Multi-source data real-time preprocessing method and system for unmanned aerial vehicle platform

Abstract

The invention discloses an unmanned aerial vehicle platform-oriented multi-source data real-time preprocessing method and system. The method comprises the following steps of: acquiring multi-sensor original data streams and clock synchronization signals, identifying a sampling rate deviation component, generating a delay compensation coefficient and establishing a data alignment field domain; scanning a data alignment field domain to form a buffer capacity distribution surface, and configuring an overflow early warning trigger point to form a grading preprocessing pipeline scheme; acquiring a sensor noise sequence to generate an effective signal bandwidth, and determining a processing time window; executing data screening to perform aggregation path planning, and identifying a data aggregation position to form a flow control anchor; executing time slice analysis to form a queue backlog rate, and fusing a clock synchronization signal to execute phase adjustment to generate a data stream speed; and comparing the co-processing parameter with a preset delay threshold, and adjusting the flow control anchor to form a self-adaptive preprocessing sequence. According to the invention, accurate time sequence alignment and efficient flow control of multi-sensor data are realized, and the real-time performance and stability of an unmanned aerial vehicle data processing system are improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) data processing technology, and in particular to a method and system for real-time preprocessing of multi-source data for UAV platforms. Background Technology

[0002] Modern drones typically carry multiple types of sensors, including inertial measurement units (IMUs), GPS positioning modules, visual cameras, and lidar. The massive amounts of multimodal data generated by these sensors provide fundamental information support for the drone's autonomous navigation and intelligent decision-making. However, due to significant differences in sampling frequencies, data formats, and transmission delays among different sensors, multi-sensor data faces severe challenges in timing alignment and synchronous processing.

[0003] Traditional multi-sensor data processing methods primarily rely on simple timestamp alignment or fixed buffering mechanisms for data synchronization. This static processing strategy cannot adapt to dynamically changing flight environments and load demands. Existing technologies lack systematic solutions for handling sampling rate deviations, transmission delay compensation, and flow control, often leading to increased data processing latency, low resource utilization, and insufficient system responsiveness. Particularly in highly dynamic flight scenarios, sudden changes in sensor data streams can easily cause processing bottlenecks and data loss, severely impacting the real-time performance and reliability of UAV flight control. Therefore, a method is urgently needed to address at least one of the aforementioned problems. Summary of the Invention

[0004] This invention provides a method and system for real-time preprocessing of multi-source data for unmanned aerial vehicle (UAV) platforms. By establishing a data alignment field and a hierarchical preprocessing pipeline, it achieves efficient synchronous processing of multi-sensor data. The method establishes a data alignment field by identifying sampling rate deviations and generating delay compensation coefficients, constructs a flow control scheduling strategy to form a hierarchical preprocessing pipeline scheme, establishes a processing time window based on real-time constraints, forms flow control anchor points through aggregation path planning, and finally generates an adaptive preprocessing sequence using queue backlog analysis. This addresses the technical problems of inconsistent timing of multi-sensor data, lack of flow control, and low processing efficiency in existing technologies.

[0005] The first aspect of this invention proposes a real-time preprocessing method for multi-source data for unmanned aerial vehicle (UAV) platforms, comprising the following steps: The system collects raw data streams and clock synchronization signals from multiple sensors of the UAV, identifies sampling rate deviation components based on the raw data streams, generates delay compensation coefficients using the clock synchronization signals, and establishes a data alignment field based on the sampling rate deviation components and the delay compensation coefficients. The data alignment field is scanned to form a buffer capacity distribution surface. Overflow warning trigger points are configured along the buffer capacity distribution surface. The flow control scheduling strategy is deduced in reverse through the warning trigger points. A hierarchical preprocessing pipeline scheme is formed based on the flow control scheduling strategy. Obtain the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and perform pruning processing on the effective signal bandwidth based on the real-time constraints to establish a processing time window; The hierarchical preprocessing pipeline scheme is used to perform data filtering within the processing time window to obtain feature distribution changes, and aggregation path planning is performed according to the feature distribution changes to identify data aggregation locations. Throughput assessment is performed on the data aggregation locations to form flow control anchors. Time slice analysis is performed on the flow control anchor to form a queue backlog rate. The queue backlog rate is then fused with the clock synchronization signal to perform phase adjustment and generate a data flow rate. Load balancing is then performed according to the data flow rate to generate collaborative processing parameters. The collaborative processing parameters are compared with a preset delay threshold to generate a scheduling deviation signal, and the flow control anchor is adjusted based on the scheduling deviation signal to form an adaptive preprocessing sequence.

[0006] A second aspect of this invention proposes a real-time preprocessing system for multi-source data of unmanned aerial vehicle (UAV) platforms, comprising: The data acquisition module is used to acquire raw data streams and clock synchronization signals from multiple sensors of the UAV, identify sampling rate deviation components based on the raw data streams, generate delay compensation coefficients using the clock synchronization signals, and establish a data alignment field based on the sampling rate deviation components and the delay compensation coefficients. The buffer analysis module is used to scan the data alignment field to form a buffer capacity distribution surface, configure overflow warning trigger points along the buffer capacity distribution surface, deduce the flow control scheduling strategy in reverse through the warning trigger points, and form a hierarchical preprocessing pipeline scheme based on the flow control scheduling strategy. The bandwidth processing module is used to acquire the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and perform clipping processing on the effective signal bandwidth based on the real-time constraints to establish a processing time window; The flow analysis module is used to perform data filtering within the processing time window using the hierarchical preprocessing pipeline scheme to obtain feature distribution changes, perform aggregation path planning according to the feature distribution changes to identify data aggregation locations, and perform throughput evaluation on the data aggregation locations to form flow control anchors. The parameter extraction module is used to perform time slice analysis on the flow control anchor to form a queue backlog rate, perform phase adjustment by fusing the clock synchronization signal with the queue backlog rate to generate a data flow rate, and perform load balancing to generate collaborative processing parameters according to the data flow rate. The control output module is used to compare the collaborative processing parameters with a preset delay threshold to generate a scheduling deviation signal, and adjust the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence.

[0007] The beneficial effects of this invention are reflected in the following aspects: First, by constructing a data alignment field and sampling rate deviation compensation mechanism, combined with a hierarchical preprocessing pipeline design, the core problem of multi-sensor timing inconsistency is solved. This technology can achieve high-precision time synchronization and dynamically compensate for delay drift during system operation, significantly improving time synchronization accuracy compared to traditional fixed compensation methods. Through timing distortion spectrum analysis and delay correlation matrix construction, the system can accurately identify and correct various timing deviations, providing a high-quality synchronization data foundation and effectively avoiding the accumulation of fusion errors caused by timing misalignment. Second, by employing effective signal bandwidth processing and feature distribution change analysis, combined with aggregation path planning and flow control anchor point mechanisms, an intelligent data flow adjustment system is established. Through dynamic identification of data aggregation locations and real-time throughput assessment, the system can predictively allocate flow and balance load, achieving optimal allocation of system resources while ensuring priority processing of critical flight data. This mechanism, through buffer capacity distribution surface scanning and overflow warning triggering, can proactively prevent the generation of data processing bottlenecks, improving the overall efficiency of data processing and system throughput. Finally, by employing queue backlog rate analysis and phase adjustment techniques, an adaptive preprocessing sequence based on backpressure injection was constructed, enabling the system to self-regulate and optimize. This mechanism, through real-time monitoring of system load status and performance indicators, can quickly respond and automatically adjust processing strategies when system performance fluctuates. Through transient flow rate analysis and data trigger point identification, the system can predict load change trends in advance and take corresponding preventative measures, ensuring the continuity and robustness of data processing for UAVs in harsh flight environments and under unforeseen circumstances, thus providing technical assurance for flight safety.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a real-time preprocessing method for multi-source data on an unmanned aerial vehicle (UAV) platform according to the present invention.

[0010] Figure 2 This is a structural block diagram of a multi-source data real-time preprocessing system for unmanned aerial vehicle (UAV) platforms according to the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0012] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0013] It should also be noted that when a component is described as "fixed to" or "set on" another component, it can be directly on the other component or there may be an intervening component present. When a component is described as "connected to" another component, it can be directly connected to the other component or there may be an intervening component present.

[0014] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0015] The technical solutions of the embodiments of this application will be described below.

[0016] like Figure 1 As shown, this embodiment of the invention provides a real-time preprocessing method for multi-source data for unmanned aerial vehicle (UAV) platforms, including the following steps S110-S160: Step S110: Collect raw data streams and clock synchronization signals from multiple sensors of the UAV, identify sampling rate deviation components based on the raw data streams, generate delay compensation coefficients using the clock synchronization signals, and establish a data alignment field based on the sampling rate deviation components and delay compensation coefficients.

