Intelligent pod control method and system for unmanned aerial vehicle

Abstract

The invention provides an intelligent pod control method and system for an unmanned aerial vehicle. According to the method, multiple frames of thermal radiation images are obtained through an infrared imaging module of an unmanned aerial vehicle pod, a dynamic temperature field atlas is generated through heat source decoupling, and spatial relevance between heat sources is analyzed through an atlas neural network. Damping torque matched with wind disturbance is generated based on the temperature gradient vector, filtering parameters of the magnetofluid holder are optimized, and meanwhile the interference intensity between energy transmission paths is calibrated. A threat propagation path is generated by combining thermal diffusion characteristics, and the pod deflection angle is dynamically adjusted, so that the center of an optical view field is synchronized with the threat centroid, and accurate tracking and anti-interference control are realized. The method is used for intelligent pod control of the unmanned aerial vehicle. Through intelligent heat source analysis and adaptive adjustment, accurate and stable tracking of a threat target by the unmanned aerial vehicle pod is realized.

Description

Technical Field

[0001] This application relates to the technical field of intelligent pod control, and particularly to an intelligent pod control method and system for unmanned aerial vehicles (UAVs). Background Art

[0002] In the night security scenario, the monitoring of illegal UAV cluster intrusion faces complex challenges: it is necessary to quickly identify multi-heat-source targets in low-illumination environments, overcome environmental interferences such as wind disturbances, and track dynamic threats. Traditional optoelectronic means are difficult to effectively distinguish dense heat sources from background noise, requiring the intelligent pod to have high-precision thermal radiation analysis capabilities, adaptive anti-interference control, and cluster target trajectory prediction functions.

[0003] The current mainstream solution adopts the "infrared imaging + deep learning" technology, extracts the features of thermal images through a convolutional neural network, combines the Kalman filter to predict the target movement trajectory, and suppresses the jitter of the pan-tilt head based on fixed-threshold filtering. For example, a certain patent proposes to use the YOLOv5 model to detect heat source targets and adjust the pod's field of view tracking through PID control.

[0004] This solution has insufficient handling of the interference of energy transfer between heat sources and the coupling effect of dynamic wind disturbances: First, it is difficult for CNN to model the complex spatial correlation between heat sources, resulting in a high false detection rate for dense targets; second, fixed-parameter filtering cannot adapt to the spectral changes of wind disturbances, and the pan-tilt head response lags; third, the impact of the heat diffusion path on threat propagation is not considered, and the trajectory prediction deviation is significant. Summary of the Invention

[0005] This application provides an intelligent pod control method and system for UAVs to solve the problem of poor control effect of intelligent pods in the prior art.

[0006] In a first aspect, this application provides an intelligent pod control method for UAVs, including: Obtaining multiple frames of thermal radiation images of the UAV through an infrared imaging module in the intelligent pod, performing heat source decoupling processing on the multiple frames of thermal radiation images, and generating a dynamic temperature field map for each heat source; Analyzing the spatial correlation between heat sources by using a graph neural network based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources; Generating a damping torque that matches the wind disturbance spectrum according to the spatial correlation between heat sources, and correcting the filtering parameters of the magnetorheological damping pan-tilt head in the intelligent pod according to the frequency response characteristics of the damping torque; Mapping the temperature gradient vector to a corresponding energy transfer path, and calibrating the interference intensity between adjacent two energy transfer paths by using the filtering parameters; Extracting heat diffusion features from the temperature gradient vector, and generating a threat propagation path according to the calibrated interference intensity and the heat diffusion features; According to the threat propagation path, dynamically plan the deflection angle of the intelligent pod so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

