Multi-modal environment perception enhancement method and system in complex environment

Abstract

The invention discloses a multi-modal environment perception enhancement method and system in a complex environment. The method comprises the following steps: acquiring multi-modal environment perception data, illumination monitoring data and target demand data, performing boundary recognition on the environment perception data to extract a perception critical point, extracting environment interference distribution from the illumination data, and performing intensity coupling to generate a perception basic field; recognizing sensing bottleneck points and generating an equivalent interference network, and extracting sensing characteristic parameters to construct a sensing response path diagram; primary and secondary channels are extracted through sensing signal shunt analysis, an optimal fusion point is identified, and a cooperative enhancement matrix is generated; recognizing a breakthrough sensing interval by the reconstructed sensing graph and generating a sensing recombination sequence; determining a sensing intersection time period to obtain a sensing compression factor to construct a sensing control instruction set; and an optimal suppression path is identified through fluctuation suppression analysis, fluctuation correction is performed on the perception recombination sequence to generate an environment perception enhancement instruction, and perception performance optimization of the unmanned aerial vehicle in a complex environment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle environment perception, and particularly relates to a multi-modal environment perception enhancement method and system in complex environment. BACKGROUND

[0002] The autonomous flight of unmanned aerial vehicles in complex environments requires the acquisition of environmental information by multiple sensors, including visual cameras, ranging radars, thermal imaging devices, etc. However, existing unmanned aerial vehicle environment perception systems generally have limited perception capabilities, especially in environments with complex light changes, poor weather conditions, or strong radio interference. The quality of sensor data decreases, resulting in insufficient perception accuracy, which affects flight safety and task completion quality.

[0003] Current environment perception technology mainly uses independent sensors or basic multi-sensor combination methods, lacks in-depth analysis and effective control of environmental disturbance factors, and cannot adaptively adjust perception configurations and algorithm strategies according to actual environmental conditions. Traditional methods often ignore the mutual influence and interference source distribution law between different detection devices, making it difficult to achieve dynamic optimization of perception efficiency, especially in the positioning of weak perception links and performance breakthrough areas. Therefore, there is an urgent need for a new environment perception enhancement technology to solve at least one of the above problems. SUMMARY

[0004] The present application discloses a multi-modal environment perception enhancement method and system in complex environment, aiming to solve the technical problems of low perception accuracy, weak anti-interference ability, and insufficient adaptability in existing unmanned aerial vehicle environment perception systems. Through multi-level perception data processing, intelligent interference suppression, and dynamic parameter optimization, the environment perception capability of unmanned aerial vehicles is comprehensively improved, providing reliable technical support for the safe flight and precise operation of unmanned aerial vehicles in complex environments.

[0005] The present application discloses a multi-modal environment perception enhancement method in complex environment, comprising the following steps: Obtain multi-modal environment perception data, light monitoring data, and target demand data, identify the boundary of the multi-modal environment perception data to extract the perception critical point, extract the environmental disturbance distribution from the light monitoring data, and generate a perception base field by intensity coupling the perception critical point and the environmental disturbance distribution; Identify the perception bottleneck point based on the perception base field, generate an equivalent interference network by analyzing the interference impedance of the perception bottleneck point, extract the perception feature parameters from the equivalent interference network, and construct a perception response path graph based on the equivalent interference network and the perception feature parameters; Performing a perception signal split analysis on the perception response path graph to extract primary and secondary channels, analyzing the signal strength distribution of the primary and secondary channels to identify an optimal fusion point, and dynamically assigning weights to the optimal fusion point to generate a collaborative enhancement matrix; Reconstructing the perception distribution based on the collaborative enhancement matrix to generate a reconstructed perception map, performing threshold breakthrough matching on the reconstructed perception map and the environmental interference distribution to identify a breakthrough perception interval, and generating a perception reorganization sequence using the breakthrough perception interval; Performing intersection and superposition analysis on the perception recombinant sequence and the target demand data to determine a perception intersection period, obtaining a perception compression factor based on the perception intersection period and the perception feature parameter, and constructing a perception control instruction set according to the perception compression factor; A fluctuation suppression analysis is performed on the perception control instruction set to identify an optimal suppression path, and fluctuation correction is performed on the perception reorganization sequence along the optimal suppression path to generate an environmental perception enhancement instruction.

[0006] A second aspect of the present invention provides a multimodal environmental perception enhancement system in a complex environment, comprising: A data acquisition module is configured to acquire multimodal environmental perception data, illumination monitoring data, and target demand data, perform boundary identification on the multimodal environmental perception data to extract perception critical points, extract environmental interference distribution from the illumination monitoring data, and perform intensity coupling between the perception critical points and the environmental interference distribution to generate a perception basic field; an interference analysis module, configured to identify a perception bottleneck point based on the perception basic field, perform interference impedance analysis on the perception bottleneck point to generate an equivalent interference network, extract perception feature parameters from the equivalent interference network, and construct a perception response path diagram based on the equivalent interference network and the perception feature parameters; A collaborative enhancement module is configured to perform a perception signal split analysis on the perception response path diagram to extract primary and secondary channels, analyze the signal strength distribution of the primary and secondary channels to identify an optimal fusion point, and dynamically assign weights to the optimal fusion point to generate a collaborative enhancement matrix; a perception reconstruction module, configured to reconstruct the perception distribution based on the collaborative enhancement matrix to generate a reconstructed perception map, perform threshold breakthrough matching between the reconstructed perception map and the environmental interference distribution to identify a breakthrough perception interval, and generate a perception reorganization sequence using the breakthrough perception interval; a compression control module, configured to perform intersection and superposition analysis on the perception recombined sequence and the target demand data to determine a perception intersection period, obtain a perception compression factor based on the perception intersection period and the perception characteristic parameters, and construct a perception control instruction set according to the perception compression factor; A fluctuation suppression module is configured to perform fluctuation suppression analysis on the set of perception control instructions to identify an optimal suppression path, and to perform fluctuation correction on the perception reorganization sequence along the optimal suppression path to generate environment perception enhancement instructions.

[0007] The beneficial effects of the present application are embodied in the following aspects: first, by constructing a perception basis field and an equivalent interference network, the multi-modal environment perception data and the illumination monitoring data are deeply fused, the perception bottleneck point and the interference source distribution are effectively identified, and a unified processing framework of environment perception and interference suppression is established, which improves the perception stability and accuracy of the unmanned aerial vehicle in a complex illumination and electromagnetic interference environment. Second, by using the cooperative enhancement matrix and the perception reorganization sequence technology, through primary and secondary channel shunt analysis and optimal fusion point identification, intelligent cooperative processing of multi-sensor data is realized, the complementary advantages of different sensors are effectively utilized, and the overall perception performance is improved, especially when the sensor is partially failed or the performance is degraded, the perception effect is still good. Finally, through dynamic optimization of the perception compression factor and the fluctuation suppression path, an adaptive perception parameter adjustment mechanism is established, which can adjust the perception strategy in real time according to the actual flight environment and task requirements, generate targeted environment perception enhancement instructions, effectively suppress the random fluctuations and systematic interference of the perception data, and improve the real-time performance and robustness of the unmanned aerial vehicle environment perception.

[0008] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 is a flow diagram of a multi-modal environment perception enhancement method in a complex environment according to the present application.

[0010] Figure 2 is a structural block diagram of a multi-modal environment perception enhancement system in a complex environment according to the present application. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0012] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative positional relationship, motion condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0013] It should also be noted that when an element is referred to as being "fixed on" or "disposed on" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element.

[0014] In addition, the descriptions of "first", "second", etc. in this application are for descriptive purposes only and should not be understood as indicating or implying their relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0015] The technical solutions of the embodiments of this application are introduced below.

[0016] like Figure 1 As shown, an embodiment of the present invention provides a method for enhancing multimodal environmental perception in a complex environment, comprising the following steps S110 to S160: Step S110: Acquire multimodal environmental perception data, illumination monitoring data, and target demand data, perform boundary identification on the multimodal environmental perception data to extract perception critical points, extract environmental interference distribution from the illumination monitoring data, and perform intensity coupling between the perception critical points and the environmental interference distribution to generate a perception basic field.

[0017] Specifically, multimodal environmental perception data, light monitoring data, and target demand data are acquired. The integrated sensor system onboard the UAV platform simultaneously collects multi-dimensional information about the flight environment. Multimodal environmental perception data is acquired using visual sensors, lidar, infrared sensors, and ultrasonic sensors. The visual sensors utilize a binocular stereo camera configuration to capture color and depth information. The lidar uses a scanning mode to acquire precise distance and contour information, covering the entire drone's surroundings. The infrared sensor captures thermal information about the environment and the outline of targets in low-light conditions. The ultrasonic sensors, arranged in a circular array, are primarily used for close-range obstacle detection and ground ranging. Light monitoring data is acquired using a light intensity sensor and a spectrometer. The light intensity sensor monitors changes in ambient light intensity in real time, while the spectrometer analyzes the spectral characteristics of the light and identifies interference components. Target demand data is input into the mission planning module and includes detailed information such as flight route coordinates, mission objective parameters, performance requirements, and flight constraints. All acquired data is synchronized using a unified timestamp to ensure temporal consistency across multiple sources.