[0017] Specifically, the process involves acquiring raw data streams and clock synchronization signals from multiple sensors on the UAV. The UAV's onboard multi-sensor system synchronously acquires multi-dimensional raw data during flight. These sensors include an IMU (Inertial Measurement Unit), GPS positioning module, visual camera, LiDAR, and barometric altimeter. The raw data stream consists of unprocessed measurement data output by each sensor at its respective sampling frequency. When the UAV flies in complex terrain, GPS signals may be lost due to obstruction, IMU data remains continuous but may drift, and visual data fluctuates in quality due to lighting conditions. These temporal inconsistencies in the raw data need to be addressed through data alignment. Each sensor outputs a raw data stream according to its own sampling frequency; different sensors have different sampling characteristics and data formats. The raw data stream includes sensor measurement values, timestamps, and sensor status indicators. Simultaneously, the flight control system's master clock synchronization signal is acquired, providing a unified time reference. This clock synchronization signal is distributed to each sensor module via a hardware clock distributor, ensuring that all sensors maintain time-base synchronization. A caching mechanism for the multi-sensor data stream is established to ensure the integrity and continuity of data acquisition.

[0018] Identifying sampling rate deviation components based on raw data streams. Using the acquired raw data streams from multiple sensors, the deviation between the actual and nominal sampling rates of each sensor is analyzed. The sampling rate deviation component reflects the difference between the actual and ideal operating states of the sensors. When a UAV flies for extended periods in high-temperature environments, the clocks of some sensors may deviate from the standard frequency due to temperature drift, leading to a systematic deviation in the sampling rate. The actual sampling number is calculated for each sensor per unit time, and the actual sampling rate is obtained by statistically analyzing the number of data packets within a continuous time window. The sampling rate deviation is calculated using the formula Δf = f_actual - f_nominal, where Δf is the sampling rate deviation, f_actual is the actual sampling rate, and f_nominal is the nominal sampling rate. The temporal variation characteristics of the sampling rate deviation are analyzed, and the dynamic changes in the deviation are tracked using a sliding window technique. Periodic patterns and random fluctuations in the sampling rate deviation are identified. Periodic components are extracted through spectral analysis, and random components are obtained through residual analysis. The statistical characteristics of the sampling rate deviation components for each sensor are calculated, including the mean, standard deviation, maximum deviation, and rate of change of deviation. Deviations are categorized into three types: steady-state deviation, transient deviation, and abnormal deviation for classification and management. The sampling rate deviation components of each sensor are determined to provide basic data for subsequent timing correction.

[0019] Delay compensation coefficients are generated using clock synchronization signals. Based on the acquired clock synchronization signals, delay compensation parameters are calculated for each sensor's data transmission and processing. These delay compensation coefficients correct for inherent delays in the data transmission paths of each sensor. When LiDAR data requires complex point cloud processing algorithms while visual data only requires simple format conversion, the difference in processing delays needs to be balanced by compensation coefficients to ensure accurate alignment of the data on the time axis. The propagation delay of the clock synchronization signal from the master clock to each sensor is measured, including hardware transmission delay and circuit processing delay. The inherent delays of each sensor's data processing link are analyzed, including analog-to-digital conversion, digital filtering, data encapsulation, and transmission interfaces. The delay compensation coefficient is calculated using the formula K_comp = (T_sync + T_process) / T_period, where T_sync is the clock synchronization delay, T_process is the data processing delay, and T_period is the sampling period. Compensation parameters are dynamically adjusted based on system operating status monitoring data, using lookup tables and interpolation calculations for rapid parameter adjustment.

[0020] In some embodiments, establishing a data alignment field based on the sampling rate deviation component and the delay compensation coefficient includes: constructing a temporal distortion map based on the sampling rate deviation component; performing inter-frame difference analysis on the temporal distortion map to identify the synchronization degradation rate; constructing a delay correlation matrix based on the synchronization degradation rate and the delay compensation coefficient; and establishing a data alignment field through the delay correlation matrix.

[0021] A temporal distortion map is constructed based on the sampling rate deviation component. Using the identified sampling rate deviation component, a visual map reflecting the degree of time series distortion is constructed. The temporal distortion map displays the synchronization deviation of different sensors on the time axis. When a UAV performs precise mapping tasks, the high-frequency data from the IMU and the low-frequency data from GPS need to be precisely synchronized. The temporal distortion map can intuitively show which time periods have synchronization problems, facilitating timely adjustments. The sampling rate deviation components of each sensor are arranged in chronological order to form a deviation time series. The cumulative effect of the deviation series is calculated; the cumulative deviation reflects the gradual amplification process of temporal distortion. The cumulative distortion is mapped onto a two-dimensional map, with the horizontal axis representing the time process, the vertical axis representing different sensors, and the color depth representing the degree of distortion. Areas with high distortion are displayed in dark colors, indicating severe time synchronization problems. The spatial distribution pattern of the distortion map is analyzed to identify time periods and sensor combinations with concentrated distortion. The gradient distribution of the distortion map is calculated; areas with large gradients correspond to locations of drastic distortion changes. The temporal distortion map provides a visual representation of the sampling rate deviation component, offering an intuitive tool for synchronization performance evaluation.

[0022] Inter-frame difference analysis is performed on the temporal distortion map to identify the synchronization degradation rate. Based on the constructed temporal distortion map, the distortion changes between consecutive time frames are analyzed. The temporal distortion map is divided into consecutive time frames at fixed time intervals to facilitate quantitative analysis. The distortion difference between adjacent time frames is calculated, reflecting the trend of synchronization performance changes. The inter-frame difference is obtained by subtracting the average distortion value of the current frame from the average distortion value of the previous frame. The statistical distribution characteristics of the inter-frame difference are analyzed, and the distribution pattern of the difference is statistically analyzed using histograms. Outliers in the inter-frame difference are identified, corresponding to sudden changes in synchronization performance. The synchronization degradation rate is calculated using the formula R_degrade=ΔD_frame / D_frame(t), where R_degrade is the synchronization degradation rate, ΔD_frame is the inter-frame distortion difference, and D_frame(t) is the average distortion value of the current frame. A positive degradation rate indicates that synchronization performance is deteriorating, while a negative degradation rate indicates that synchronization performance is improving. The temporal evolution trend of the synchronization degradation rate is calculated, and the trend analysis predicts the development direction of synchronization performance. The degree of degradation is divided into three levels: mild, moderate, and severe, providing a basis for graded treatment of system regulation.

[0023] A time-delay correlation matrix is ​​constructed based on the synchronization degradation rate and delay compensation coefficient. Using the identified synchronization degradation rate and the generated delay compensation coefficient, a matrix representation of the time-delay correlation between sensors is established. The synchronization degradation rate and delay compensation coefficient of each sensor are organized into vector form; the degradation rate vector reflects the change in synchronization performance, and the compensation coefficient vector reflects the delay correction capability. The correlation between the degradation rate and the compensation coefficient is calculated; a high correlation indicates a strong association. The elements of the time-delay correlation matrix are obtained through a comprehensive calculation of the degradation rate, compensation coefficient, and inter-sensor correlation coefficient. The inter-sensor correlation coefficient is obtained through correlation analysis of the data time series. Symmetrical elements of the matrix are averaged, and sparse elements are set to zero to improve computational efficiency. The basic characteristics of the time-delay correlation matrix are analyzed to ensure its numerical stability and information integrity.

[0024] A data alignment domain is established using a time-delay correlation matrix. Based on the constructed time-delay correlation matrix, a unified alignment domain for multi-sensor data is generated. The data alignment domain is a virtual time correction space. When time deviations exist in the multi-sensor data of the UAV, the alignment domain provides a unified time reference framework, enabling all sensor data to be fused under the same time reference. The main time-delay correlation patterns are extracted using matrix decomposition, with each pattern corresponding to the most important correlation feature. An alignment algorithm is designed based on the main correlation patterns, prioritizing sensor combinations with strong correlations. The coordinates of the alignment domain are calculated using a weighted combination of the reference time and correlation features, with the weighting coefficients determined by the corresponding components of the correlation matrix. A grid representation of the alignment domain is established, with grid nodes corresponding to the time alignment status of different sensors. The gradient distribution of the domain is calculated, with the gradient direction indicating the optimal time adjustment direction. The domain parameters are updated in real time based on new time-delay correlation information, maintaining the dynamic adaptability of the alignment domain. The alignment domain provides a unified time reference framework for multi-sensor data, achieving precise time alignment and eliminating the impact of sampling rate deviations and transmission delays on the accuracy of data fusion.

[0025] Step S120: Scan the data alignment field to form a buffer capacity distribution surface, configure overflow warning trigger points along the buffer capacity distribution surface, deduce the flow control scheduling strategy in reverse through the warning trigger points, and form a hierarchical preprocessing pipeline scheme based on the flow control scheduling strategy.