[0007] Optionally, extract the heat diffusion characteristics from the temperature gradient vector, and generate a threat propagation path according to the calibrated interference intensity and the heat diffusion characteristics, including: Perform time series decomposition on the temperature gradient vectors in the dynamic temperature field map of each heat source, extract multiple characteristics such as the change trajectory, fluctuation pattern, and cumulative change amount of the temperature gradient direction at consecutive time points, and respectively form a direction change sequence, an intensity fluctuation sequence, and a heat diffusion energy value; among them, according to the change trajectory of the temperature gradient direction at consecutive time points, form a direction change sequence; according to the fluctuation pattern of the temperature gradient magnitude at consecutive time points, form an intensity fluctuation sequence; according to the cumulative change amount of the temperature gradient vector within a preset time window, form a heat diffusion energy value; Combine the direction change sequence, the intensity fluctuation sequence, and the heat diffusion energy value into a heat diffusion feature vector for each heat source; According to the calibrated interference intensity, construct an interference intensity table between heat sources, where each table entry stores the interference intensity value between a pair of heat sources, and calculate the feature similarity of the heat diffusion feature vectors of any two heat sources, and the similarity is determined by fusing the trajectory repeatability and fluctuation consistency of the any two heat sources; Multiply the feature similarity by the interference intensity value of the corresponding heat source pair in the interference intensity table to obtain the propagation correlation weight between heat sources; Taking the heat source with the largest heat diffusion energy value as the starting node, from the unvisited heat sources, select the adjacent heat source with the highest propagation correlation weight with the current heat source, add the selected heat source to the path sequence, and update it as the current heat source. Repeat the above steps until the propagation correlation weight is lower than the preset threshold or there are no selectable adjacent heat sources, and output the path sequence as the threat propagation path.

[0008] Optionally, according to the threat propagation path, dynamically plan the deflection angle of the intelligent pod so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path, including: Obtain the spatial coordinates of all heat sources in the threat propagation path to form a coordinate set, and calculate the geometric centroid of the coordinate set, where the geometric centroid includes the centroid coordinates of multiple components; Obtain the current pointing angle of the intelligent pod, where the current pointing angle includes the horizontal yaw angle and the vertical pitch angle; Convert the centroid coordinates to the pod coordinate system through coordinate transformation, and calculate the horizontal yaw angle correction amount and the vertical pitch angle correction amount according to the converted centroid coordinates and the spatial relative position of the intelligent pod, and use them as the target pointing angle; Decompose the difference between the current pointing angle and the target pointing angle into angular velocity components, perform kinematic constraints on the angular velocity components according to the inertial parameters of the magneto - rheological damping gimbal, output a continuous angle change sequence that meets the constraints, and control the intelligent pod to deflect according to the continuous angle change sequence so that the optical field of view center aligns with the geometric centroid coordinates.

[0009] Optionally, obtain multiple frames of thermal radiation images of the unmanned aerial vehicle through the infrared imaging module in the intelligent pod, perform heat source decoupling processing on the multiple frames of thermal radiation images, and generate a dynamic temperature field map of each heat source, including: Capture multiple frames of thermal radiation images of the unmanned aerial vehicle at consecutive time points through the infrared imaging module of the intelligent pod, and each frame of thermal radiation image contains temperature values of multiple pixel points; Based on the radiation intensity distribution and spatial position stability of the multiple frames of thermal radiation images, track the heat source contour of each independent heat source in consecutive frames; Use the tracked heat source contour to determine the radiation contribution of each independent heat source in the multiple frames of thermal radiation images; According to the radiation contribution, assign a unique identifier to each heat source and extract the temperature distribution data of each heat source in each frame of image; integrate the temperature distribution data of each heat source in time series to generate a dynamic temperature field map of this heat source.

[0010] Optionally, based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources, use a graph neural network to analyze the spatial correlation between heat sources, including: Extract the temperature gradient vectors from the dynamic temperature field maps of each heat source; Input the temperature gradient vectors as node features into the graph neural network, and the graph neural network learns the dependence relationships between node features; Through iterative calculations of the graph neural network, output the association weights between nodes, and the association weights represent the degree of mutual influence of temperature changes between heat sources; Generate a description of the spatial correlation between heat sources according to the association weights.