[0018] Boundary recognition is performed on multi-modal environment perception data to extract perception critical points. The data acquired by multi-modal sensors is fused in a unified coordinate system, and the data of different sensors is mapped to the same spatial reference system through coordinate transformation. Boundary detection is performed on the fused environment data, and a gradient detection method is used to identify the change characteristics of the data in space. The first and second order gradients of the data are calculated, and the positions with sharp gradient changes correspond to object boundaries and environmental turning points. In the urban flight environment, the edges of buildings, road boundaries, and the junctions of vegetation and open land are typical positions with sharp gradient changes. An adaptive gradient threshold is set, and the threshold is dynamically adjusted according to the local data characteristics to ensure that the boundary features can be effectively extracted under different environmental conditions. Density clustering analysis is performed on the detected boundary feature points, the clustering radius is set to 0.5 meters, and the minimum number of clustering points is set to 5. The geometric parameters of each boundary segment are calculated, including boundary length, average curvature, main direction angle, and boundary continuity index. The boundary segments that meet the critical conditions of the geometric parameters are selected as the perception critical points, and the specific standards are that the boundary length is greater than 1 meter, the curvature change is greater than 30 degrees, and the direction angle turning is greater than 45 degrees.

[0019] The environmental disturbance distribution is extracted from the light monitoring data. Time series analysis is performed on the light data collected by the light intensity sensor and spectrometer to identify the change pattern and abnormal fluctuation of light intensity. The sliding window method is used to calculate the mean, variance and trend of light intensity, and the sharp change of light in the window indicates the existence of disturbance source. The main frequency components of light change are extracted through frequency domain analysis to identify the characteristic patterns of periodic disturbance and random disturbance. The spectral data is grouped according to the visible, near-infrared and infrared three bands, and the energy distribution characteristics of each band are analyzed. When there is an abnormal energy concentration phenomenon in a certain band, such as strong reflection spot or direct light, these positions correspond to the areas where the disturbance sources are located. Identify the abnormal peak values and energy concentration areas in the spectrum, which correspond to the positions of disturbance sources such as strong reflection and direct light. The spatial orientation and distance information of the disturbance source is calculated using the measurement data of multiple sensor positions. According to the light attenuation characteristics, the intensity level and effective influence range of the disturbance source are calculated. The discrete disturbance measurement points are expanded to a continuous disturbance distribution field covering the entire environment space using an interpolation method.

[0020] In some embodiments, the intensity coupling of the perception critical point and the environmental disturbance distribution generates a perception base field, including: identifying an intensity change rate sequence at the perception critical point; cross-verification positioning the intensity change rate sequence and the environmental disturbance distribution to verify anchor points; deriving coupling coefficients along the verified anchor points; constructing a perception base field according to the distribution range of the coupling coefficients.

[0021] At the perception critical point, the intensity rate of change sequence is identified. At the exact spatial position of each perception critical point, the corresponding interference intensity value in the environmental interference distribution field is extracted. A circular sampling region is established around the critical point, and a sampling grid is set to obtain the intensity distribution information in the neighborhood. The intensity gradient in multiple radiation directions outward from the critical point center is calculated, and the intensity change curve with distance is recorded. For example, at the critical point of the building edge, the interference intensity towards the building interior direction is lower, and the intensity towards the open area direction is higher, forming an obvious intensity gradient change. The first-order rate of change and the second-order rate of change of the intensity are calculated by using the numerical differentiation method, the first-order rate of change reflects the speed of the intensity change, and the second-order rate of change reflects the acceleration characteristics of the change. The rate of change data in multiple directions is statistically processed to generate the intensity rate of change sequence of the critical point. The rate of change sequence is filtered to eliminate the influence of high-frequency noise and measurement error on the sequence characteristics. The main change mode is identified through statistical analysis of the sequence, and the feature identifier of the rate of change sequence is established.

[0022] The intensity rate of change sequence is cross-verified with the environmental interference distribution to locate the verification anchor point. The intensity rate of change sequence of the perception critical point is taken as a feature template, and a pattern matching search is performed in the overall environmental interference distribution. The similarity between the sequence and each position in the distribution field is calculated by using correlation analysis, and the positions with high similarity are marked as matching regions. The direction consistency and amplitude matching degree of the rate of change sequence are further verified in the matching region, and the accuracy of the matching is ensured through multiple tests. The position points that pass the test are subjected to spatial clustering processing, and the position with the highest similarity is selected as a representative point in each clustering cluster. These representative points are the verification anchor points, which represent the corresponding matching positions of the intensity change characteristics of the perception critical point in the environmental space. For example, the rate of change characteristics of the road boundary critical point may find similar matches in the parking lot edge, building outline and other positions to form verification anchor points. The verification anchor point provides a spatial reference and a reference coordinate, ensuring the spatial consistency of the coupling calculation.

[0023] Illustratively, the coupling coefficient is derived along the verification anchor point, including: setting key sampling points around the verification anchor point; obtaining local intensity rate of change and impedance rate of change from the key sampling points; constructing a rate of change curve according to the intensity rate of change and the impedance rate of change; and converting the evolution trend of the rate of change curve into a coupling coefficient.

[0024] Key sampling points are set around the verification anchor points. With each verification anchor point as the geometric center, a multi-layer concentric circle sampling strategy is adopted to set key sampling points. The inner circle radius is set to 0.5 meters, the middle circle radius is set to 1.5 meters, and the outer circle radius is set to 2.5 meters, forming a three-layer sampling structure. 12 sampling points are evenly distributed on the circumference of each layer, and there are a total of 36 key sampling points in three layers, with an angular interval of 30 degrees between the sampling points. According to the importance level of the verification anchor points, the sampling density is adjusted, and 6 additional sampling points are added to the inner circle for anchor points with high importance, forming a high-density sampling mode of 42 points. Each sampling point is precisely positioned by GPS and inertial navigation system, with a positioning accuracy of centimeter level. The sampling points are classified and numbered according to distance and angle, and the association index between the sampling points and the verification anchor points is established. Considering the influence of terrain undulation and obstacle shielding on sampling, the positions of some shielded sampling points are adjusted to ensure that all sampling points can obtain effective measurement data.

[0025] Local intensity variation rate and impedance variation rate are obtained from key sampling points. At the position of each key sampling point, the current environmental disturbance intensity and medium impedance parameters are synchronously measured by multi-modal sensors. The intensity difference between adjacent sampling points is used to calculate the local intensity gradient, improving the accuracy of numerical calculation. The intensity variation rate is decomposed into radial and tangential components, with the radial component representing the intensity attenuation characteristics along the distance direction and the tangential component representing the intensity distribution characteristics along the angle direction. Meanwhile, the environmental medium characteristic parameters at the sampling points are obtained, including dielectric constant, magnetic permeability and other key physical quantities. The variation of each impedance parameter between adjacent sampling points is calculated to obtain the radial and tangential components of the local impedance variation rate. The correlation between intensity variation rate and impedance variation rate is analyzed, and the correlation degree and coupling characteristics are calculated. Taking the anchor point near the building as an example, the intensity variation rate increases rapidly when approaching the building, and the impedance variation rate also shows a corresponding jump, showing a strong correlation between the two.

[0026] Variation rate curves are constructed based on intensity variation rate and impedance variation rate. The intensity variation rate of key sampling points is taken as the abscissa, and the corresponding impedance variation rate is taken as the ordinate, and the data distribution is plotted in the two-dimensional coordinate system. The function relationship between the variation rates is established by curve fitting method, and different fitting models are tried to determine the best description function. The accuracy index of the fitting model is calculated, and the model with the best fitting effect is selected as the final variation rate curve. The geometric characteristics of the fitted curve are analyzed, and the key parameters such as the average slope, the maximum curvature point, and the coordinates of the inflection point of the curve are calculated. The monotonicity interval and extreme position of the variation rate are determined through the derivative analysis of the curve. The symmetry and periodicity characteristics of the curve are analyzed to identify the regularity pattern of the variation rate distribution. The residual error of the fitted curve is analyzed to evaluate the description ability of the curve model to the actual data. The variation rate curve reflects the internal correlation between intensity variation and impedance variation.