[0026] Specifically, a buffer capacity distribution surface is formed by scanning the data alignment field. Using the data alignment field, the buffer capacity characteristics of each region are analyzed through scanning techniques. The buffer capacity distribution surface reflects the temporal and spatial distribution of buffer demand for data streams from different sensors. When UAVs perform complex reconnaissance missions, the large amount of image data generated by high-resolution cameras and the small amount of high-frequency data from IMUs exhibit completely different buffer occupancy patterns. The distribution surface can intuitively show which areas require more buffer resources. The data alignment field is divided into regular scanning grids to facilitate systematic analysis of the characteristics of each region. Within each grid cell, statistical flow density and buffer occupancy are calculated. Flow density reflects the amount of data passing through the region per unit time. The buffer capacity utilization rate of each grid cell is calculated; the utilization rate equals the ratio of actual occupied capacity to total capacity. A continuous buffer capacity distribution surface is generated through capacity interpolation calculations between adjacent grid cells. Peak and valley regions on the distribution surface are identified; peak regions correspond to high buffer demand, and valley regions correspond to low buffer demand. The temporal variation pattern of the buffer capacity distribution is analyzed, and the periodicity and suddenness of capacity demand are identified through time-series analysis.

[0027] Overflow warning trigger points are configured along the buffer capacity distribution surface. Based on the generated buffer capacity distribution surface, warning trigger mechanisms are set up at locations where the capacity is close to saturation. Overflow warning trigger points are monitoring points that automatically alarm when the buffer usage approaches a dangerous threshold. When drones fly in dense urban areas, multiple sensors work simultaneously, generating massive amounts of data. Warning trigger points can issue timely warnings before the buffer overflows, preventing data loss. Potential overflow risk areas are identified using a threshold determination method, and an alarm is triggered when the buffer capacity utilization exceeds a set threshold. The overflow risk assessment formula is R_overflow = C_used / C_total, where R_overflow is the overflow risk coefficient, C_used is the used buffer capacity, and C_total is the total buffer capacity. When the risk coefficient exceeds the set threshold, an alarm of the corresponding level is set, with different levels corresponding to different response measures. The configuration density of warning trigger points is determined based on the degree of change in the distribution surface, with increased trigger point density in areas of drastic change. The spatial distribution optimization of warning trigger points is calculated, and the integrity of the warning network is ensured through coverage analysis. Each warning trigger point includes attribute information such as trigger threshold, response time, impact range, and processing priority. The early warning trigger point is connected to the flow control system through an interrupt mechanism to ensure timely response to buffer overflow risks.

[0028] In some embodiments, the step of reversely deriving the flow control scheduling strategy via the early warning trigger point includes: extracting the buffer occupancy distribution from the early warning trigger point; determining the data congestion region based on the buffer occupancy distribution; performing throughput density assessment on the data congestion region to generate a high-load monitoring area; and determining the flow control scheduling strategy based on the high-load monitoring area.

[0029] Extract buffer occupancy distribution from early warning trigger points. Analyze the system's buffer usage patterns using trigger records and status information from these points. Collect real-time status data for all early warning trigger points, including current buffer occupancy, maximum occupancy, and occupancy change rate. Combine response time and impact range information for each trigger point to calculate buffer occupancy density, which is equal to the ratio of occupancy to the trigger point's impact range. Extend the discrete trigger point data into a continuous buffer occupancy distribution using spatial interpolation. Generate a visual representation of the buffer occupancy distribution, with color intensity indicating varying occupancy levels. Analyze the statistical characteristics of the buffer occupancy distribution, calculating its mean, variance, and other statistical parameters. Identify anomalous regions in the occupancy distribution, corresponding to locations where occupancy deviates significantly from the normal range. Track the dynamic changes in the buffer occupancy distribution through time series analysis, identifying trends and periodic patterns in occupancy growth. Establish classification criteria for the occupancy distribution, categorizing distribution areas into four types: low occupancy, medium occupancy, high occupancy, and saturated occupancy.

[0030] For example, determining the data congestion area based on the buffer occupancy distribution includes: assessing the identification difficulty through the buffer occupancy distribution to determine a positioning strategy, wherein the identification difficulty includes the number of peak occupancy, distribution dispersion, and queue depth; setting region division parameters according to the positioning strategy; and spatially mapping the buffer occupancy distribution based on the region division parameters to determine the data congestion area.

[0031] The difficulty of identification is assessed by evaluating the buffer occupancy distribution to determine the localization strategy. Based on the complexity of the buffer occupancy distribution, an appropriate congestion area identification method is selected. Data congestion areas are those where buffer utilization is abnormally high and data processing speed cannot keep up with input speed. When a UAV simultaneously runs multiple algorithms such as autonomous navigation, target recognition, and environmental modeling, the data requirements of different algorithms overlap at certain points in time, leading to data backlog in some buffers and forming processing bottlenecks. The number of peaks in the occupancy distribution is analyzed; a large number of peaks indicates that congestion locations are dispersed and identification is difficult. The dispersion of the occupancy distribution is calculated, and the dispersion is quantified using variance and coefficient of variation. The depth distribution of the buffer queue is statistically analyzed; areas with large queue depths correspond to locations with severe data backlogs. The identification difficulty is calculated using the formula D_difficulty=w1×N_peaks+w2×CV_distribution+w3×D_queue, where D_difficulty is the identification difficulty, N_peaks is the number of occupancy peaks, CV_distribution is the distribution dispersion, D_queue is the queue depth, and w1, w2, and w3 are weighting coefficients. The appropriate localization strategy is selected based on the numerical value of the recognition difficulty: a simple threshold method is used for low difficulty, cluster analysis is used for medium difficulty, and a comprehensive analysis method is used for high difficulty. A decision rule for strategy selection is established, and the most suitable recognition algorithm is automatically selected based on the difficulty analysis results.

[0032] Region segmentation parameters are set according to the positioning strategy. Based on the determined positioning strategy, key parameters of the region segmentation algorithm are configured. For the threshold method strategy, parameters such as occupancy threshold and connectivity radius are set. The threshold setting adopts an adaptive method, dynamically adjusting according to the overall load level of the current system. For the clustering analysis strategy, parameters such as the number of clusters, distance metric, and convergence criteria are set. The number of clusters is automatically determined through data analysis methods, and Euclidean distance is used as the distance metric. For the comprehensive analysis strategy, multiple decision conditions and weight coefficients are set. Parameter settings are optimized through historical data analysis to ensure the accuracy and stability of the recognition results. An adaptive parameter adjustment mechanism is established to adjust parameter settings based on the quality feedback of the recognition results. After the parameter settings are completed, the algorithm is tested to ensure that the algorithm can meet the requirements of real-time processing.

[0033] Data congestion regions are determined by spatial mapping of buffer occupancy distribution based on region partitioning parameters. Using the set region partitioning parameters, the buffer occupancy distribution is converted into discrete congestion regions. The selected region partitioning algorithm is executed to divide the continuous occupancy distribution into several discrete region units. The occupancy statistics of each region unit are calculated, including average occupancy rate, maximum occupancy rate, and occupancy change rate. Congestion is determined for region units based on these statistical characteristics; units meeting the congestion criteria are marked as congested regions. Adjacent congested region units are merged to form continuous congestion domains. The geometric attributes of the congestion domains are calculated, including area, center location, and boundary complexity. A hierarchical structure of congestion regions is established, subdividing large congestion domains into multiple sub-regions. Congestion regions are numbered and classified according to their congestion severity as mild, moderate, and severe. A spatial distribution map and attribute table of congestion regions are generated, providing detailed regional information for flow control decisions.

[0034] High-load monitoring zones are generated by assessing throughput density in data congestion areas. Based on the identified congestion areas, the data processing throughput and load density of each zone are calculated. The data inflow and outflow rates within the congestion zone are measured; the inflow rate reflects the speed of data generation, and the outflow rate reflects the speed of data processing. The throughput density calculation formula is T_density=(R_in-R_out) / A_region, where T_density is the throughput density, R_in is the data inflow rate, R_out is the data outflow rate, and A_region is the area of ​​the region. Positive throughput density indicates data backlog, while negative throughput density indicates sufficient processing capacity. Regions with high throughput density are identified as high-load areas, requiring focused monitoring and optimization. The load saturation of high-load areas is calculated, reflecting how close the current load is to the maximum processing capacity. The development trend of high-load areas is predicted using load prediction methods, based on the data change rate. The prediction results are compared with set monitoring thresholds to determine the areas requiring enhanced monitoring. A spatial distribution map of the high-load monitoring zones is generated, marking monitoring priorities and warning levels.