[0011] Optionally, generate a damping torque that matches the wind disturbance spectrum according to the spatial correlation between heat sources and correct the filtering parameters of the magneto - rheological damping gimbal in the intelligent pod according to the frequency response characteristics of the damping torque, including: Based on the spatial correlation between heat sources, construct a spatial relationship model, where the spatial relationship model represents the position and degree of mutual influence of each heat source as connection weights; Simulate the dynamic response of the wind disturbance spectrum at different frequencies through the spatial relationship model to convert the connection weights into torque generation factors, and calculate the torque amplitudes and phases at multiple frequency points according to the torque generation factors; Synthesize a damping torque sequence according to the torque amplitude and phase, wherein the spectral characteristics of the damping torque sequence are consistent with those of the wind disturbance spectrum; Compare the amplitude response and phase response in the frequency response characteristics of the extracted damping torque sequence with the current filtering parameters of the magnetorheological damping pan-tilt in the intelligent pod to generate a parameter deviation; Adjust the cut-off frequency and gain coefficient in the current filtering parameters according to the parameter deviation to generate corrected filtering parameters.

[0012] Optionally, map the temperature gradient vector to a corresponding energy transfer path, and use the filtering parameters to calibrate the interference intensity between two adjacent energy transfer paths, including: Use each vector element in the temperature gradient vector as a starting point, and perform path tracing along the vector direction to generate multiple energy transfer paths; Calculate the overlapping area between two adjacent energy transfer paths, and quantify the overlapping area as the initial interference intensity; Use the cut-off frequency and gain coefficient in the filtering parameters to scale and adjust the initial interference intensity to determine the calibrated interference intensity, wherein the process of the scaling adjustment includes: filtering out high-frequency interference components using the cut-off frequency, and amplifying or attenuating the amplitude of the initial interference intensity using the gain coefficient.

[0013] In a second aspect, the present application provides an intelligent pod control system for an unmanned aerial vehicle, including: A first generation module, configured to obtain multiple frames of thermal radiation images of the unmanned aerial vehicle through an infrared imaging module in the intelligent pod, perform heat source decoupling processing on the multiple frames of thermal radiation images, and generate a dynamic temperature field map of each heat source; An analysis module, configured to analyze the spatial correlation between heat sources by using a graph neural network based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources; A correction module, configured to generate a damping torque matching the wind disturbance spectrum according to the spatial correlation between heat sources, and correct the filtering parameters of the magnetorheological damping pan-tilt in the intelligent pod according to the frequency response characteristics of the damping torque; A calibration module, configured to map the temperature gradient vector to a corresponding energy transfer path, and calibrate the interference intensity between two adjacent energy transfer paths by using the filtering parameters; A second generation module, configured to extract heat diffusion features from the temperature gradient vector, and generate a threat propagation path according to the calibrated interference intensity and the heat diffusion features; A planning module, configured to dynamically plan the deflection angle of the intelligent pod according to the threat propagation path, so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

[0014] In the embodiment of the present application, multiple frames of thermal radiation images of the drone are obtained through an infrared imaging module in an intelligent pod, and the multiple frames of thermal radiation images are subjected to heat source decoupling processing to generate a dynamic temperature field map of each heat source; based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources, a graph neural network is used to analyze the spatial correlation between the heat sources; according to the spatial correlation between the heat sources, a damping torque matching the wind disturbance spectrum is generated, and the filtering parameters of a magneto-rheological damping gimbal in the intelligent pod are corrected according to the frequency response characteristics of the damping torque; the temperature gradient vectors are mapped to corresponding energy transfer paths, and the interference intensity between two adjacent energy transfer paths is calibrated by using the filtering parameters; thermal diffusion features are extracted from the temperature gradient vectors, and a threat propagation path is generated according to the calibrated interference intensity and the thermal diffusion features; according to the threat propagation path, the deflection angle of the intelligent pod is dynamically planned so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