[0027] The evolution trend of the rate of change curve is converted into a coupling coefficient. Based on the geometric feature parameters of the rate of change curve, a quantitative calculation model of coupling strength is established. The overall evolution trend of the curve is analyzed. An upward trend indicates positive coupling enhancement, a downward trend indicates negative coupling weakening, and a fluctuating trend indicates the instability of the coupling relationship. The local slope values of each section of the curve are calculated. The sections with large absolute slope values correspond to positions with intense coupling changes. Key feature points of the curve are identified, including the starting point, the ending point, the extreme point, and the inflection point. These feature points reflect the turning and critical states of the coupling relationship. A comprehensive calculation formula for the coupling coefficient C = a x |average slope| + β x maximum curvature + γ x |curve area| is established, where C is the coupling coefficient, a = 0.4, β = 0.3, and γ = 0.3 are empirical weight coefficients. The calculated coupling coefficients are subjected to logarithmic transformation and normalization processing, making the numerical distribution more uniform and the range controlled within the [0, 1] interval. For example, in a complex building environment, the coupling coefficient between the building edge and the strong reflection interference source is usually high, while the coupling coefficient in the open area is relatively low.

[0028] A perception base field is constructed according to the distribution range of the coupling coefficient. The coupling coefficient values of all verification anchor points are counted, the distribution mean, standard deviation, and quartile of the coefficient are calculated, and the classification boundaries of coupling strength are determined. The coupling coefficient is divided into three levels according to the numerical size: strong coupling (0.7-1.0), medium coupling (0.4-0.7), and weak coupling (0.0-0.4). The inverse distance weighted interpolation method is used to extend the discrete coefficient values of the verification anchor points to a continuous spatial distribution field. The interpolation weight index is set to 2, and the search radius is set to 10 meters. In the generated perception base field, each location in space has a clear coupling coefficient value. High coefficient areas represent strong coupling between perception critical points and environmental interference, and low coefficient areas represent weak coupling. The perception base field is visualized by contour lines and color mapping technology, providing intuitive spatial reference for subsequent bottleneck point identification and response path construction.

[0029] In step S120, the perception bottleneck points are identified based on the perception base field, the interference impedance analysis of the perception bottleneck points is performed to generate an equivalent interference network, the perception feature parameters are extracted from the equivalent interference network, and the perception response path graph is constructed based on the equivalent interference network and the perception feature parameters.

[0030] Specifically, the perception bottleneck points are identified based on the perception base field. The spatial distribution characteristics of the coupling coefficient are analyzed from the perception base field to identify the areas with abnormal changes in coupling strength. The locations with significantly low coupling coefficients in the perception base field are searched, which correspond to bottleneck areas with limited perception ability. The coupling coefficient field is processed using numerical analysis methods, and the connected low-coupling areas are extracted as candidate bottleneck regions. In urban environments, narrow streets between high-rise building groups, underground passage entrances, and areas near large metal structures often form perception bottleneck regions, where GPS signal attenuation is severe and visual sensors are affected by obstructions. Basic features of each candidate region are calculated, including area, location coordinates, and statistical properties of the coupling coefficient. Threshold screening is used to select regions with areas exceeding the minimum limit and significantly low coupling coefficients as effective bottleneck regions. The location with the lowest coupling coefficient in each effective bottleneck region is found as the center of the bottleneck point. Considering the spatial relationship between the UAV flight path and the bottleneck point, key bottleneck points near the planned route are identified. The identified bottleneck points are classified according to their potential impact on flight safety, with impact levels divided into high, medium, and low.

[0031] The interference impedance analysis of the perception bottleneck points generates an equivalent interference network. A local analysis area is established at each perception bottleneck point location for interference source characteristic analysis centered on the bottleneck point. The environmental impedance characteristics around the bottleneck point are obtained through sensor measurements, including signal propagation attenuation and reflection characteristics at different frequency bands. The signal strength distribution within the communication and navigation system operating frequency band of the UAV is measured to identify the main interference factors causing signal attenuation. The measured impedance data is converted into an equivalent circuit model, which combines basic resistance, inductance, and capacitance elements to describe complex environmental characteristics. An impedance matrix Z = R + jX is established, where R is the impedance real part, X is the impedance imaginary part, and j is the imaginary unit. The metal outer walls of buildings correspond to large capacitors, power lines produce strong magnetic fields corresponding to inductors, and the conductivity of the ground and walls corresponds to resistors. The mutual influence relationship between bottleneck points is established when the influence ranges of two bottleneck points overlap. The transmission characteristics, including signal propagation loss and delay time between two points, are calculated. All bottleneck points and their mutual relationships are organized into a network topology, with each bottleneck point as a network node and mutual influence relationships as connection edges. The dispersed interference sources are uniformly represented as an equivalent interference network through network modeling.

[0032] In some embodiments, the extraction of perception feature parameters from the equivalent interference network includes: analyzing the node connection mode of the equivalent interference network to identify primary connections and secondary connections; calculating the connection strength distribution based on the primary connections and the secondary connections; deriving a feature spectrum from the connection strength distribution; and extracting core components from the feature spectrum as perception feature parameters.

[0033] The node connection mode recognition of the equivalent interference network mainly includes primary connections and secondary connections. The connection relationships between all nodes in the equivalent interference network are statistically analyzed, and the weight and influence strength of each connection edge are calculated. All edges are sorted by the size of the connection weight, and the connection with a large weight corresponds to a strongly coupled interference transmission path. The weight threshold T = μ + σ is calculated, where T is the classification threshold, μ is the mean of all connection weights, and σ is the standard deviation of the weight. The edges with a weight higher than the threshold T are marked as primary connections, and the edges with a weight lower than the threshold T are marked as secondary connections. Primary connections usually connect important bottleneck nodes and carry the main interference transmission function. For example, the connection between a large shopping center and a nearby parking lot is a primary connection because both are strong interference sources and are close in space. Secondary connections play a supplementary role and provide alternative transmission paths when primary connections fail. The topological mode of the connections is analyzed, and different network structures such as star-shaped connections, chain-shaped connections, and mesh-shaped connections are identified. The network is divided into several tightly connected sub-networks, and the connections within each sub-network are primary connections, while the connections between sub-networks are secondary connections. The proportion of the number of primary connections and secondary connections is counted to evaluate the complexity of the network connection structure.

[0034] The connection strength distribution is calculated based on primary connections and secondary connections. The weight data of primary connections and secondary connections are statistically analyzed respectively, and the mean, variance, and distribution characteristics of each are calculated. A spatial distribution model of connection strength is established, which maps the connection weight to the spatial coordinate system of the network. The density distribution of connection strength in space is calculated by the kernel density estimation method, and the strong connection concentration area and the weak connection sparse area are identified. The connection strength density function ρ(x, y) = Σwi·K((x-xi) / h,(y-yi) / h) is calculated, where wi is the weight of the ith connection, K is the kernel function, and h is the bandwidth parameter. The directional characteristics of connection strength are analyzed, and the statistical properties of connection weight in different directions are calculated. The frequency histogram of connection strength is established, and the main peak value and dispersion degree of the weight distribution are identified. The spatial autocorrelation of connection strength is calculated, and the correlation degree of connection strength in adjacent regions is analyzed. The discrete connection strength data is extended to a continuous intensity field distribution by interpolation methods. The connection strength distribution reflects the spatial distribution pattern and intensity variation law of energy transmission in the interference network.

[0035] Deduce feature spectrum from connection strength distribution. Convert connection strength distribution data to frequency domain for spectral analysis, extract frequency domain feature information. Calculate power spectrum of connection strength distribution, identify main peak frequency and corresponding power intensity in frequency spectrum. Preprocess intensity distribution data to reduce errors and interference during analysis. Analyze bandwidth characteristics of frequency spectrum, calculate effective signal bandwidth and distribution range of main frequency components. Identify main and secondary frequency components in frequency spectrum, analyze the relationship between frequency components. Calculate center frequency of frequency spectrum using power weighting method, determine frequency barycenter based on power values of each frequency point. Calculate center frequency and frequency dispersion of frequency spectrum, evaluate the frequency concentration of connection strength distribution. Describe the shape characteristics and distribution patterns of frequency spectrum through spectral parameters. Establish the correlation between spectral features and network topology characteristics, analyze the influence of network structure on spectral features. Filter the frequency spectrum to remove noise components and retain useful feature information.

[0036] Extract core components from feature spectrum as perceptual feature parameters. Decompose feature spectrum into components, extract main components and key features of frequency spectrum. Calculate the contribution rate of each main component, select the top few main components with higher contribution rate as core components. Analyze the frequency range and power distribution corresponding to the core components, determine the frequency center and bandwidth characteristics of each component. Calculate amplitude, phase and other feature parameters of core components, form a multi-dimensional feature parameter vector. Organize the feature values of each core component into a multi-dimensional feature parameter vector to form a structured parameter representation. Further separate independent signal components in the frequency spectrum to improve the independence and distinguishability of feature parameters. Standardize the extracted perceptual feature parameters to eliminate the influence of dimension differences on analysis. Organize the final extracted perceptual feature parameters into a structured parameter table, containing complete information such as parameter name, value and physical meaning.