[0035] Flow control scheduling strategies are determined based on high-load monitoring zones. Using the generated high-load monitoring zone information, targeted flow control and scheduling strategies are formulated. Flow control scheduling strategies are management schemes that dynamically adjust data processing priorities and resource allocation according to system load conditions. When UAVs encounter communication bandwidth limitations while performing emergency rescue missions, the scheduling strategy prioritizes location and communication data, appropriately delaying image processing and storage operations. The load characteristics and causes of high-load monitoring zones are analyzed to identify the main data sources and processing bottlenecks causing the load. When formulating flow control scheduling strategies, the response time and processing priority of early warning trigger points are comprehensively considered. Trigger points with short response times adopt a fast response strategy, while trigger points with high processing priorities are allocated more computing resources. Corresponding scheduling strategies are designed based on the temporal characteristics of the load: a resource expansion strategy is used for continuous loads, and a flow smoothing strategy is used for bursty loads. The resource requirements and performance expectations of different scheduling strategies are calculated, and the optimal combination of flow control scheduling strategies is selected. The strategy combination includes measures such as data source flow limiting, buffer expansion, processing priority adjustment, and load balancing. An implementation schedule and resource allocation plan are formulated to ensure smooth strategy execution and system stability.

[0036] A hierarchical preprocessing pipeline scheme is developed based on flow control scheduling strategies. Utilizing the derived flow control scheduling strategy, a multi-level data preprocessing pipeline architecture is designed. This hierarchical preprocessing pipeline scheme establishes differentiated processing pipelines based on data importance and processing urgency. When the UAV simultaneously receives data from multiple sensors, flight safety-related IMU data enters the high-priority pipeline for the fastest processing, while environmental monitoring data enters the low-priority pipeline for background processing. Data streams are divided into three levels—high-priority, medium-priority, and low-priority—based on data importance and timeliness requirements. High-priority data includes flight safety-related sensor data, medium-priority data includes navigation and mission execution data, and low-priority data includes environmental monitoring and log recording data. A dedicated preprocessing pipeline is designed for each priority level, encompassing data reception, buffering, filtering, compression, and transmission. The high-priority pipeline employs a pass-through processing mode to reduce processing latency; the medium-priority pipeline uses a batch processing mode to balance latency and efficiency; and the low-priority pipeline uses a background processing mode to fully utilize idle resources. This hierarchical preprocessing pipeline scheme achieves differentiated management of data processing, improving the overall processing efficiency and responsiveness of the system.

[0037] Step S130: Obtain the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and establish a processing time window by pruning the effective signal bandwidth based on the real-time constraints.

[0038] Specifically, the sensor noise sequence and real-time constraints are obtained. By analyzing the output signals of each sensor, noise and useful signal components are extracted. The sensor noise sequence contains temporal variation patterns of various interference signals. When a UAV flies over a city with a complex electromagnetic environment, GPS signals are affected by multipath noise from building reflections, and radio communication is affected by base station interference, generating electromagnetic noise. These noise sequences need to be identified and filtered to ensure data quality. Frequency domain analysis is used to decompose the sensor signals into multiple frequency components, identifying noise and signal frequency bands. The noise sequence includes various types such as thermal noise, quantization noise, transmission noise, and environmental interference noise. Statistical analysis methods are used to identify the characteristic parameters and variation patterns of various noise types. Simultaneously, the system's real-time constraints are obtained, including the maximum data processing delay, minimum processing frequency, and upper limit of response time. Flight control-related data has the most stringent real-time requirements, followed by navigation data, while monitoring data has relatively relaxed real-time requirements. The minimum processing frequency is determined based on the stability requirements of the control loop; different control loops have different requirements for processing frequency. The upper limit of response time considers the system's safety boundary and performance margin, adding a safety margin to the normal delay.

[0039] The effective signal bandwidth is generated by mapping the noise sequence to the data alignment field. Using the acquired sensor noise sequence and the data alignment field, the frequency range of the effective signal is calculated. The effective signal bandwidth refers to the frequency range that can carry useful information after removing noise interference. When a UAV's lidar operates in rainy or foggy weather, water droplet scattering can generate strong noise in certain frequency bands. The system needs to identify these contaminated frequency bands and use only clean frequency bands for ranging calculations. The spectral characteristics of the noise sequence are projected onto the spatial coordinates of the alignment field using mathematical fitting techniques. Regions with lower noise energy are identified in the alignment field; these regions correspond to frequency bands with better signal quality. The effective signal bandwidth is calculated as B_effective = B_total - B_noise, where B_effective is the effective signal bandwidth, B_total is the total sensor bandwidth, and B_noise is the noise-occupied bandwidth. The noise-occupied bandwidth is obtained by accumulating the frequency bands where the noise power spectral density exceeds a threshold. The distribution pattern of the effective signal bandwidth in the alignment field is calculated; the effective bandwidth varies between different sensors and different time periods. A continuous effective signal bandwidth distribution map is generated using spatial interpolation methods. Analyze the temporal stability of bandwidth distribution to identify regions and time periods with significant bandwidth fluctuations. Classify the effective signal bandwidth into three levels: high quality, medium quality, and low quality for effective management.

[0040] In some embodiments, the step of establishing a processing time window based on the real-time constraint requirement for pruning the effective signal bandwidth includes: performing priority separation on the effective signal bandwidth to extract key data components; performing delay filtering on the key data components to generate a priority transmission set; performing a matching analysis between the priority transmission set and the real-time constraint requirement to obtain time feature values; and determining the processing time window based on the time feature values.

[0041] The effective signal bandwidth is prioritized to extract key data components. Based on the quality level and data importance of the effective signal bandwidth, bandwidth resources are prioritized. Key data components refer to the information most important for flight safety and mission execution. When the UAV performs a precision landing mission, altitude, attitude, and position information are key data components, requiring the highest quality bandwidth resources and the fastest processing speed. Environmental temperature, humidity, and other monitoring data are considered general data components. The effective signal bandwidth is divided into three priorities according to the security importance of the data: key data, important data, and general data. Key data includes parameters directly affecting flight safety such as attitude angles, angular velocities, altitude, and airspeed, allocated the highest priority and highest quality bandwidth resources. Important data includes parameters affecting the flight mission such as GPS position, heading angle, and engine status, allocated medium priority and medium quality bandwidth. General data includes monitoring parameters such as environmental temperature, humidity, and battery voltage, allocated the lowest priority and remaining bandwidth resources. The bandwidth allocation for key data components is determined through mathematical calculations. The bandwidth allocation for each priority data component is calculated to ensure that key data receives sufficient bandwidth. A dynamic priority adjustment mechanism is used to adjust priority weights according to flight status and mission phase.

[0042] A priority transmission set is generated by performing delay filtering on key data components. Using the extracted key data components, data is prioritized for transmission based on transmission delay characteristics. The priority transmission set refers to the set of data that meets real-time requirements and has transmission priority. When the UAV performs obstacle avoidance maneuvers, the processing delay of sensor data must be controlled at the millisecond level. The system will prioritize the transmission and processing of key data such as distance sensor and IMU data, while temporarily delaying the processing of image and log data. The transmission delay of each data stream in the key data components is measured, and the delay includes three parts: acquisition delay, processing delay, and transmission delay. Acquisition delay is determined by the sensor response time, processing delay by the computational complexity of the data processing algorithm, and transmission delay by network bandwidth and transmission distance. The delay filtering condition is that the total transmission delay is less than a set delay threshold. The delay threshold is set according to the real-time requirements of the data; different types of data have different delay requirements. Data streams that meet the delay condition are added to the priority transmission set. Data streams that do not meet the delay condition undergo delay optimization processing, including compression algorithm optimization, transmission path adjustment, and caching strategy improvement. The transmission efficiency of the priority transmission set is improved through transmission technology optimization.

[0043] A matching analysis is performed between the priority transmission set and real-time constraints to obtain time characteristic values. Based on the generated priority transmission set and real-time constraints, the degree of matching between the two is analyzed. The total data volume and transmission time requirement of the priority transmission set are calculated and compared with the time limit of the real-time constraints. The matching degree is calculated using the formula M_match = T_available / T_required, where M_match is the matching degree, T_available is the available processing time, and T_required is the required processing time. A matching degree greater than 1 indicates sufficient time, less than 1 indicates insufficient time, and equal to 1 indicates just the right time. The distribution characteristics of the matching degree are analyzed to identify time-critical data types and processing stages. Time characteristic values ​​are calculated, including maximum processing time, average processing time, time variability, and time stability. Maximum processing time reflects the time requirement under worst-case conditions, and average processing time reflects the time consumption under typical conditions. The time variability reflects the degree of fluctuation in processing time, and time stability reflects the predictability of processing time. Through statistical analysis of the time characteristic values, the main factors affecting processing time are identified.