[0015] The technical solution of the present application has the following beneficial effects: Independent heat sources are separated through multiple frames of thermal radiation images, and background noise interference is eliminated, providing a high-precision thermodynamic data basis for subsequent analysis. The interaction of temperature gradient vectors between heat sources is quantified, the problem of misdetection of dense targets is solved, and the recognition accuracy of cluster targets is improved. A damping torque with frequency response matching is adaptively generated to suppress the gimbal jitter caused by wind disturbance and enhance the dynamic stability. By optimizing the calculation of the interference intensity between paths through filtering parameters, the trajectory prediction deviation caused by the cross influence of thermal radiation is reduced. The thermal diffusion features are fused with the calibrated interference intensity to predict the threat diffusion trend and achieve active trajectory tracking. The optical field of view is adjusted to be synchronized with the threat centroid to ensure continuous target locking and improve the tracking response speed and accuracy.

[0016] Furthermore, the present application extracts the direction change, fluctuation mode, and energy accumulation characteristics of the temperature gradient vector through time series decomposition to construct a thermal diffusion feature vector; combines the interference intensity table with the feature similarity to calculate the propagation correlation weight, and generates a threat propagation path starting from the heat source with the maximum energy. Subsequently, the geometric centroid is calculated based on the coordinates of the heat sources on the path, and a gimbal deflection sequence is generated through coordinate transformation and kinematic constraints to achieve the spatial synchronization of the optical field of view and the centroid.

[0017] Through the fusion of thermal diffusion spatio-temporal features and dynamic calculation of propagation weights, the present application accurately predicts the evolution trend of the threat path; combines centroid tracking and kinematic constraint control to achieve rapid and stable alignment of the gimbal with the threat core, significantly improving the tracking robustness and anti-interference ability for cluster targets.

[0018] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0020] Figure 1 Shows a flowchart of an intelligent pod control method for an unmanned aerial vehicle provided by the present application; Figure 2 Shows a structural schematic diagram of an intelligent pod control system for an unmanned aerial vehicle provided by the present application; Figure 3 Shows a structural schematic diagram of a computing device provided by the present application. Detailed implementation manners

[0021] To enable those skilled in the art to better understand the solution of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application.

[0022] In some processes described in the specification, claims and the above accompanying drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The operation numbers such as 101, 102, etc. are only used to distinguish different operations, and the numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0023] In the night unmanned aerial vehicle cluster monitoring scenario, the existing "infrared imaging + deep learning" - based solution has significant limitations: First, traditional convolutional neural networks are difficult to effectively model the dynamic energy transfer relationship between heat sources, resulting in a high false detection rate for dense heat source targets; Second, the fixed - parameter filtering algorithm cannot adapt to the time - varying frequency spectrum characteristics of wind disturbances, causing response lag and tracking jitter of the magnetohydrodynamic damping gimbal; Third, the existing methods lack quantitative analysis of the heat diffusion path, resulting in systematic deviations in threat propagation trajectory prediction, seriously affecting the collaborative tracking effect of cluster targets. These defects essentially stem from the fragmented treatment of thermodynamic relevance, environmental disturbance coupling, and energy propagation mechanism in the existing technology.

[0024] In view of the above problems, the present invention proposes an intelligent pod control method based on dynamic coupling analysis of heat sources. By constructing a spatial correlation model of temperature gradient vectors using a graph neural network and combining an adaptive damping adjustment mechanism that matches the wind disturbance spectrum, high-precision decoupling and stable tracking of heat source targets are achieved. Specifically, first, multi-frame thermal radiation images are subjected to heat source decoupling to generate a dynamic temperature field map, and then the spatio-temporal characteristics of temperature gradient vectors are used to construct an energy transfer path and calibrate the interference intensity. Finally, a threat propagation path is generated and the deflection angle of the pod is dynamically planned. This method innovatively integrates thermodynamic characteristic analysis, environmental disturbance suppression, and target motion prediction: the problem of misdetection of dense heat sources is solved by a graph neural network, the instability of the pan-tilt caused by wind disturbance is overcome by correcting the frequency response characteristics of magneto-fluid damping, and the accurate reconstruction of the threat propagation path is achieved based on heat diffusion characteristics and interference intensity calibration. Compared with the prior art, this solution improves the tracking accuracy of cluster targets in a complex night environment and reduces the amplitude of field-of-view jitter caused by wind disturbance, significantly enhancing the reliability and adaptability of the UAV pod in the security monitoring scenario.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying 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 efforts belong to the scope of protection of the present application.