[0037] The perception response path graph is constructed based on the equivalent interference network and the perception characteristic parameters. The nodes of the equivalent interference network are mapped to a two-dimensional plane coordinate system according to the spatial position, and the relative position relationship between the nodes is maintained. According to the numerical size of the perception characteristic parameters, each node in the network is assigned a weight, and the weight value reflects the influence degree of the position on the overall perception performance. A plurality of candidate paths from the current position of the unmanned aerial vehicle to the target position are established, and each path passes through different combinations of nodes in the network. The path cost Cost = Σ(wi x di) is calculated, where Cost is the total path cost, wi is the weight of the ith node, and di is the distance between adjacent nodes. The shortest path search method is used to find the path with the minimum perception cost, while considering the safety and feasibility of the path. In the city distribution task, the unmanned aerial vehicle may have three main path options: detouring around high-rise buildings, crossing over the park, and flying along the river. The system will select the optimal solution according to the distribution of perception bottleneck points on each path. Key nodes for perception response are identified in the path graph, which correspond to positions where the environmental perception capability changes significantly. When the environmental conditions change or new bottleneck points appear, the path graph can recalculate and adjust the path selection. Through the visualization of the path graph, the operator can obtain intuitive information about the perception environment state and the flight path.

[0038] In step S130, the perception signal shunt analysis of the perception response path graph is performed to extract the primary and secondary channels, the signal intensity distribution of the primary and secondary channels is analyzed to identify the optimal fusion point, and the dynamic weight distribution of the optimal fusion point is generated to generate the cooperative enhancement matrix.

[0039] Specifically, the perceptual signal shunt analysis is performed on the perceptual response path graph to extract the primary and secondary channels. The shunt characteristics of signal transmission are analyzed from the perceptual response path graph to identify the flow distribution mode of perceptual information on different paths. The flow capacity and actual carrying capacity of each edge in the path graph are calculated by graph theory analysis method to identify the main transmission channel with high flow and the secondary transmission channel with low flow. In combination with the path cost Cost calculated in S120, the signal transmission performance of each channel is evaluated, including transmission capacity, path stability, connection reliability, and path cost, and other key indicators. The channel importance score S = w1 x capacity + w2 x stability + w3 x reliability + w4 x (1 / Cost) is calculated, where S is the channel importance score, w1, w2, w3, w4 are weight coefficients with values of 0.3, 0.3, 0.2, and 0.2, respectively, and the lower the path cost, the higher the importance of the channel. According to the importance score, all channels are sorted, and the top 30% of the channels are marked as primary channels, and the last 70% of the channels are marked as secondary channels. The signal transmission channels along the main streets usually become the primary channels because these paths have good signal coverage and less interference, while the paths that pass through dense building groups are usually secondary channels, and the signal is easily affected by shielding and reflection. The connectivity and redundancy of the primary and secondary channels are analyzed to ensure that the secondary channels can provide backup transmission capacity when the primary channels fail. A channel load balancing mechanism is established to dynamically allocate signal bearing tasks of the primary and secondary channels according to real-time flow conditions.

[0040] The signal intensity distribution of the primary and secondary channels is analyzed to identify the optimal fusion point. Signal intensity measurement is performed on the primary and secondary channels, and multi-point sampling is used to obtain signal intensity data at various positions within the channels. A spatial distribution model of signal intensity is established to analyze the variation and distribution characteristics of intensity along the length direction of the channel. Key nodes of signal convergence or divergence are identified through intensity variation analysis, which usually correspond to important positions for signal processing. Candidate fusion points are searched at the intersection of the primary and secondary channels and positions with similar signal intensities, with a focus on areas with high channel importance scores. The fusion advantages of each candidate point are evaluated, considering factors such as signal intensity matching degree, spatial position rationality, and transmission path efficiency. The airspace above the shopping center is often the intersection of multiple signal channels, providing good fusion conditions, while the narrow channels between high-rise buildings have high signal intensity but poor fusion conditions. The optimal fusion point is determined through comprehensive evaluation as the core node for weight allocation. Considering the dynamic changes during the flight of the unmanned aerial vehicle, the fusion point selection is adjusted in real time according to the flight state and environmental changes.

[0041] In some embodiments, the dynamic weight allocation to the optimal fusion point generates a synergistic enhancement matrix, including: identifying upstream information sources and downstream information sources according to the optimal fusion point implementing information flow tracking analysis; using the upstream information sources to implement information feedback enhancement on the downstream information sources to form enhanced information sources; implementing grid layout of the enhanced information sources to generate a distributed grid; and implementing weight coding on the distributed grid to generate a synergistic enhancement matrix.

[0042] According to the optimal fusion point, upstream information sources and downstream information sources are identified by implementing information flow tracking analysis. Taking the optimal fusion point as the center, bidirectional tracking is performed along the connection edges of the perception response path diagram to identify the source and destination of the information flow. In the upstream direction, all source nodes providing information input to the fusion point are found, which constitute the upstream information source set. In the downstream direction, all target nodes receiving the information output of the fusion point are found, which constitute the downstream information source set. The information contribution and quality characteristics of each upstream information source are analyzed, and the information sources are sorted according to the importance of the contribution degree. The transmission characteristics of the information flow are analyzed to identify the delay time and synchronization requirements of the information transmission. The association between upstream and downstream information sources is established to analyze the transmission path and influence chain of the information flow. The information sources are classified, and functionally similar and geographically close information sources are classified into the same category. The upstream information sources usually include various sensor nodes and monitoring devices, and the downstream information sources are data processing centers and decision control units, and the information flow reflects the complete chain from perception to decision. A complete information transmission network topology is established through information flow tracking.

[0043] For example, the use of the upstream information sources to implement information feedback enhancement on the downstream information sources to form enhanced information sources includes: using the upstream information sources as information providers to transmit enhanced information to the downstream information sources; implementing cyclic control on the enhanced information to form a stable enhancement state; continuously optimizing the information enhancement precision based on the stable enhancement state; and when the enhancement precision meets the requirements, maintaining the stable enhancement state to form an enhanced information source.

[0044] The upstream information source is configured as an information provider to transmit enhanced information to the downstream information source. The data transmission link from the upstream information source to the downstream information source is configured, and the upstream source includes GPS modules, image sensors, laser radars, and other sensing devices, and the downstream source is a flight control system, a navigation module, and other control units. A data packet transmission protocol is established, a 32-byte information header is set to identify the data type and priority, and the payload part carries the specific enhanced data content. The LZ4 compression algorithm is deployed to compress the image enhancement data in real time, and the original 1MB image data is compressed to 300KB, reducing the transmission bandwidth occupation. A primary and backup dual-channel transmission architecture is configured, the transmission delay of the main channel is controlled within 50 milliseconds, and the backup channel automatically takes over when the main channel fails. A transmission priority queue is established, the GPS differential correction information is set as the highest priority, the image enhancement data is set as the medium priority, and the state monitoring information is set as the low priority. The TCP protocol is used to ensure the reliable transmission of the enhanced information, and the acknowledgement retransmission mechanism is used to handle data loss. A 2MB transmission buffer is set to cache the enhanced information to be transmitted when the UAV performs multi-sensor synchronous acquisition tasks.

[0045] The enhanced information is implemented with a cycle control to form a stable enhancement state. A PID feedback control loop is established, the P controller responds to the enhancement effect deviation, the I controller eliminates the steady-state error, and the D controller predicts the trend. The target value of the enhancement effect is set to 0.85, the enhancement intensity is increased when the actual enhancement effect is lower than 0.80, and the enhancement intensity is reduced when the actual enhancement effect is higher than 0.90. The accuracy, completeness, and delay of the enhanced information are monitored, the accuracy is required to be above 95%, the completeness is required to be above 98%, and the delay is controlled within 100 milliseconds. A 50-millisecond control cycle is used, the enhancement effect data is collected once every cycle, and the control parameters are adjusted. Fast control links are deployed to handle 10-millisecond-level transient changes, and slow control links are deployed to handle 100-millisecond-level gradual changes. A 20-point moving average filter is used to eliminate measurement noise, and when the enhancement effect of 50 consecutive control cycles is maintained within the target range ±0.05, it is confirmed that the system enters a stable enhancement state. In the stable state, the system enhancement effect fluctuation amplitude is less than ±2%, and the control output remains stable.

[0046] Based on the stable enhancement state, the information enhancement precision is continuously optimized. On the premise of maintaining the 0.85 enhancement effect, the precision index is further improved. The precision data is collected once every second, and the precision distribution characteristics and improvement space in the stable state are analyzed. The core parameters affecting the precision are identified: the current value of the image enhancement gain coefficient is 1.2, the current value of the laser radar filtering threshold is 0.3, and the current value of the data fusion weight is [0.4, 0.3, 0.3]. The particle swarm algorithm is used to optimize the parameter configuration, and the adjustment amplitude is limited within 3% of the current value to avoid damaging the stable state. A 30-second sliding window is used to evaluate the optimization effect, and when the precision improvement is less than 0.05% for 5 consecutive times, the parameter adjustment is stopped. In the urban environment, the image gain is optimized from 1.2 to 1.35 to adapt to the shadow changes, and the laser radar threshold is adjusted from 0.3 to 0.25 to reduce the noise influence. Through multi-objective optimization, the precision and stability are balanced, the precision weight is set to 0.7, and the stability weight is set to 0.3.