[0044] The processing time window is determined based on time characteristic values. Optimal data processing time window parameters are set using the calculated time characteristic values. The processing time window is the time period allocated by the system to complete data processing tasks. When a UAV needs to process multiple sensor data simultaneously in complex tasks, the size of the time window directly affects whether the system can complete all processing tasks in a timely manner. A window that is too small may lead to incomplete processing, while a window that is too large may affect real-time response. An upper limit for the time window is set based on the maximum processing time to ensure data processing can be completed even in the worst-case scenario. A standard value for the time window is set based on the average processing time to meet processing requirements in most cases. The time window calculation formula is T_window = α × T_max + β × T_average, where T_window is the processing time window, T_max is the maximum processing time, T_average is the average processing time, and α and β are weighting coefficients. The weighting coefficients are set according to the system's reliability requirements; α focuses on worst-case protection, and β considers average performance. The impact of the rate of change of time on the window size is considered; the window size is appropriately increased to provide a buffer when the rate of change of time is large. The impact of time stability on window adjustment is considered; a fixed window can be used when stability is high, and a dynamic window is used when stability is low. The time window parameters for different flight phases are calculated. Smaller time windows are used during takeoff and landing to improve response speed, while larger time windows are used during cruise to improve processing quality. The window size is dynamically optimized based on the current processing load and performance requirements through real-time adjustment of the time window.

[0045] Step S140: Use a hierarchical preprocessing pipeline scheme to perform data filtering within the processing time window to obtain feature distribution changes, perform aggregation path planning according to feature distribution changes to identify data aggregation locations, and perform throughput assessment on data aggregation locations to form flow control anchors.

[0046] Specifically, a hierarchical preprocessing pipeline scheme is used to perform data filtering within processing time windows to obtain changes in feature distribution. Sensor data is processed in a hierarchical manner using this pipeline scheme and processing time windows. Changes in feature distribution reflect the evolution of the statistical characteristics of the data in time and space. When a UAV transitions from a stable cruise state to complex maneuvering flight, the variance of IMU data increases significantly, the update frequency of GPS data may decrease due to signal obstruction, and the brightness distribution of camera data changes due to changes in viewing angle. These feature change patterns can reflect the transition of flight state. Input data streams are allocated to corresponding preprocessing pipelines according to priority: high-priority data enters the fast channel, medium-priority data enters the standard channel, and low-priority data enters the batch processing channel. Feature extraction is performed on the data within each time window, including the basic statistical parameters and trends of the data. Through a sliding processing mechanism of time windows, feature information is continuously extracted from multiple windows. The feature differences between adjacent time windows are calculated by comparing the features of the current window with those of the previous window, reflecting the temporal variation pattern of data features. The statistical characteristics of feature distribution changes are analyzed, including the magnitude, direction, and frequency of change. Identify anomalous patterns in feature distribution changes, which correspond to sensor malfunctions, environmental interference, or sudden changes in system load. Track the evolution path of the feature distribution by analyzing the change trajectories in the feature space.

[0047] In some embodiments, the step of identifying data aggregation locations by performing aggregation path planning according to the feature distribution changes includes: dividing the data density window in the feature distribution changes to generate a density window set; performing aggregation path planning based on the density window set to obtain a target aggregation area; performing format conversion and modulation processing on the target aggregation area to generate aggregation load; and using the aggregation load to perform traffic transfer processing to identify data aggregation locations.

[0048] The data density window set is generated by partitioning the data space based on the spatial distribution characteristics of the feature distribution variation. The spatial boundaries of the window partitioning are determined according to the distribution range of data points in the feature distribution variation. A window partitioning method is adopted, using smaller windows in dense data areas and larger windows in sparse data areas. The number of data points and average density within each density window are calculated, and the density distribution characteristics of the window are statistically analyzed. Three types of windows—high-density, medium-density, and low-density—are identified, with high-density windows corresponding to candidate regions of data aggregation. The density relationship between adjacent windows is analyzed, and the boundaries of windows with significant density variations are identified. Adjacent windows with similar density characteristics are merged to form continuous density regions. A density window set is generated, containing the location, size, density value, and type identification information of all windows.

[0049] Target clustering areas are obtained through clustering path planning based on density window sets. Using the generated density window sets, paths for data to cluster towards high-density areas are calculated. Data clustering locations refer to the spatial locations where multiple data streams naturally converge and have the highest processing efficiency. When a UAV simultaneously operates multiple subsystems such as navigation, obstacle avoidance, and mission execution, relevant data from different sensors converge at certain processing nodes; these nodes are the data clustering locations. Starting from each data point location, the distance and path characteristics to each high-density window are calculated. Path planning algorithms are used to calculate the path from the data point to the high-density window, with paths planned along the direction of increasing density. The convergence points of multiple clustering paths are analyzed; these convergence points correspond to the intersections of multiple data streams. The clustering intensity of the convergence points is calculated using the number of convergence paths and the flow rate. The convergence point with the highest clustering intensity is selected as the primary target clustering area, and convergence points with relatively high clustering intensity are selected as secondary target clustering areas. The spatial distribution pattern of the target clustering areas is analyzed to identify the shape, size, and boundary characteristics of the clustering areas.

[0050] For example, the step of performing format conversion and modulation processing on the target cluster area to generate cluster load includes: using the target cluster area as a conversion center to perform format compatibility detection on surrounding data blocks to obtain format distribution data; performing conversion overhead analysis on the format distribution data to generate overhead distribution; continuously narrowing the conversion range using the overhead distribution to improve processing efficiency; and determining the conversion area as cluster load when the processing efficiency meets real-time requirements.

[0051] The target cluster area is used as the conversion center to perform format compatibility testing on surrounding data blocks to obtain format distribution data. Starting from the center of the cluster area, the data format distribution is detected outward in an expanding manner. The clustered load is a set of data that has undergone format unification and processing optimization. When the UAV's vision system needs to fuse 24-bit color images from RGB cameras, 16-bit grayscale images from infrared cameras, and 32-bit distance data from LiDAR, these different formats of data need to be converted into a unified format to form the clustered load. The number and distribution of data blocks of different formats are counted within each detection range. Data block formats include basic types such as integer format, floating-point format, string format, and binary format. The compatibility degree between various formats is calculated. Formats with high compatibility have low conversion overhead, while formats with low compatibility require complex conversion processing. The complexity of format conversion is analyzed. Simple format conversion only requires data type conversion, while complex format conversion requires data structure reorganization and changes in encoding methods. The format distribution data characteristics at different distance ranges are statistically analyzed to generate a spatial map of the format distribution. Clustering and dispersion patterns of the format distribution are identified. Clustering patterns correspond to concentrated areas of data with the same format.

[0052] Cost analysis is performed on the format distribution data to generate a cost distribution. The specific costs of converting data for each format distribution are analyzed, including CPU computation time, memory usage, and storage space requirements. When processing high-resolution image data, converting from raw RAW format to compressed JPEG format requires significant CPU computation for image compression algorithms, while converting from 16-bit integers to 32-bit floating-point numbers only requires a simple numeric type conversion. The actual performance of format conversion is measured, and cost data for different conversion operations is obtained through benchmarking. Conversion time costs are calculated, including format parsing time, data reconstruction time, and encoding / writing time. Conversion space costs are evaluated, including intermediate cache space and temporary storage space requirements. Conversion resource costs are analyzed, including CPU utilization, memory bandwidth, and I / O operation frequency. The total conversion cost at each location in the computation space is calculated, equal to the sum of the conversion costs of all data blocks at that location. A visual representation of the cost distribution is generated, with high-cost areas displayed in warm colors and low-cost areas displayed in cool colors.

[0053] Utilizing cost distribution to continuously narrow the conversion range improves processing efficiency. Based on the generated cost distribution, an iterative optimization method is used to narrow the processing range for format conversion. Regions with lower costs are identified from the cost distribution map as preferred conversion ranges, as these regions have higher conversion processing efficiency. When the system needs to complete a large number of data format conversions within a limited time, low-cost data blocks are prioritized for processing, while high-cost, complex format data is temporarily deferred to ensure that core data can be converted in a timely manner. A threshold adjustment method is used to gradually increase the cost threshold to narrow the conversion range. A relationship model between the cost threshold and processing efficiency is established, and efficiency optimization is achieved by adjusting the threshold. After each adjustment, the processing efficiency within the conversion range is recalculated, and efficiency is measured by the amount of converted data completed per unit time. The trend of processing efficiency changes is monitored, and threshold adjustment is stopped when the efficiency improvement becomes smaller. When the processing efficiency reaches the preset target, range adjustment is stopped, and the current conversion range is the optimal range. The impact of range adjustment on data coverage is analyzed to ensure that the main data content is still covered after adjustment.

[0054] When processing efficiency meets real-time requirements, the conversion area is determined as the clustered load. Based on the optimized conversion range and processing efficiency assessment, the final clustered load area is determined. The processing efficiency of the current conversion range is calculated and compared with the system's real-time requirements. When the flight control system requires data processing latency to be no more than 10 milliseconds, the system will strictly check whether the current conversion area can complete all necessary format conversion operations within the time limit. The current processing efficiency needs to meet or exceed the minimum processing efficiency standard required by the system. A multi-dimensional indicator system for efficiency evaluation is established, including processing speed, resource consumption, and output quality. Real-time performance testing is conducted to verify processing efficiency by simulating real load conditions. When the efficiency check passes, the current conversion area is determined as the final range of the clustered load. Boundary optimization is performed on the determined clustered load area, using a boundary smoothing algorithm to eliminate irregular boundary shapes. Key performance parameters of the clustered load are calculated, including total data volume, average processing time, and expected completion time.