[0026] Figure 1 The following is a flowchart of an intelligent pod control method for a UAV provided in an embodiment of the present application, as Figure 1 shown, the method includes: 101. Obtain multi-frame thermal radiation images of the UAV through an infrared imaging module in the intelligent pod, and perform heat source decoupling processing on the multi-frame thermal radiation images to generate a dynamic temperature field map of each heat source.

[0027] Optionally, step 101 may specifically include: 1011. Capture multi-frame thermal radiation images of the UAV at consecutive time points through the infrared imaging module of the intelligent pod, and each frame of thermal radiation image includes temperature values of multiple pixels; 1012. Based on the radiation intensity distribution and spatial position stability of the multi-frame thermal radiation images, track the heat source contour of each independent heat source in consecutive frames; 1013. Use the tracked heat source contour to determine the radiation contribution of each independent heat source in the multi-frame thermal radiation images; In the above solution, the thermal radiation image refers to the two-dimensional temperature distribution data captured by the infrared imaging module, and each pixel value corresponds to the radiation intensity of a certain point in the scene. Heat source decoupling refers to the process of separating the radiation data of each independent heat source from the overlapping thermal radiation image containing multiple heat sources. The heat source contour refers to the boundary of the continuous region in the thermal radiation image where the temperature is significantly higher than the background, which is a closed polygon composed of adjacent high-temperature pixels. The radiation contribution refers to the temperature influence weight of a single heat source in the mixed pixels, and the true temperature value of each heat source is separated through decoupling calculation. The dynamic temperature field map contains the radiation gradient vector distribution on the target surface caused by the engine power fluctuation.

[0028] In the embodiment of the present application, first, in step 1011, the infrared imaging module in the intelligent pod continuously captures the thermal radiation energy on the surface of the drone at a fixed time interval, such as 0.1 second / frame; the detector consists of tens of thousands of independent pixels, such as a 384×288 array, to form physical pixel points. Each pixel receives the infrared radiation photons in the target area and converts them into an electrical signal intensity value through photoelectric conversion; after the signal is quantified by the analog-to-digital converter, combined with the preset material emissivity, such as the motor aluminum alloy ε = 0.88 and the environmental temperature compensation parameter, according to the formula , the absolute temperature value of each pixel is calculated, where S is the electrical signal intensity, k is the detector calibration coefficient, and σ is the Stefan constant; finally, the system maps according to the pixel spatial position into a two-dimensional temperature matrix, directly generating a frame of thermal radiation image containing 110,592, that is, 384×288 pixel point temperature values, and repeating this process at consecutive time points to obtain a thermal radiation image containing the temperature values of multiple pixel points.

[0029] Secondly, in step 1012, the gradient amplitude of the single-frame thermal image is calculated according to the radiation intensity distribution of the multi-frame thermal radiation images, and the temperature mutation region is extracted as the candidate heat source; then, a heat source centroid motion model is established according to the spatial position stability of the frame thermal radiation image, and according to this model, when it satisfies less than the maximum allowable displacement threshold and the variance of the centroid movement direction <15° within 3 consecutive frames, it is determined as a stable heat source; the heat source contours that meet the stability conditions are optimally matched through an algorithm to achieve the trajectory locking and tracking of the same heat source between multiple frames.