[0047] When the enhancement precision meets the requirements, the stable enhancement state is maintained to form the enhancement information source. An accuracy compliance determination mechanism is established, and when the enhancement precision remains above 90% for 10 consecutive minutes, it is confirmed to meet the requirements. The state maintenance mode is activated, and the optimized parameter configuration is locked, and a ±1% fine-tuning strategy is used to maintain the stable enhancement state. A state monitoring system is deployed to monitor the key parameter deviation in real time, and when the image gain deviation exceeds ±0.05 or the filtering threshold deviation exceeds ±0.02, it is automatically corrected. A three-level early warning mechanism is established, and a 3% decrease in precision sends a reminder, a 5% decrease sends a warning, and an 8% decrease sends an emergency signal. Automatic calibration is performed every 24 hours to correct hardware parameters such as GPS base station drift, IMU temperature deviation, and lens focal length change. A primary and backup system is deployed, and when the primary system fails, the backup system takes over within 2 seconds to maintain the continuous service of the enhancement information source. The system version EnhSrc_V2.3, parameter snapshot, performance benchmark, and other technical archives are recorded. The enhancement information source is encapsulated through REST API, supporting standard calling interfaces such as GET query state and POST update parameters.

[0048] The enhanced information sources are arranged in a grid to form a distributed grid. The enhanced information sources are arranged in a grid according to spatial positions and functional characteristics to establish a regular grid topology. The regular grid arrangement ensures the uniformity of the connection between the grid units and the completeness of the coverage. The size and resolution of the grid are determined according to the requirements of the perception task to balance the relationship between the coverage range and the computational complexity. The principle of nearby connection is adopted to reduce the communication overhead, and a reasonable connection topology is designed through connection optimization. When some nodes fail, the grid structure is automatically adjusted, and the adjacent normal nodes take over the functional area of the failed nodes. The grid nodes are usually distributed along a specific path to form a chain or tree-shaped grid structure to ensure effective coverage of the target area. A task scheduling algorithm is used to allocate computing tasks, and when the node load is too high, the new task is forwarded to the adjacent node with lower load.

[0049] A weight coding is implemented for the distributed grid to generate a collaborative enhancement matrix. Each node in the distributed grid is assigned a weight coefficient, and the weight reflects the importance and contribution of the node in the collaborative enhancement. The node weight W = a x information value + b x network status is calculated, where W is the node weight, a and b are weight coefficients with values of 0.6 and 0.4, respectively, the information value reflects the importance of the node in processing information, and the network status reflects the importance of the node in the network. A weight coding matrix is established, and the matrix elements represent the influence weight and collaborative relationship between nodes. Global adjustment of the weight is achieved through matrix operation to make the collaborative enhancement effect of the entire grid reach the best state. Sparse matrix storage technology is used to reduce the storage space and computational complexity of the weight matrix. When the node response time exceeds the preset threshold, the weight of the node is appropriately reduced. The weight of the node that has been working stably for a long time is appropriately increased. The node weights in the entire network are recalculated and distributed periodically. A large weight matrix is processed through distributed computing, and the results are collected after parallel computing by each node.

[0050] In step S140, a perception distribution reconstruction is performed based on the collaborative enhancement matrix to generate a reconstructed perception map. The reconstructed perception map is matched with the environmental interference distribution to identify a breakthrough perception interval through threshold breakthrough. The perception reconstruction sequence is generated using the breakthrough perception interval.

[0051] Specifically, the reconstruction of the perception distribution is generated based on the cooperative enhancement matrix. The weight distribution information is extracted from the cooperative enhancement matrix, and the numerical characteristics and spatial distribution patterns of each element in the matrix are analyzed. The main enhancement direction and the secondary enhancement direction are identified through matrix analysis, and the main component information of the matrix is extracted. The matrix weights are reorganized and distributed to extract the main enhancement direction and intensity information. The reconstructed weights are mapped to the two-dimensional coordinate system of the UAV perception space, and each weight value corresponds to the perception intensity of a position in the space. The discrete weight points are expanded to a continuous perception intensity distribution through spatial interpolation technology, and a smooth reconstruction result is generated. The reconstruction of the perception distribution converts the abstract mathematical representation of the cooperative enhancement matrix into an intuitive spatial perception capability distribution map, which shows the strength and weakness of the perception capability of the UAV at different positions such as the commercial district, the edge of the industrial park, and the interior of the residential area. The reconstructed perception distribution is filtered to remove high-frequency noise and outliers, and the main features of the perception distribution are retained. Through morphological processing method, the perception distribution is structured, the connected region and isolated region are identified, and the complete reconstructed perception map is formed.

[0052] In some embodiments, the threshold breakthrough matching of the reconstructed perception map and the environmental interference distribution to identify the breakthrough perception interval comprises: implementing hierarchical decomposition of the reconstructed perception map to obtain a perception density layer; performing cross-correlation analysis on the perception density layer and the environmental interference distribution to generate a correlation response map; based on the correlation response map, implementing peak aggregation identification to form an aggregation node array; and determining the breakthrough perception interval through connectivity reconstruction of the aggregation node array.

[0053] The reconstructed perception map is implemented hierarchical decomposition to obtain a perception density layer. A multi-scale decomposition technique is used on the reconstructed perception map to decompose the perception intensity distribution into a hierarchical structure of different spatial frequencies. The perception map is decomposed into a coarse scale layer and multiple fine scale layers using a hierarchical decomposition method, each level corresponding to different spatial resolution characteristics. The perception density distribution of each level is calculated, and the perception density reflects the concentration degree of perception intensity in unit area. The density estimation method is used to calculate the perception density distribution and determine the local density value. Each density layer is divided into foreground and background regions through threshold segmentation, and the foreground region corresponds to the high-density perception region. The perception density layer processes the complex perception distribution characteristics according to different levels of detail, and the coarse scale layer shows the large-scale perception distribution trend, and the fine scale layer shows the local perception detail changes. The change rule and correlation of the perception density on different scale levels are analyzed. Each density layer is labeled and coded to form a hierarchical density layer management system. Through hierarchical decomposition and separation processing, the complex perception distribution characteristics are decomposed into multiple density layers that are easy to process.

[0054] The cross-correlation analysis of the perception density layer and the environmental interference distribution generates a correlation response map. The perception density layer and the environmental interference distribution are spatially aligned to ensure that the correlation analysis is performed in the same coordinate system. The correlation analysis method is used to calculate the correlation between the density layer and the interference distribution, identifying matching and non-matching regions in space. The local correlation coefficient is calculated using a sliding window technique over the entire spatial range, generating a spatial distribution map of the correlation coefficient. The correlation response of the perception density and the interference distribution is calculated through convolution operations, identifying matching regions. The extreme points and gradient changes in the correlation response map are analyzed to identify regions with the strongest and weakest correlations. The cross-correlation analysis compares the spatial matching degree of the perception ability distribution and the environmental interference distribution, identifying areas with strong perception ability and low interference. The correlation response values are divided into three levels of strong, medium, and weak correlation according to their intensity. The threshold boundaries of correlation intensity are determined through statistical analysis, and the optimal classification threshold is determined using clustering analysis. The correlation response map is smoothed to remove false correlation peaks caused by noise.

[0055] Based on the correlation response map, peak aggregation identification is performed to form an array of aggregation nodes. Local maximum points are searched in the correlation response map, which correspond to the positions with the strongest correlation between perception density and environmental interference distribution. Peak detection techniques are used to remove secondary peaks among multiple adjacent peaks, leaving only the true extreme peaks. All significant peaks in the correlation response map are detected using a peak identification algorithm, and a peak height exceeding a preset threshold is considered an effective peak. When the distance between adjacent peaks is less than a set threshold and the height difference is within the allowable range, multiple peaks are grouped into an aggregation group. Spatial clustering analysis is performed on the identified peaks, grouping peaks with similar positions into aggregation node groups. The center of gravity of each aggregation node group is calculated as the representative position of the aggregation node, and the center of gravity coordinates are calculated using a weighted average method. Peak aggregation identification finds the hottest area with the highest matching degree of perception and interference in the correlation response map. The spatial distribution pattern of aggregation nodes is analyzed to identify the adjacency relationship and connectivity characteristics between nodes. All aggregation nodes are arranged in an array structure according to their spatial position, forming an index and search mechanism for the node array.

