA's on-chip Block RAM or external SRAM, forming an event sequence buffer to be processed. The buffer adopts a circular queue structure to ensure that it can stably provide data for subsequent processing stages under continuous high-speed input event stream conditions.
[0047] II. Initial Screening for Spatiotemporal Correlation
[0048] The initial screening of spatiotemporal correlations is performed at the hardware level (i.e., the FPGA logic layer). The core purpose of this step is to eliminate isolated background noise events that do not have a clear physical motion correlation, and only gather local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters.
[0049] Specifically, this embodiment does not rely on a single specific clustering algorithm (such as K-means or DBSCAN), but instead employs a generalized multidimensional spatial clustering characteristic discrimination mechanism to search for the clustering characteristics of discrete events in a four-dimensional space (x, y, t, p) jointly formed by the time and spatial dimensions. Its implementation is as follows:
[0050] First, for each newly arriving event e_i=(x_i, y_i, t_i, p_i) in the buffer, the FPGA logic defines a neighborhood with a radius of R_s (in pixels) in the spatial domain and a window width of T_w (in microseconds) in the temporal domain, centered on this event. Then, the number of events N_neighbor falling within this spatiotemporal neighborhood is counted.
[0051] If N_neighbor is less than the preset spatial density threshold N_th, the event is determined to be an isolated background noise event and is directly eliminated without triggering any subsequent calculations. If N_neighbor is greater than or equal to N_th, the event is determined to be a signal event with a clear physical motion correlation and is included in the candidate set of suspected target clusters.
[0052] The aforementioned spatiotemporal neighborhood discrimination logic is implemented using hardware parallel computing—multiple computing units in the FPGA simultaneously process event streams from different spatial regions, thereby achieving real-time removal of noise events. It should be noted that the specific values of R_s, T_w, and N_th can be adaptively adjusted based on target characteristics (such as target angular velocity, event trigger rate, etc.) and sensor parameters; this embodiment does not impose a unique limitation on this.
[0053] After the initial screening based on the spatiotemporal correlations described above, a large number of isolated noise events have been eliminated. The remaining events form several local clusters in the spatiotemporal dimension, with each cluster corresponding to a potential suspected target.
[0054] III. Dynamic Local Gates and Multi-Model State Observers
[0055] The cluster of suspected targets formed after initial screening based on spatiotemporal correlation is input into the Dynamic Local Gate, and a multi-model state observer is introduced to extract the kinematic abstract features of the targets.
[0056] The construction method of dynamic local gate is as follows: For each suspected target cluster, firstly, the centroid position (i.e., the weighted average of spatial coordinates) and average timestamp of all events within the cluster are calculated as the observation position of the target at the current moment. Then, a local gate range is defined with this observation position as the center. The size of this gate is not fixed, but is adaptively adjusted according to the historical motion state of the target—when the target is in a high-dynamic motion state (such as rapid maneuvering or large-angle turning), the gate range is appropriately expanded to ensure that the target is not lost; when the target is in a low-dynamic motion state (such as uniform linear motion), the gate range is appropriately reduced to suppress false alarms and background interference.
[0057] After the dynamic local gate is defined, a multi-model state observer is introduced to estimate the state of the target within the gate. In this embodiment, the multi-model state observer adopts an architecture that combines an interactive multiple model (IMM) filter with an extended Kalman filter (EKF).
[0058] Specifically, the interactive multi-model filter (IMF) runs multiple parallel motion models simultaneously, including but not limited to: constant velocity (CV) models, constant acceleration (CA) models, and coordinated turn (CT) models. Each model predicts the target's state at the next moment and calculates its likelihood probability based on the current observations. The IMF uses a Markov chain to softly switch between models, adaptively adjusting model weights according to the real-time changes in the likelihood probabilities of each model, thereby achieving high-precision tracking of the motion state of maneuvering targets.
[0059] Within each model, an extended Kalman filter is used to handle the nonlinearity of the observation equations. Since the observation information provided by the event camera is the pixel coordinates of the target on the image plane (converted to angular position after distortion correction), while the target's state vector contains kinematic parameters such as position, velocity, and acceleration, the observation equations exhibit nonlinear characteristics. The extended Kalman filter approximates the nonlinear filtering problem as a linear filtering problem by performing a first-order Taylor expansion (i.e., linearization) on the nonlinear observation equations, thereby recursively estimating the target's state vector and its covariance matrix.
[0060] The final output of the interactive multi-model filter is a probabilistically weighted fusion of the estimation results from multiple models. Its output target kinematic abstract features include at least: target angular position (i.e., the azimuth and pitch angles of the target in the sensor's field of view), multi-order kinematic derivatives (including velocity and acceleration terms), and a covariance confidence matrix (a measure of uncertainty characterizing the aforementioned kinematic abstract features). Through the processing of the multi-model state observer, coordinate jumps and roll noise interspersed in the massive event stream are effectively removed, extracting stable and realistic kinematic features.
[0061] IV. Discarding Original Pixel Coordinates and Generating a Simplified Structure
[0062] After completing the above kinematic feature extraction, this embodiment performs a key data simplification operation: discarding the original physical pixel coordinates of all events within the dynamic local gate, that is, no longer retaining any original event point data belonging to the target.