[0030] Finally, in step 1013, according to the tracked heat source contour, the scanning line algorithm is used to fill the heat source polygon boundary for each heat source to generate a binary mask matrix; then, the radiation value of the heat source is directly extracted in the non-overlapping region. When multiple heat source contours overlap, the radiation contribution is allocated according to the physical characteristics.

[0031] For example, in the night urban security scenario, the pod monitors the invading drone cluster in area A.

[0032] Specifically, the infrared module captures 5 consecutive thermal images with a frame interval of 0.1 second, detecting 3 moving heat sources as heat sources 1 - 3 and 1 static background heat source as heat source 4. By the optical flow method, it is identified that the displacement of heat source 1 in adjacent frames is only 2 pixels, which meets the stability threshold, and the contour is composed of 50 boundary pixels; NMF decomposition discovers a certain mixed pixel, that is, the temperature value of 82 °C at the coordinate [120, 300] is contributed 65% (weight 0.65) by heat source 1 and 35% (weight 0.35) by heat source 2. Finally, the dynamic map of heat source 1 is output: the temperature rises from 78 °C to 85 °C within 2 seconds, and the spatial gradient is 1.2 °C / pixel.

[0033] This step realizes the precise separation and dynamic modeling of heat sources through spatio-temporal joint analysis: using displacement stability determination to exclude background interference, combining matrix decomposition to quantify the radiation contributions of each heat source in the mixed pixel, and generating a high-fidelity spatio-temporal evolution map of the temperature field. Its advantage lies in eliminating the superposition effect of dense heat sources and providing a reliable independent heat source dynamic data basis for subsequent correlation analysis.

[0034] 102. Based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources, use a graph neural network to analyze the spatial correlation between the heat sources; Optionally, step 102 may specifically include: 1021. Extract the temperature gradient vectors from the dynamic temperature field maps of each heat source; 1022. Input the temperature gradient vectors as node features into the graph neural network, and the graph neural network learns the dependence relationships between the node features; 1023. Through the iterative calculation of the graph neural network, output the correlation weights between the nodes, and the correlation weights represent the degree of mutual influence of temperature changes between the heat sources; 1024. Generate a description of the spatial correlation between the heat sources according to the correlation weights.

[0035] In the above solution, the dynamic temperature field map refers to the time-series temperature distribution data of each generated heat source, including the spatial position and temperature value matrix. The temperature gradient vector refers to the physical quantity that describes the direction and intensity of temperature change in the temperature field, including the direction component (angle) and the magnitude component (change rate). The graph neural network refers to a deep learning model used to process graph-structured data and can learn the relationships between nodes. Node features refer to the attribute information vectors of each node (i.e., heat source) in the graph. The correlation weight refers to the degree of mutual influence of temperature changes between the heat sources. The description of spatial correlation refers to the analysis result of mutual influence presented in the form of a heat source relationship graph and a text report.

[0036] In the embodiments of the present application, first, spatial differential processing is performed on the dynamic temperature field maps of each heat source through step 1021, that is, the horizontal gradient is T ( x+1, y ) T ( x 1, y ),vertical gradient is T ( x , y +1) T ( x , y 1); where T(x, y) is the temperature value at position (x, y), the horizontal gradient is the temperature change rate in the x direction, and the vertical gradient is the temperature change rate in the y direction. Then calculate the gradient magnitude and angle of each heat source, that is , , where is the temperature change intensity, is the temperature change direction, is the vertical direction temperature gradient component, is the horizontal direction temperature gradient component. Output the above calculation results as the temperature gradient vector of each heat source.