[0056] The breakthrough perception interval is determined by connectivity reconstruction of the aggregated node array. Based on the spatial distribution of the aggregated node array, the connectivity relationship between nodes is analyzed using graph theory. Each aggregated node is treated as a vertex of the graph, and the connection edge is set according to the spatial distance and correlation strength between nodes. The connection strength S=(1 / d)×c is calculated, where S is the connection strength, d is the distance between nodes, and c is the correlation strength. The closer the distance between nodes and the stronger the correlation, the higher the connection strength. The entire node array is decomposed into several independent connected subgraphs through connectivity component analysis, and each connected subgraph corresponds to a potential breakthrough perception interval. The geometric characteristics of each connected subgraph are analyzed, and the surrounding range and spatial distribution characteristics of the subgraph are calculated. Through quality evaluation, the connected subgraph with more nodes, high connectivity density, and appropriate spatial range is selected as the effective breakthrough perception interval. The breakthrough perception interval is the spatial region where the perception ability can overcome environmental interference and achieve effective work. In these regions, the sensor system of the unmanned aerial vehicle can obtain the best perception effect. The boundary extraction and region segmentation of the determined breakthrough perception interval are performed to generate an accurate interval boundary description. Each interval is assigned a unique identifier and attribute tag, and the spatial range, importance level, and related aggregated node information of the interval are recorded.

[0057] The perception reorganization sequence is generated using the breakthrough perception interval. According to the spatial distribution and importance ranking of the breakthrough perception interval, a priority strategy for perception reorganization is designed. The importance ranking considers the area size, correlation strength, and spatial position of the interval, with larger area and higher correlation strength intervals having higher priority. The spatial adjacency relationship and functional association between various breakthrough intervals are analyzed to identify interval combinations that can be processed cooperatively. The timing arrangement of perception reorganization is configured, considering the flight path and time constraints of the unmanned aerial vehicle, and the execution order of reorganization operations is reasonably arranged. The reorganization sequence R=[r1, r2,..., rn] is constructed, where ri represents the ith reorganization operation, including operation type, target interval, execution time, and expected effect. The specific operation method of perception reorganization is designed, including sensor parameter adjustment, data processing strategy improvement, and fusion weight optimization. When the reorganization operation effect is lower than the expected threshold, the operation parameters are automatically adjusted or the operation is skipped. The overall effect is evaluated after completing 5 reorganization operations, and if necessary, the remaining operations are reordered. The perception reorganization sequence formulates a series of perception system adjustment operations based on the characteristics of the breakthrough perception interval, and different perception parameters are configured in different intervals to obtain the best perception effect. The perception reorganization sequence is converted into a format executable by the unmanned aerial vehicle system, including detailed operation steps and parameter settings. Through the execution of the reorganization sequence, the unmanned aerial vehicle perception system realizes self-adaptive adjustment and optimization under different environmental conditions.

[0058] Step S150, intersect superposition analysis of the perception reorganization sequence and the target demand data is performed to determine a perception intersection period, a perception compression factor is obtained based on the perception intersection period and the perception feature parameter, and a perception control instruction set is constructed according to the perception compression factor.

[0059] Specifically, intersect superposition analysis of the perception reorganization sequence and the target demand data is performed to determine a perception intersection period. Time information and space information are extracted from the perception reorganization sequence, and the time distribution characteristics and execution priority of the reorganization operation are analyzed. The target demand data of the unmanned aerial vehicle is unfolded according to a time axis, including flight task time nodes, key operation time points, and performance requirement periods. The perception reorganization sequence and the target demand data are mapped into a unified time coordinate system through time alignment. The intersection relationship of the two time sequences is analyzed, and the time period in which the perception reorganization demand and the target task demand exist simultaneously is identified. The perception intersection period refers to the part in which the time period in which the perception system needs to be adjusted and optimized overlaps with the time period in which the unmanned aerial vehicle executes a key task, such as the time window in which the unmanned aerial vehicle needs to adjust the image sensor gain and optimize the data fusion parameter while executing a precise hovering shooting task. The intersection period is sorted according to importance, and factors such as task importance, time urgency, and resource constraints are considered. The duration and interval distribution of the intersection period are analyzed, and the key intersection period and the secondary intersection period are identified. The identification result of the intersection period is adjusted in real time considering the dynamic changes in the flight process of the unmanned aerial vehicle. The determined perception intersection period is arranged in time sequence to form a complete intersection period sequence.

[0060] In some embodiments, the obtaining of the perception compression factor based on the perception intersection period and the perception feature parameter comprises: constructing a period density cloud map according to the perception intersection period; identifying a density aggregation center in the period density cloud map; performing compression feasibility analysis on the density aggregation center to obtain a compression capacity; and fusing the compression capacity and the perception feature parameter to form the perception compression factor.

[0061] The time period density cloud map is constructed according to the perception intersection time period. The distribution of the perception intersection time period on the time axis is converted into a density distribution function, and the intersection time period is weighted based on the importance priority. The density distribution of the time period in the time-intensity two-dimensional space is calculated by using the density estimation method. The time axis represents the time of the intersection time period, and the intensity axis represents the importance level of the time period. The density distribution is visualized as a cloud map. Different colors and density levels represent different density levels. The time period density cloud map is similar to the precipitation probability distribution map in weather forecast, which shows the concentration of perception adjustment demand on the time axis. The high-density area represents that the perception adjustment operation in the time period is more frequent and important. The cloud map is smoothed to remove the density fluctuations and noise caused by sparse data. The global distribution characteristics of the cloud map are analyzed to identify the main mode and the secondary mode of the density distribution. The characteristic parameters of the cloud map are calculated by statistical analysis, including the density peak value, the distribution range and other key indicators. The entire time period range is divided into several sub-regions, each corresponding to different density characteristics and processing strategies.

[0062] The density aggregation center is identified in the time period density cloud map. The peak value detection method is used to search for local maximum points in the time period density cloud map, which correspond to the center positions of the density aggregation. The precise position and stability of the peak value are determined, and the density threshold is set to filter the low-density aggregation center. When the density value exceeds the average density and is a local maximum value, it is determined as a density aggregation center. The identified aggregation centers are classified and processed, and the aggregation centers with similar positions are merged into an aggregation center group. The influence range of each aggregation center is calculated, and the action radius of the aggregation center is determined by density level analysis. The time stability of the aggregation center is analyzed, and the position change and intensity change of the aggregation center in the time evolution process are tracked. The density aggregation center represents the time point with the most concentrated perception adjustment demand, which usually corresponds to the key moment when multiple perception parameters need to be adjusted simultaneously during the execution of complex tasks by the unmanned aerial vehicle, such as the take-off climbing stage requiring simultaneous adjustment of height sensor, GPS accuracy and visual recognition parameters. The importance of the aggregation center is sorted, and the importance is based on the comprehensive evaluation of the density peak height, the influence range size and the time stability.

[0063] Compression feasibility analysis on density aggregation centers generates compression capacity. Based on the identified density aggregation centers, the feasibility and effect of perceptual data compression near the center are analyzed. The data characteristics of the aggregation center are evaluated, including the correlation, redundancy and information amount of the data, which directly affect the effect of compression. The frequency characteristics of the data near the aggregation center are analyzed to identify the components that can be lossy compressed and lossless compressed. The compression feasibility evaluation considers multiple dimensions such as compression ratio, distortion, computational complexity and real-time requirements. The compression capacity P = a x efficiency + b x quality is calculated, where a and b are weight coefficients with values of 0.6 and 0.4, efficiency reflects compression speed and resource consumption, and quality reflects data fidelity after compression. The maximum compression capacity of different aggregation centers is determined through theoretical analysis and experimental verification, forming a relationship model between capacity and center characteristics. The impact of the compression process on the accuracy of perception is analyzed, and the relationship between compression benefits and system performance is balanced. Considering the hardware resource limitations, including processor performance, storage space and transmission bandwidth, the actual achievable compression capacity is determined. The compression capacity is dynamically evaluated according to the real-time system load and performance requirements.

[0064] The compression capacity and perceptual feature parameters are fused to form a perceptual compression factor. Compression capacity reflects the technical compression ability of each region, and perceptual feature parameters reflect the importance of perceptual data. The fusion of the two ensures that the system neither over-compresses in key perception areas affecting accuracy, nor wastes resources in non-important areas. The compression capacity of each density aggregation center is normalized to map the capacity values of different centers to a unified numerical range. The core components of the perceptual feature parameters are extracted, including the spectral center frequency, power distribution characteristics and signal stability indicators, to obtain a representative feature vector. The normalized compression capacity is used as the basic weight and the feature vector components are used as adjustment coefficients using a weighted linear combination method. The fusion weight is dynamically allocated according to the importance level and spatial distribution characteristics of the density aggregation center. The correlation between compression capacity and perceptual feature parameters is analyzed, and when they show strong positive correlation, the fusion effect is enhanced, and when negative correlation occurs, compensation adjustment is made. The perceptual compression factor directly controls the system's data transmission strategy, sensor power allocation and communication bandwidth scheduling. High compression factor regions use data compression and power reduction strategies, while low compression factor key perception areas maintain high-precision data transmission. The compression factor values of multiple aggregation centers are generated to form a complete perceptual compression factor distribution.