[0055] This system utilizes load balancing to identify data aggregation locations. It calculates traffic allocation schemes for transferring aggregated loads to different candidate locations, taking into account factors such as location processing capacity, current load status, and network connection quality. When the UAV's main flight control processor becomes overloaded due to executing complex maneuvering algorithms, the system will transfer some sensor data processing tasks to the less loaded navigation processor or image processing module, avoiding system performance degradation caused by single-point overload through intelligent traffic allocation. A comprehensive evaluation system for candidate locations is established to assess the sufficiency of computing resources and data processing efficiency at each location. The real-time load status of candidate locations is analyzed, including CPU utilization, memory usage, and I / O bandwidth usage. The matching degree of processing capacity between candidate locations is evaluated to ensure that the transferred data types match the processing expertise of each location. A load balancing algorithm is used to allocate aggregated loads, ensuring a relatively balanced distribution of load across all locations by monitoring the load level of each candidate location in real time. A dynamic load adjustment mechanism is established to adjust the traffic allocation strategy in real time based on the system's operating status. The transmission paths and overhead of traffic transfer are analyzed, selecting the transfer path with the highest transmission efficiency and lowest overhead. Calculate the load saturation at each location after the transfer, and use saturation analysis to identify ideal aggregation locations with sufficient processing capacity and moderate load. Identify the optimal data aggregation location, which should have sufficient processing capacity, low transmission overhead, and good load balancing characteristics.

[0056] Throughput assessment is performed on data aggregation locations to form flow control anchors. Using the identified data aggregation locations, the data processing throughput capacity of each location is calculated. Flow control anchors are key control nodes with flow regulation and load balancing functions. When UAVs encounter data processing bottlenecks while performing complex tasks, flow control anchors can dynamically adjust the data flow direction, redistributing data from overloaded nodes to idle nodes to alleviate congestion. The data inflow and outflow at the aggregation location are measured. Inflow reflects the rate at which data converges to that location, and outflow reflects the rate at which data is processed from that location. The throughput assessment formula is T_throughput = min(R_in, R_capacity), where T_throughput is the actual throughput, R_in is the data inflow rate, and R_capacity is the location's processing capacity. The actual throughput is taken as the minimum of the inflow rate and processing capacity, ensuring that it does not exceed the processing capacity limit. The temporal variation characteristics of throughput are analyzed to identify peak and trough times. The utilization level of throughput is calculated. When the utilization level is close to full load, the aggregation location is approaching its processing capacity limit and flow control is required. Aggregation locations with stable throughput and moderate utilization are selected as flow control anchor points. Flow control anchors perform data flow regulation and load balancing functions, maintaining system stability by controlling the inflow and outflow rates of data. The control capability and influence range of each anchor are calculated; anchors with strong control capabilities can handle larger flow fluctuations.

[0057] Step S150: Perform time slice analysis on the flow control anchor to form the queue backlog rate, perform phase adjustment by fusing the queue backlog rate with the clock synchronization signal to generate the data flow rate, and perform load balancing to generate collaborative processing parameters according to the data flow rate.

[0058] Specifically, time-slice analysis is performed on the flow control anchor to generate the queue backlog rate. The working time of the flow control anchor is divided into fixed-length time slices for systematic monitoring and analysis. The queue backlog rate reflects the congestion level of the data processing queue. When the UAV is performing multi-task parallel processing, tasks such as image recognition, path planning, and communication transmission will queue in the processing queue. The backlog rate can quantify the busyness of the queue. The data queue status of the control anchor is monitored within each time slice, recording the number of data packets, total data volume, and waiting time in the queue. The instantaneous length and average length of the queue are calculated. The instantaneous length reflects the queue status at a certain moment, and the average length reflects the overall queue status within the time slice. The formula for calculating the queue backlog rate is R_backlog = Q_current / Q_capacity, where R_backlog is the queue backlog rate, Q_current is the current queue length, and Q_capacity is the queue capacity. When the backlog rate is close to 1, it indicates that the queue is close to full capacity and flow control is required; when the backlog rate is low, it indicates that the queue processing capacity is sufficient. Analyze the time-varying patterns of the backlog rate to identify its upward, downward, and fluctuating trends. Use data smoothing methods to handle short-term fluctuations in the backlog rate, highlighting long-term trends. Calculate the rate of change of the backlog rate; rapid changes require timely responses and adjustments. Record the backlog rate distribution across different flow control anchors to identify system bottlenecks and load hotspots.

[0059] In some embodiments, the step of performing phase adjustment by fusing the clock synchronization signal with the queue backlog rate to generate data stream speed includes: extracting the rate spike point of the queue backlog rate; dividing the buffer saturation segment according to the rate spike point; setting a back-off buffer duration for the buffer saturation segment based on the clock synchronization signal; and reconstructing the buffer saturation segment using the back-off buffer duration to form data stream speed.

[0060] Extract rate spikes in queue backlog. Based on the calculated queue backlog time series, identify key time points of rapid backlog growth. Perform first-order differencing on the backlog series to obtain the rate of change sequence. Analyze the statistical distribution of the rate of change, calculating statistical parameters such as the mean, standard deviation, and quantiles. The condition for identifying rate spikes is dR / dt > μ + 2σ, where dR / dt is the rate of change in backlog, μ is the mean rate, and σ is the standard deviation rate. When the rate of change exceeds the mean plus twice the standard deviation, the time point is marked as a rate spike. The reliability of the spikes is further verified using a sliding window technique, with a window length of 10 time slices. Examine the rate change pattern before and after the spike within the window to confirm the persistence and significance of the spike. Filter out false spikes caused by noise, retaining spikes caused by genuine system load changes.

[0061] The backlog rate sequence is divided into buffer saturation segments based on rate spikes. Using identified rate spikes, the backlog rate sequence is divided into segments with different levels of saturation. Buffer saturation segments reflect the system's load level at different times. When the UAV switches from simple straight-line flight to complex spiral ascent maneuvers, the processor load increases sharply, forming a high-saturation segment, while a low-saturation segment forms during stable cruise. The backlog rate sequence is divided into multiple consecutive time periods using rate spikes as boundaries. The average level and trend of the backlog rate within each time period are analyzed, and the time periods are classified according to the backlog rate level. High-saturation segments correspond to periods with consistently high backlog rates, medium-saturation segments correspond to periods with moderate backlog rates, and low-saturation segments correspond to periods with relatively low backlog rates. The duration and backlog rate variation of each saturation segment are calculated to assess the severity of the saturation segment. Consecutive high-saturation segments are identified, indicating that the system is continuously under high load. The transition patterns between saturation segments are analyzed to identify the triggering conditions for the transition from low to high saturation.

[0062] The backoff buffer duration is set for buffer saturation sections based on the clock synchronization signal. Utilizing the periodic characteristics of the clock synchronization signal, an appropriate backoff time is set for the identified buffer saturation sections. The data flow rate is the data processing rate optimized through time reconstruction and phase adjustment. When the UAV's sensor data flow is congested, the system improves operational efficiency by inserting backoff buffer time to distribute the data flow density. The periodic stability and phase characteristics of the clock synchronization signal are analyzed to determine the standard period length. The proportion of the backoff buffer duration is determined based on the saturation level of the buffer saturation section; higher saturation sections require longer backoff times. The backoff buffer duration is calculated based on the clock period and saturation level to ensure sufficient backoff time to alleviate buffer pressure. When the clock signal is stable, the standard backoff duration is used; when the clock signal is unstable, the backoff duration is appropriately extended to enhance system stability. A dynamic adjustment mechanism for the backoff duration is established to adjust the backoff parameters according to the real-time system load.

[0063] The data flow rate is determined by reconstructing buffer saturated segments using a backoff buffer duration. Based on the set backoff buffer duration, the time and speed of the buffer saturated segments are reconstructed. A backoff buffer duration is inserted at the beginning of each buffer saturated segment to create a buffer space for data processing. The data flow timing within the buffer segment is reallocated, distributing the originally tightly packed data packets across a longer time window. The average speed of the reconstructed data flow is calculated; the speed equals the total data volume divided by the reconstructed time length. The reconstruction process ensures the timing and integrity of the data without altering its content or logical order. The stability of the reconstructed data flow rate is analyzed, and speed fluctuations are assessed based on the degree of speed change. A data flow rate quality monitoring mechanism is established to ensure that the reconstructed speed meets the system's real-time requirements. Key parameters and performance indicators of the reconstruction process are recorded to provide a reference for system optimization.