[0037] Secondly, through step 1022, first encode the temperature gradient vector of each heat source, that is, the average gradient magnitude, the standard deviation of gradient change, and the main gradient direction angle together with the spatial position coordinates into a node feature vector as the input layer of the neural network; subsequently, the graph neural network enables each node to collect the feature information of its neighboring nodes through a multi-layer message passing mechanism and perform non-linear transformation and feature fusion operations in the hidden layer. This process allows the network to automatically capture the heat conduction patterns between heat sources. For example, when the main gradient direction of a certain heat source continuously points to another heat source, the network will strengthen the connection weight between the two nodes; finally, iteratively update the node state through multi-layer stacked graph convolutional layers, enabling the network to gradually learn the dependency relationships of the temperature change transfer law features between heat sources, such as the radiative influence of a high-temperature heat source on a neighboring low-temperature heat source, the synchronous temperature change trend between heat sources of the same material, etc.

[0038] Next, in the initial iteration layer of the graph neural network through step 1023, each heat source node transmits its temperature gradient features, including change intensity and direction information, to adjacent nodes through fully connected edges. The neural network analyzes the similarity of temperature change patterns through a feature comparison algorithm. For example, when the gradient directions of the main motor and the battery node continuously maintain a difference of less than 15°, a preliminary association signal is generated. Then, an attention screening mechanism is introduced in the intermediate iteration layer to focus on analyzing heat source node pairs with physical connections, such as heat conduction structures or spatial proximity, such as a distance less than 20 cm. The effective connections are strengthened through the analysis of the temporal correlation of temperature changes. If the temperature rise and fall changes of two heat sources show a correlation coefficient above 0.8 in the time series, their association weight is significantly increased. Finally, in the output layer, the cumulative association evidence is processed through a non-linear activation function, and a continuous weight value in the interval [0, 1], such as 0.86 representing strong mutual influence, is assigned to each pair of heat source nodes. This weight accurately quantifies the transfer efficiency of the temperature change of heat source A causing the temperature change of heat source B, completing the mathematical characterization of the degree of temperature mutual influence.

[0039] Finally, through step 1024, first, the numerical association weights are mapped into a visualized heat matrix graph. The matrix coordinate axes are sorted according to the physical positions of the heat sources, and the mutual influence relationship between the heat sources is intuitively displayed by the shade of the color. For example, dark red represents strong influence, and light blue represents weak influence. At the same time, a dynamic relationship network graph is generated, where the node size is proportional to the total weight output of the heat source, the edge width is proportional to the association weight value, and the core heat conduction path is marked. Finally, a multi-dimensional spatial association description is automatically generated based on the weight threshold to identify three types of key associations: the dominant heat source, that is, the highest output weight; the strongly coupled heat source pair, that is, the two-way weight > 0.8; and the isolated heat source, that is, the total association degree < 0.3.

[0040] For example, a certain unmanned aerial vehicle contains three heat sources: the main motor (node A), the battery (node B), and the controller (node C).

[0041] Specifically, first, the temperature gradient extraction shows that the main gradient direction of A is towards B (angle 120°), and the gradient of C is evenly distributed (σ = 0.15). Then, an initial fully connected graph is constructed through the graph neural network. After training, the output association weights are 0.82 (strong association), 0.31 (weak association), and 0.19 (extremely weak association). Finally, the generated spatial association description indicates that the main motor has a significant thermal impact on the battery, and it is recommended to increase the thermal insulation layer between the two; the controller is relatively independent and does not need to be adjusted.

[0042] This method automatically discovers the hidden association relationships between heat sources in complex systems, reveals the temperature propagation paths, and helps optimize the thermal management design; the generated association description intuitively displays the key influencing sources and the affected targets, providing a decision-making basis for preventing overheating failures; the adaptive learning ability of the graph neural network effectively improves the accuracy of thermal association analysis.