[0065] The perception control instruction set is constructed according to the perception compression factor. Based on the calculated perception compression factor, the control strategy and parameter configuration scheme of the UAV perception system are designed. The compression factor is converted into specific control parameters, including sensor sampling frequency, data processing accuracy and communication bandwidth allocation. A hierarchical structure of the control instruction is constructed, which is divided into three levels of system-level instruction, module-level instruction and device-level instruction. The optimal instruction combination is determined through instruction optimization to minimize system resource consumption under the premise of ensuring perception performance. The system-level instruction controls the overall perception strategy, such as switching between high-precision mode, energy-saving mode and emergency mode, the module-level instruction controls the working parameters of each sensor module, and the device-level instruction controls the power and frequency settings of the specific hardware. The dependency relationship and execution order between instructions are set to ensure the coordinated execution of instructions and the stable operation of the system. A fault-tolerant mechanism for the instruction is designed, which can automatically switch to an alternative instruction when part of the instruction fails. The constructed perception control instruction set is converted into a standard format recognizable by the UAV system, which contains detailed execution parameters and operation procedures. The perception control instruction set provides a dynamic configuration and resource optimization control framework for the UAV autonomous perception system, achieving the best balance between perception performance and resource consumption.

[0066] In step S160, a fluctuation suppression analysis is performed on the perception control instruction set to identify an optimal suppression path, and a fluctuation correction is performed on the perception reorganization sequence along the optimal suppression path to generate an environment perception enhancement instruction.

[0067] In some embodiments, the fluctuation suppression analysis on the perception control instruction set to identify the optimal suppression path includes: detecting fluctuation amplitude characteristics and fluctuation frequency characteristics through the perception control instruction set; constructing a fluctuation suppression matrix according to the fluctuation amplitude characteristics; performing frequency domain suppression processing on the fluctuation suppression matrix based on the fluctuation frequency characteristics to generate a suppression response path group; and selecting a path with the optimal suppression effect from the suppression response path group to determine the optimal suppression path.

[0068] The fluctuation amplitude feature and the fluctuation frequency feature are detected by the perception control instruction set. The parameters of each instruction in the perception control instruction set are continuously monitored, and the time-varying data of the parameters are obtained by using the sampling technology. The amplitude distribution characteristics of the instruction parameters are calculated by statistical analysis, including the maximum amplitude, the average amplitude, and the amplitude variation range and other key indicators. The envelope curve of the fluctuation signal is extracted by using envelope detection, and the variation trend and fluctuation characteristics of the amplitude are analyzed. The amplitude features are organized into a feature vector, including the peak value, the effective value, and the standard deviation and other parameters. The frequency characteristics of the fluctuation signal are identified by frequency spectrum analysis, and the power contribution and relative importance of each frequency component are calculated. The fluctuation amplitude feature reflects the intensity of the fluctuation of the perception parameters, and the fluctuation frequency feature reflects the fast and slow rhythm of the fluctuation. When the unmanned aerial vehicle flies in a strong wind environment, the attitude control instruction will have high-frequency small-amplitude fluctuations, and when it flies in complex terrain, the obstacle avoidance instruction will have low-frequency large-amplitude fluctuations. The frequency components are grouped according to low frequency, medium frequency, and high frequency, and the low-frequency component corresponds to the slow dynamic of the system, and the high-frequency component corresponds to the rapid disturbance. The main peak frequencies in the power spectrum are identified, and these frequencies correspond to the characteristic oscillation modes of the system. The coupling relationship between the amplitude and frequency characteristics is analyzed, and the dominant oscillation mode and the secondary oscillation mode are identified.

[0069] A fluctuation suppression matrix is constructed according to the fluctuation amplitude feature. Based on the detected fluctuation amplitude feature, a corresponding suppression strategy and control parameter are designed. The suppression matrix is constructed by considering the amplitude feature and the system stability requirement, and the matrix elements reflect the suppression intensity corresponding to different amplitude levels. Different suppression intensities are allocated to fluctuations of different amplitude levels, and strong suppression strategies are used for fluctuations with large amplitudes, and weak suppression strategies are used for fluctuations with small amplitudes. The corresponding relationship between the amplitude level and the suppression coefficient is determined by table lookup, and the table is formulated based on historical data and experimental results. An adaptive algorithm is used to dynamically adjust the suppression coefficient according to the real-time amplitude variation, to ensure the time-varying optimization of the suppression effect. The suppression matrix realizes the collaborative suppression control of multiple parameters, and simultaneously processes the mutual influence of multiple fluctuation components. The stability of the matrix is analyzed to ensure the stable operation of the suppression system, and to avoid system oscillation caused by excessive suppression. The fluctuation suppression matrix converts the abstract amplitude feature into specific suppression operations, and provides differentiated processing strategies for perception disturbances of different intensities.

[0070] The fluctuation suppression matrix is converted to the frequency domain for processing, and the frequency selectivity of the suppression strategy is optimized through frequency domain analysis. A filter design method is used to design dedicated suppression filters for different frequency components, and the filter parameters are optimized and determined according to the frequency characteristics. Low-pass filters are designed to suppress high-frequency disturbances, and band-stop filters are designed to suppress disturbances in specific frequency bands, and multiple frequency band suppression is achieved through filter combination. A closed-loop suppression system is designed through the frequency domain suppression transfer function to ensure system stability and dynamic performance. The frequency response characteristics of the suppression system are analyzed to obtain suppression effect and dynamic performance data. A path modeling method is used to construct the suppression response path, with nodes representing frequency components and connections representing suppression relationships. All possible combinations of suppression paths are generated by traversing the suppression response graph through a path search algorithm. The performance parameters of each path, such as suppression delay, suppression strength, and frequency coverage, are calculated. Similar paths are grouped according to their characteristics, and each group contains multiple alternative paths.

[0071] The optimal suppression path is determined by selecting the path with the optimal suppression effect from the suppression response path group. The path performance is evaluated in four dimensions: suppression depth, response speed, frequency selectivity, and stability. The path score E is calculated as E = w1 x suppression depth + w2 x response speed + w3 x frequency selectivity + w4 x stability, where E is the comprehensive score of the path, and w1, w2, w3, w4 are weight coefficients with values of 0.4, 0.3, 0.2, and 0.1, respectively. The comprehensive score of each path in the suppression response path group is calculated, and the weighted average comprehensive performance index is calculated. The path with the highest score is selected as the candidate optimal path through performance comparison. The path selection considers constraints such as processor occupancy, response time, and memory usage to ensure that the selected path is feasible within the system resource limits. When the system load or environmental disturbance intensity changes significantly, the system automatically recalculates and switches to a more suitable suppression path. The control parameters of the optimal suppression path are converted into system executable control instructions to achieve seamless integration from path selection to specific control.

[0072] Fluctuations in the perception recombined sequence are corrected along the optimal suppression path to generate environmental perception enhancement commands. Control nodes in the suppression path are mapped to corresponding positions in the perception recombined sequence, ensuring that corrections are synchronized with the UAV's flight status and mission requirements. The suppression strategy and control parameters contained in the optimal suppression path guide the specific correction operations, and the suppression coefficients corresponding to the path nodes are directly applied to the sequence correction. Adaptive filtering technology is used to suppress fluctuations in the recombined sequence in real time, with filter parameters dynamically adjusted based on the UAV's flight status. A feedback control system continuously monitors and adjusts the correction results, specifically compensating for perceptual disturbances generated by maneuvers such as turns, climbs, and descents. Predictive control methods are used to predict future fluctuation trends and, combined with the UAV's trajectory planning information, preemptively perform corrections. A multi-level correction strategy is implemented to address fluctuations in different frequency ranges and intensity levels, achieving refined suppression of interference sources such as propeller vibration, engine noise, and airflow disturbances. The corrected perception recombined sequence is converted into environmental perception enhancement commands, which contain sensor configuration details such as camera stabilization parameters, lidar scanning frequency, infrared sensor gain adjustment, and data fusion weights. Corresponding command parameters are set for different flight phases, focusing on close-range perception accuracy during takeoff and landing, optimizing long-range target detection capabilities during cruising, and strengthening perception support for fine control during hovering. Through a full-process process that integrates multimodal sensor collaboration, perception bottleneck optimization, dynamic weight allocation, and fluctuation suppression correction, the drone achieves enhanced multimodal environmental perception in complex environments.