[0064] Load balancing is performed based on data flow rate to generate collaborative processing parameters. Based on the generated data flow rate information, processing tasks and computing resources are allocated using a load balancing algorithm. The collaborative processing parameters define the collaborative configuration and performance indicators between multiple processing units. When the UAV needs to process data from GPS, IMU, camera, and radar simultaneously, the collaborative processing parameters determine the collaborative performance status between each processing unit. The differences in data flow rates among various flow control anchors are analyzed to identify the distribution patterns of high-speed and low-speed flows. The total processing capacity and current load distribution of the computing system are calculated; the processing capacity is obtained by summing the computing performance of each processing unit. A load balancing strategy is designed to transfer some tasks from high-load control anchors to low-load control anchors. The load balancing weight calculation formula is W_balance = V_flow / V_total, where W_balance is the load balancing weight, V_flow is the flow rate of a single control anchor, and V_total is the total flow rate of all control anchors. Control anchors with higher weights undertake more processing tasks, while those with lower weights undertake fewer processing tasks. The amount of data transferred and the transmission overhead are calculated, and the transfer scheme with the lowest overhead is selected. The time consumption of the task allocation process is calculated based on the time slice analysis results to obtain the task allocation delay; the resource allocation delay is determined by the resource scheduling time statistics of the load balancing process; the communication protocol delay is generated by combining the transmission time measurement of the clock synchronization signal fusion process; and the synchronization time delay is calculated by using the time deviation analysis of the phase adjustment operation. Cooperative processing parameters are generated, including key performance indicators such as task allocation delay, resource allocation delay, communication protocol delay, and synchronization time delay.

[0065] Step S160: Compare the collaborative processing parameters with the preset delay threshold to generate a scheduling deviation signal, and adjust the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence.

[0066] Specifically, a scheduling deviation signal is generated by comparing the collaborative processing parameters with preset delay thresholds. This signal reflects the gap between the current system performance and the target performance. When the UAV is performing an emergency obstacle avoidance mission, if the task allocation delay exceeds the preset threshold, causing the processing unit to be unable to respond to the obstacle avoidance calculation requirements in a timely manner, the scheduling deviation signal will immediately indicate that the system needs to rebalance the task queue or add parallel processing units to meet real-time requirements. Corresponding delay thresholds are set according to the criticality of different processing stages. The task allocation stage, as the primary factor determining the system's response speed, has the strictest delay requirements; the resource allocation stage affects computational efficiency and has the next strictest requirements; communication protocol delay is related to module collaboration efficiency and has a moderate tolerance; synchronization time delay mainly affects long-term stability and has relatively lenient requirements. The deviation between the actual delay value of each collaborative processing parameter and the corresponding preset threshold is calculated. This deviation reflects the gap between the performance of each processing stage and the target requirements. The scheduling deviation signal is calculated using the formula E_schedule = Σ|Pi_actual - Pi_threshold|, where E_schedule is the comprehensive scheduling deviation signal, Pi_actual is the actual delay of the i-th item (including task allocation, resource allocation, communication protocol, and synchronization time), and Pi_threshold is the corresponding preset delay threshold. A positive deviation component indicates that the delay exceeds the threshold, requiring targeted optimization of the corresponding processing stage; a negative deviation component indicates that the performance of that item is good, with room for further optimization of other stages. The temporal variation pattern of the deviation signal is analyzed to identify the persistence, periodicity, and suddenness of each delay deviation. Statistical methods are used to analyze the distribution characteristics and correlation of the deviation signal to evaluate the performance status and bottleneck locations of each processing stage in the system.

[0067] In some embodiments, adjusting the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence includes: performing time-domain expansion of the scheduling deviation signal to obtain the transient flow change rate; performing peak detection on the transient flow change rate to identify data excitation points; constructing a backpressure injection sequence based on the data excitation points; and using the backpressure injection sequence to perform compensating modulation on the flow control anchor to generate an adaptive preprocessing sequence.

[0068] The transient flow rate of change is obtained by time-domain expansion of the scheduling deviation signal. The deviation signal is expanded sequentially over time to form continuous time-series data. The transient flow rate of change describes the rate of change of system load over a short period. When a UAV suddenly accelerates from low-speed hovering to high-speed forward flight, the amount of sensor data increases dramatically, and the transient flow rate of change can capture this rapid load change pattern. A time window sliding technique is used to segment the deviation signal for analysis, facilitating the identification of change characteristics in different time periods. The rate of change of the deviation signal is calculated within each time window, reflecting the dynamic trend of system performance deviation. High-order differential operations are performed on the deviation signal to extract the rate of change information such as the first and second derivatives. The transient flow rate of change is calculated as dF / dt = (E_schedule(t) - E_schedule(t - Δt)) / Δt, where E_schedule(t) is the scheduling deviation signal at the current moment, E_schedule(t - Δt) is the scheduling deviation signal at the previous moment, Δt is the time interval, and dF / dt is the transient flow rate of change. Analyze the frequency domain characteristics of the rate of change to identify its main frequency components. Calculate the statistical characteristics of the rate of change, including parameters such as peak value, mean, and amplitude. Identify abnormal patterns in the rate of change, which correspond to sudden load changes or fault states in the system. Identify the time correlation and periodicity characteristics of the rate of change through analytical methods.

[0069] Peak detection is used to identify data trigger points in transient flow rate changes. Calculated transient flow rate changes are used to identify key data trigger moments through peak detection. Data trigger points are critical moments when the system load changes significantly. These moments occur when multiple UAV sensors simultaneously encounter data bursts (e.g., a camera suddenly capturing a large number of targets, or radar detecting a dense group of obstacles). The system needs to focus its control on these points. A threshold standard for peak detection is set, based on the statistical distribution of the rate of change. An adaptive threshold method is used to dynamically adjust the detection threshold according to the real-time statistical characteristics of the rate of change. When the rate of change exceeds the set standard, that time point is identified as a peak. The threshold setting strikes a balance between detection sensitivity and false alarm rate. A local extremum search algorithm is used to identify peak points within the threshold range. Peak points correspond to key moments of transient flow rate changes. Detected peaks are verified and filtered to eliminate false peaks caused by noise. The morphological characteristics of the peaks are analyzed, including parameters such as peak height, peak width, and peak slope. Detected peak points are marked as data trigger points, representing moments when the system load changes significantly. The excitation intensity at the excitation point is calculated, and the intensity is calculated by combining the peak characteristics.

[0070] A backpressure injection sequence is constructed based on data excitation points. Using the identified data excitation points, the injection timing and intensity configuration of backpressure control are designed. The backpressure injection sequence is a proactive flow control mechanism applied before the system load peak. When it is predicted that the UAV is about to enter a complex high-load maneuver phase, the system will reduce the processing frequency of non-critical data in advance, reserving sufficient processing resources for critical control data. The time interval and intensity distribution of excitation points are analyzed to determine the timing and frequency of backpressure injection. The corresponding backpressure injection intensity is set according to the intensity level of the excitation points; stronger excitation points require stronger backpressure control. The backpressure injection intensity is calculated based on the characteristics of the excitation points to ensure that the injection intensity matches the degree of load change. A backpressure injection time series is designed, including parameters such as injection time, injection intensity, and injection duration. A predictive injection strategy is adopted to initiate backpressure control in advance before the excitation point arrives. Through a multi-level backpressure injection mechanism, different levels of control measures are adopted according to the severity of the excitation point. The backpressure injection sequence provides the UAV flight control system with proactive load adjustment capabilities, ensuring stable operation of the system under high load conditions.

[0071] An adaptive preprocessing sequence is generated by compensating and modulating the flow control anchors using a backpressure injection sequence. Based on the constructed backpressure injection sequence, compensating modulation processing is applied to the flow control anchors. The adaptive preprocessing sequence is a data processing flow that dynamically adjusts according to real-time load conditions. When the UAV flies in a complex environment, the system automatically adjusts the order, frequency, and accuracy of data processing based on sensor load, processor performance, and task priority, ensuring that critical flight data is always prioritized. The control signals of the backpressure injection sequence are input to each flow control anchor, modulating the anchor's operating parameters and control strategy. The flow limit parameters of the anchor are adjusted according to the strength and direction of the backpressure signal; strong backpressure signals correspond to strict flow limits. The buffer capacity of the anchors is dynamically modulated, increasing the buffer capacity during high load periods and decreasing it during low load periods. The data processing priority of the anchors is adjusted, increasing the priority of important data and decreasing the priority of less important data. The compensation modulation coefficients are determined through mathematical calculations, dynamically based on the backpressure intensity. A multi-anchor collaborative modulation mechanism ensures that the modulation actions of each anchor are coordinated. Finally, a modulated adaptive preprocessing sequence is generated, containing the new configuration parameters and processing flow for all anchors. The adaptive preprocessing sequence can respond in real time to load changes during UAV flight, and ensure the priority processing of critical flight data through dynamic modulation, thereby ensuring flight safety and mission execution reliability.