[0063] Instead, the system generates a streamlined structure that contains only the following core information fields:
[0064] (1) Target ID: A unique identifier used to identify the currently tracked target;
[0065] (2) Target Angular Position: includes the azimuth angle θ and elevation angle φ of the target in the sensor's field of view;
[0066] (3) Kinematic Derivatives: These include at least the velocity vector v (including the angular velocity component) and the acceleration vector a (including the angular acceleration component).
[0067] (4) Covariance Confidence Matrix: used to characterize the uncertainty of the above position and motion derivative estimates, that is, the variance of each state estimate and their covariance;
[0068] (5) Timestamp: The global time stamp corresponding to this structure.
[0069] The aforementioned simplified structure is encapsulated in a compact binary format, with each structure containing only a few hundred bytes of data. Compared to the tens of thousands of original data points representing photons in the original event stream, this simplified structure compresses massive amounts of data into a few mathematical vectors representing the motion state of objects without losing spatial dynamics accuracy, achieving data substitution at the source level.
[0070] V. Distribution of the Simplified Structure
[0071] After generating the simplified structure, the spaceborne system packages it according to a preset communication data packet encapsulation format. This encapsulation format mandates the inclusion of a position field, multi-order motion derivative fields, and a covariance confidence matrix field. The packaged data packet is then transmitted downlink to the ground station via the spaceborne telemetry transmitter.
[0072] Because the data size of the simplified structure is extremely small (the overall data stream is at the KB level), the original event stream data, which originally required broadband downlink communication, is transformed into telemetry data packets with extremely low bandwidth. Taking a satellite continuously tracking multiple space targets as an example, if there are N targets, and each target outputs one simplified structure per second (the output frequency can be adjusted according to the target's dynamic characteristics), the downlink data rate is only N × several hundred bytes per second, which is far less than the data volume corresponding to millions of events per second in the original event stream, fundamentally breaking through the transmission bottleneck of satellite-to-ground data interaction.
[0073] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.
Claims
1. A high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams, characterized in that, The method includes: Receives a wide-field-of-view raw asynchronous event stream generated by a spatial sensor; Perform initial spatiotemporal correlation screening at the hardware bottom layer or edge node to eliminate isolated background noise events and only cluster local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters; The suspected target cluster is input into a dynamic local gate, and a multi-model state observer is introduced to extract the kinematic abstract features of the target. The kinematic abstract features include at least the target angular position and multiple kinematic derivatives. Discard the original physical pixel coordinates within the local gate, and generate and send only a simplified structure containing the target angular position, multi-order motion derivatives, and covariance confidence.
2. The high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The initial screening of spatiotemporal correlation is a generalization to find the clustering characteristics of discrete events in a multidimensional space composed of the time and space dimensions. The subsequent calculation mechanism is triggered only when the signal has a clear physical motion correlation.
3. The high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The multi-model state observer includes an extended Kalman filter and / or an interactive multi-model filter to strip away massive coordinate jumps and roll noise and extract stable and realistic kinematic features.
4. The high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The size of the dynamic local gate is adaptively adjusted according to the target's motion state. When the target is in a high dynamic motion state, the gate range is expanded, and when the target is in a low dynamic motion state, the gate range is reduced.
5. A high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The multi-order kinematic derivatives include velocity and acceleration terms, and the covariance confidence is used as a measure of uncertainty to characterize the abstract kinematic features.
6. The high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The simplified structure transforms the original event stream data, which previously required broadband downlink communication, into low-bandwidth telemetry data packets at the KBs level.
7. A high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The initial screening of spatiotemporal correlation is performed at the hardware level. It performs joint discrimination of spatial and temporal neighborhoods based on the timestamp and pixel address of the event, and completes the real-time removal of noisy events through hardware parallel computing.
8. A high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, After generating the simplified structure, the method further includes: packaging the simplified structure according to a preset communication data packet encapsulation format, wherein the encapsulation format is required to include a position field, a multi-order motion derivative field, and a covariance confidence matrix field.
9. A high signal-to-noise ratio downsampling method for spaceborne photoelectric event streams as described in claim 1, characterized in that, The spatial sensor is an asynchronous vision sensor, and each event in the original asynchronous event stream contains at least pixel coordinates, timestamps, and polarity change information.
10. A high signal-to-noise ratio downsampling system for spaceborne photoelectric event streams, characterized in that, The system includes: The event stream receiving module is used to receive the wide-field-of-view raw asynchronous event stream generated by the space sensor; The spatiotemporal preliminary screening module is deployed at the hardware bottom layer or edge node to perform spatiotemporal correlation preliminary screening, eliminate isolated background noise events, and only aggregate local events that conform to the preset spatiotemporal distribution pattern into suspected target clusters. The kinematic abstraction module is used to input the suspected target cluster into a dynamic local gate and introduce a multi-model state observer to extract the kinematic abstract features of the target. The kinematic abstract features include at least the target angular position and multiple kinematic derivatives. The data simplification and delivery module is used to discard the original physical pixel coordinates within the local gate and generate and deliver only a simplified structure containing the target angular position, multi-order motion derivatives, and covariance confidence.