[0073] In order to implement the multimodal environment perception enhancement method in a complex environment corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The following is a block diagram of a multimodal environment perception enhancement system 200 for a complex environment provided by an embodiment of the present application. For ease of explanation, only the parts related to the present embodiment are shown. The multimodal environment perception enhancement system 200 for a complex environment provided by an embodiment of the present application includes: The data acquisition module 201 is configured to acquire multimodal environmental perception data, illumination monitoring data, and target demand data, perform boundary identification on the multimodal environmental perception data to extract perception critical points, extract environmental interference distribution from the illumination monitoring data, and perform intensity coupling between the perception critical points and the environmental interference distribution to generate a perception basic field. an interference analysis module 202 for identifying a perception bottleneck point based on the perception basic field, performing interference impedance analysis on the perception bottleneck point to generate an equivalent interference network, extracting perception feature parameters from the equivalent interference network, and constructing a perception response path diagram based on the equivalent interference network and the perception feature parameters; The synergistic enhancement module 203 is configured to perform perceptual signal shunt analysis on the perceptual response path graph to extract a primary channel and a secondary channel, analyze signal intensity distribution of the primary channel and the secondary channel to identify an optimal fusion point, perform dynamic weight distribution on the optimal fusion point to generate a synergistic enhancement matrix; The perceptual reconstruction module 204 is configured to perform perceptual distribution reconstruction based on the synergistic enhancement matrix to generate a reconstructed perceptual graph, perform threshold breakthrough matching between the reconstructed perceptual graph and the environmental interference distribution to identify a breakthrough perceptual interval, and generate a perceptual reorganization sequence by using the breakthrough perceptual interval. The compression control module 205 is configured to perform intersection superposition analysis between the perceptual reorganization sequence and the target demand data to determine a perceptual intersection time period, obtain a perceptual compression factor based on the perceptual intersection time period and the perceptual characteristic parameter, and construct a perceptual control instruction set according to the perceptual compression factor. The fluctuation suppression module 206 is configured to perform fluctuation suppression analysis on the perceptual control instruction set to identify an optimal suppression path, and perform fluctuation correction on the perceptual reorganization sequence along the optimal suppression path to generate an environmental perceptual enhancement instruction.

[0074] The above-described multi-modal environmental perceptual enhancement system 200 in a complex environment can implement a multi-modal environmental perceptual enhancement method in a complex environment according to the above-described method embodiments. The optional items in the above-described method embodiments are also applicable to this embodiment, and will not be described in detail herein. The remaining content of the present embodiment can refer to the content of the above-described method embodiments, and will not be described in detail herein.

[0075] The above only describes some or preferred embodiments of the present application, and neither the text nor the drawings can limit the scope of protection of the present application. Any equivalent structural transformation based on the content of the present application and the drawings, or direct / indirect application in other related technical fields is included in the scope of protection of the present application.

Claims

1. A multi-modal environment perception enhancement method in a complex environment, characterized in that, The method comprises the following steps: acquiring multi-modal environment perception data, illumination monitoring data and target demand data, performing boundary recognition on the multi-modal environment perception data to extract a perception critical point, extracting an environment disturbance distribution from the illumination monitoring data, coupling the perception critical point and the environment disturbance distribution to generate a perception base field; identifying a perception bottleneck point based on the perception base field, performing interference impedance analysis on the perception bottleneck point to generate an equivalent interference network, extracting a perception feature parameter from the equivalent interference network, and constructing a perception response path graph based on the equivalent interference network and the perception feature parameter; performing perception signal shunt analysis on the perception response path graph to extract a primary and secondary channel, analyzing the signal intensity distribution of the primary and secondary channel to identify an optimal fusion point, and performing dynamic weight distribution on the optimal fusion point to generate a synergistic enhancement matrix; performing perception distribution reconstruction based on the synergistic enhancement matrix to generate a reconstructed perception graph, performing threshold breakthrough matching on the reconstructed perception graph and the environment disturbance distribution to identify a breakthrough perception interval, and generating a perception reorganization sequence using the breakthrough perception interval; performing intersection superposition analysis on the perception reorganization sequence and the target demand data to determine a perception intersection period, acquiring a perception compression factor based on the perception intersection period and the perception feature parameter, and constructing a perception control instruction set according to the perception compression factor; performing fluctuation suppression analysis on the perception control instruction set to identify an optimal suppression path, and performing fluctuation correction on the perception reorganization sequence along the optimal suppression path to generate an environment perception enhancement instruction.

2. The method of claim 1, wherein, The method of coupling the perception critical point and the environment disturbance distribution to generate a perception base field comprises: identifying an intensity change rate sequence at the perception critical point; cross-verification positioning a verification anchor point by comparing the intensity change rate sequence and the environment disturbance distribution; deriving a coupling coefficient along the verification anchor point; constructing a perception base field according to the distribution range of the coupling coefficient.

3. The method of claim 1, wherein, The method of extracting a perception feature parameter from the equivalent interference network comprises: analyzing the node connection mode of the equivalent interference network to identify a primary connection and a secondary connection; calculating a connection intensity distribution according to the primary connection and the secondary connection; deriving a feature frequency spectrum through the connection intensity distribution; extracting a core component from the feature frequency spectrum as a perception feature parameter.

4. The method of claim 1, wherein, The method of performing dynamic weight distribution on the optimal fusion point to generate a synergistic enhancement matrix comprises: tracking and analyzing information flow according to the optimal fusion point to identify an upstream information source and a downstream information source; forming an enhanced information source by using the upstream information source to implement information feedback enhancement on the downstream information source; generating a distributed grid by grid layout of the enhanced information source; generating a synergistic enhancement matrix by weight coding of the distributed grid.

5. The method of claim 1, wherein, The method of performing threshold breakthrough matching on the reconstructed perception graph and the environment disturbance distribution to identify a breakthrough perception interval comprises: performing hierarchical decomposition and stripping on the reconstructed perception graph to obtain a perception density layer; performing cross-correlation analysis on the perception density layer and the environment disturbance distribution to generate a correlation response graph; forming an aggregation node array based on the correlation response graph. The connectivity reconstruction by the array of aggregation nodes determines a breakthrough perception interval.

6. The method of claim 1, wherein, The perception compression factor is obtained based on the perception intersection period and the perception characteristic parameter. A period density cloud map is constructed according to the perception intersection period. A density aggregation center is identified in the period density cloud map. Compression capacity is generated by performing compression feasibility analysis on the density aggregation center. A perception compression factor is formed by fusing the compression capacity and the perception characteristic parameter.

7. The method of claim 1, wherein, The fluctuation suppression analysis on the set of perception control instructions identifies an optimal suppression path, including: Fluctuation amplitude characteristics and fluctuation frequency characteristics are detected by the set of perception control instructions. A fluctuation suppression matrix is constructed according to the fluctuation amplitude characteristics. An inhibition response path group is generated by performing frequency domain suppression processing on the fluctuation suppression matrix based on the fluctuation frequency characteristics. The optimal suppression path is determined according to the inhibition response path group with the optimal suppression effect.

8. The method of claim 2, wherein, The coupling coefficient is derived along the verification anchor point, including: Key sampling points are set around the verification anchor point. Local intensity variation rate and impedance variation rate are obtained from the key sampling points. A variation rate curve is constructed according to the intensity variation rate and the impedance variation rate. The evolution trend of the variation rate curve is converted into a coupling coefficient.

9. The method of claim 4, wherein, The upstream information source is used to perform information feedback enhancement on the downstream information source to form an enhanced information source, including: The upstream information source is used as an information provider to transmit enhanced information to the downstream information source; The enhanced information is subjected to cyclic control to form a stable enhancement state; The information enhancement accuracy is continuously optimized based on the stable enhancement state; When the enhancement accuracy meets the requirements, the stable enhancement state is maintained to form an enhanced information source.

10. A multi-modal environment perception enhancement system in complex environments, characterized in that, It includes: A data acquisition module is configured to acquire multi-modal environmental perception data, illumination monitoring data, and target demand data, identify boundaries of the multi-modal environmental perception data to extract perception critical points, extract environmental disturbance distribution from the illumination monitoring data, and perform intensity coupling between the perception critical points and the environmental disturbance distribution to generate a perception base field; An interference analysis module is configured to identify perception bottleneck points based on the perception base field, generate an equivalent interference network by performing interference impedance analysis on the perception bottleneck points, extract perception characteristic parameters from the equivalent interference network, and construct a perception response path graph based on the equivalent interference network and the perception characteristic parameters; A cooperative enhancement module is configured to perform perception signal shunt analysis on the perception response path graph to extract primary and secondary channels, analyze signal intensity distribution of the primary and secondary channels to identify an optimal fusion point, and generate a cooperative enhancement matrix by performing dynamic weight allocation on the optimal fusion point; A perception reconstruction module is configured to perform perception distribution reconstruction based on the cooperative enhancement matrix to generate a reconstructed perception graph, perform threshold breakthrough matching between the reconstructed perception graph and the environmental disturbance distribution to identify a breakthrough perception interval, and generate a perception reorganization sequence using the breakthrough perception interval. The compression control module is configured to: perform intersection overlay analysis on the perception reorganization sequence and the target demand data to determine a perception intersection period; acquire a perception compression factor based on the perception intersection period and the perception characteristic parameter; and construct a perception control instruction set according to the perception compression factor. The fluctuation suppression module is configured to: perform fluctuation suppression analysis on the perception control instruction set to identify an optimal suppression path; and perform fluctuation correction on the perception reorganization sequence along the optimal suppression path to generate an environmental perception enhancement instruction.

Images