[0072] To implement the above-described method embodiments, a real-time preprocessing method for multi-source data on an unmanned aerial vehicle (UAV) platform is provided to achieve the corresponding functionalities and technical effects. See also... Figure 2 , Figure 2This diagram illustrates a structural block diagram of a real-time multi-source data preprocessing system 200 for an unmanned aerial vehicle (UAV) platform, according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The real-time multi-source data preprocessing system 200 for an UAV platform, according to an embodiment of this application, includes: The data acquisition module 201 is used to acquire raw data streams and clock synchronization signals from multiple sensors of the UAV, identify sampling rate deviation components based on the raw data streams, generate delay compensation coefficients using the clock synchronization signals, and establish a data alignment field based on the sampling rate deviation components and the delay compensation coefficients. Buffer analysis module 202 is used to scan the data alignment field to form a buffer capacity distribution surface, configure overflow warning trigger points along the buffer capacity distribution surface, deduce the flow control scheduling strategy in reverse through the warning trigger points, and form a hierarchical preprocessing pipeline scheme based on the flow control scheduling strategy. The bandwidth processing module 203 is used to acquire the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and perform clipping processing on the effective signal bandwidth based on the real-time constraints to establish a processing time window; The flow analysis module 204 is used to perform data filtering within the processing time window using the hierarchical preprocessing pipeline scheme to obtain feature distribution changes, perform aggregation path planning according to the feature distribution changes to identify data aggregation locations, and perform throughput evaluation on the data aggregation locations to form flow control anchors. The parameter extraction module 205 is used to perform time slice analysis on the flow control anchor to form a queue backlog rate, perform phase adjustment by fusing the clock synchronization signal with the queue backlog rate to generate a data flow rate, and perform load balancing to generate collaborative processing parameters according to the data flow rate. The control output module 206 is used to compare the collaborative processing parameters with a preset delay threshold to generate a scheduling deviation signal, and adjust the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence.

[0073] The aforementioned multi-source data real-time preprocessing system 200 for UAV platforms can implement the multi-source data real-time preprocessing method for UAV platforms described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0074] The above description is only a part or preferred embodiment of this application. Neither the text nor the drawings should limit the scope of protection of this application. All equivalent structural transformations made using the content of this application's specification and drawings under the overall concept of this application, or direct / indirect applications in other related technical fields, are included within the scope of protection of this application.

Claims

1. A real-time preprocessing method for multi-source data for unmanned aerial vehicle (UAV) platforms, characterized in that, include: The system collects raw data streams and clock synchronization signals from multiple sensors of the UAV, identifies sampling rate deviation components based on the raw data streams, generates delay compensation coefficients using the clock synchronization signals, and establishes a data alignment field based on the sampling rate deviation components and the delay compensation coefficients. The data alignment field is scanned to form a buffer capacity distribution surface. Overflow warning trigger points are configured along the buffer capacity distribution surface. The flow control scheduling strategy is deduced in reverse through the warning trigger points. A hierarchical preprocessing pipeline scheme is formed based on the flow control scheduling strategy. Obtain the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and perform pruning processing on the effective signal bandwidth based on the real-time constraints to establish a processing time window; The hierarchical preprocessing pipeline scheme is used to perform data filtering within the processing time window to obtain feature distribution changes, and aggregation path planning is performed according to the feature distribution changes to identify data aggregation locations. Throughput assessment is performed on the data aggregation locations to form flow control anchors. Time slice analysis is performed on the flow control anchor to form a queue backlog rate. The queue backlog rate is then fused with the clock synchronization signal to perform phase adjustment and generate a data flow rate. Load balancing is then performed according to the data flow rate to generate collaborative processing parameters. The collaborative processing parameters are compared with a preset delay threshold to generate a scheduling deviation signal, and the flow control anchor is adjusted based on the scheduling deviation signal to form an adaptive preprocessing sequence.

2. The method according to claim 1, characterized in that, The step of establishing a data alignment field based on the sampling rate deviation component and the delay compensation coefficient includes: A temporal distortion map is constructed based on the sampling rate deviation component; Inter-frame difference analysis is performed on the temporal distortion map to identify the synchronization degradation rate; Construct a delay correlation matrix based on the synchronization degradation rate and the delay compensation coefficient; A data alignment field is established using the aforementioned time-delay correlation matrix.

3. The method according to claim 1, characterized in that, The process of deriving the flow control scheduling strategy in reverse from the early warning trigger point includes: Extract the buffer occupancy distribution from the warning trigger points; The data congestion region is determined based on the buffer occupancy distribution; The throughput density of the data congestion area is evaluated to generate a high-load monitoring area; The flow control scheduling strategy is determined based on the high-load monitoring area.

4. The method according to claim 1, characterized in that, The process of establishing a processing time window for pruning the effective signal bandwidth based on the real-time constraints includes: Priority separation is performed on the effective signal bandwidth to extract key data components; The key data components are subjected to delayed filtering to generate a priority transmission set; The time feature values ​​are obtained by performing a matching analysis between the priority transmission set and the real-time constraint requirements; The processing time window is determined based on the time characteristic value.

5. The method according to claim 1, characterized in that, The step of identifying data aggregation locations by performing aggregation path planning based on the changes in the feature distribution includes: In the feature distribution variation, data density window partitioning is performed to generate a density window set; Based on the density window set, clustering path planning is performed to obtain the target clustering area; The target aggregation region is subjected to format conversion and modulation processing to generate an aggregation payload; The aggregated load is used to perform traffic transfer processing to identify the location of data aggregation.

6. The method according to claim 1, characterized in that, The step of performing phase adjustment to generate data stream speed by fusing the clock synchronization signal with the queue backlog rate includes: Extract the rate spike in the queue backlog rate; The buffer saturation segment is divided according to the aforementioned rate surge point; The backoff buffer duration is set for the buffer saturation segment based on the clock synchronization signal; The data flow rate is formed by reconstructing the buffer saturation segment using the backoff buffer duration.

7. The method according to claim 1, characterized in that, The step of adjusting the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence includes: The scheduling deviation signal is expanded in the time domain to obtain the transient flow rate of change. Peak detection is performed on the transient flow rate of change to identify the data excitation point; Construct a reverse pressure injection sequence based on the data excitation points; The flow control anchor is compensated and modulated using the backpressure injection sequence to generate an adaptive preprocessing sequence.

8. The method according to claim 3, characterized in that, The determination of data congestion areas based on the buffer occupancy distribution includes: The location strategy is determined by assessing the difficulty of identifying the buffer occupancy distribution, where the difficulty of identification includes the number of peak occupancy values, the dispersion of the distribution, and the queue depth. Set the region division parameters according to the positioning strategy; Based on the region division parameters, the buffer occupancy distribution is spatially mapped to determine the data congestion region.

9. The method according to claim 5, characterized in that, The step of performing format conversion and modulation processing on the target aggregation region to generate the aggregation payload includes: The target cluster area is used as a conversion center to perform format compatibility testing on surrounding data blocks to obtain format distribution data. The cost distribution is generated by performing a conversion cost analysis on the formatted distribution data. The overhead distribution is used to continuously narrow the conversion range and improve processing efficiency. When the processing efficiency meets the real-time requirements, the conversion area is determined to be a clustered load.

10. A real-time preprocessing system for multi-source data for unmanned aerial vehicle (UAV) platforms, characterized in that, include: The data acquisition module is used to acquire raw data streams and clock synchronization signals from multiple sensors of the UAV, identify sampling rate deviation components based on the raw data streams, generate delay compensation coefficients using the clock synchronization signals, and establish a data alignment field based on the sampling rate deviation components and the delay compensation coefficients. The buffer analysis module is used to scan the data alignment field to form a buffer capacity distribution surface, configure overflow warning trigger points along the buffer capacity distribution surface, deduce the flow control scheduling strategy in reverse through the warning trigger points, and form a hierarchical preprocessing pipeline scheme based on the flow control scheduling strategy. The bandwidth processing module is used to acquire the sensor noise sequence and real-time constraints, map the noise sequence to the data alignment field to generate an effective signal bandwidth, and perform clipping processing on the effective signal bandwidth based on the real-time constraints to establish a processing time window; The flow analysis module is used to perform data filtering within the processing time window using the hierarchical preprocessing pipeline scheme to obtain feature distribution changes, perform aggregation path planning according to the feature distribution changes to identify data aggregation locations, and perform throughput evaluation on the data aggregation locations to form flow control anchors. The parameter extraction module is used to perform time slice analysis on the flow control anchor to form a queue backlog rate, perform phase adjustment by fusing the clock synchronization signal with the queue backlog rate to generate a data flow rate, and perform load balancing to generate collaborative processing parameters according to the data flow rate. The control output module is used to compare the collaborative processing parameters with a preset delay threshold to generate a scheduling deviation signal, and adjust the flow control anchor based on the scheduling deviation signal to form an adaptive preprocessing sequence.

Images