[0043] 103. Generate a damping torque that matches the wind disturbance spectrum based on the spatial correlation between heat sources, and correct the filtering parameters of the magnetorheological damping gimbal in the intelligent pod according to the frequency response characteristics of the damping torque; Optionally, step 103 may specifically include: 1031. Based on the spatial correlation between the heat sources, construct a spatial relationship model, where the spatial relationship model represents the position and mutual influence degree of each heat source as connection weights; 1032. Simulate the dynamic response of the wind disturbance spectrum at different frequencies through the spatial relationship model, so as to convert the connection weights into torque generation factors, and calculate the torque amplitudes and phases at multiple frequency points according to the torque generation factors; 1033. Synthesize a damping torque sequence according to the torque amplitudes and phases, where the spectral characteristics of the damping torque sequence are consistent with the characteristics of the wind disturbance spectrum; 1034. Compare the amplitude response and phase response in the frequency response characteristics of the extracted damping torque sequence with the current filtering parameters of the magnetorheological damping gimbal in the intelligent pod to generate a parameter deviation amount; 1035. Adjust the cut-off frequency and gain coefficient in the current filtering parameters according to the parameter deviation amount to generate corrected filtering parameters.

[0044] In the above solution, the spatial relationship model refers to representing the position and mutual influence degree of each heat source as connection weights. The torque generation factor refers to the conversion coefficient that converts the heat source correlation intensity into a physical and mechanical response, which maps the dimensionless correlation weight to a dimensional force that can act on the gimbal. The damping torque refers to making the pointing deviation of the optical axis of the intelligent pod under the side wind disturbance of 12 m / s less than the preset threshold. The damping torque sequence refers to a set of torque numerical values that change with time, and its fluctuation characteristics are completely matched with the external wind disturbance, and are used to actively cancel the influence of wind vibration. The frequency response characteristic refers to the response ability of the magnetorheological damping gimbal to vibrations of different frequencies, including the response speed and suppression effect. The filtering parameter refers to the core tuning parameter in the magnetorheological damping control system, which determines which frequencies of vibration interference the system filters.

[0045] In the embodiment of the present application, first, according to step 1031, the exact position coordinates (x, y, z) of each heat source in the three-dimensional space are obtained based on the intelligent pod coordinate system, and these positions are mapped to the nodes in the spatial network; then the mutual influence degree values calculated in step 102 are directly converted into the connection weight values between the nodes; then a connection edge dynamic generation mechanism is established. When the distance between two heat source nodes is less than a preset physical threshold, such as 10 cm and the mutual influence degree is greater than 0.4, a connection edge with mechanical properties is generated therebetween, and the line width and color depth of this edge are proportional to the connection weight, forming an intuitive spatial relationship model. For example, a weight of 0.8 generates a 8-pixel-wide red solid line, and a weight of 0.3 generates a 3-pixel-wide blue dashed line.

[0046] Secondly, through step 1032, the wind disturbance spectrum response simulation of the spatial model is carried out. The wind spectrum from 0 to 100 Hz is divided into 100 frequency points, and the vibration response of the model is analyzed at the 50 Hz frequency point. According to the weight value of the connection edge and the heat source position relationship, the sensitivity distribution of different frequency bands is determined to convert the correlation weight into a torque generation factor. For example, a weight of 0.7 - 1.0 corresponds to a high-frequency factor of 1.0, 0.4 - 0.7 corresponds to a medium-frequency factor of 0.6, and less than 0.4 corresponds to a low-frequency factor of 0.3. Then calculate the product of the generation factor and the reference coefficient to obtain the specific torque amplitude, where the reference coefficient comes from the wind tunnel experiment data table; calculate a 0.3° phase lag per centimeter of distance increase according to the node spacing to obtain the phase angle. For example, for a node spacing of 10 cm and a weight of 0.85, an amplitude of 0.25 Nm and a phase of -30° are output at the 50 Hz point.

[0047] Then, through step 1033, according to the torque amplitude, phase and energy distribution characteristics of the wind disturbance spectrum, fine frequency band complementation is carried out. New interpolation points are added every 1 Hz in the 30 - 50 Hz core area where the wind energy is concentrated. The amplitude and phase of the new points are calculated by the linear interpolation algorithm. For example, the amplitude at 33 Hz is taken as the average of the amplitudes at 32 Hz and 34 Hz, and the phase is taken as the nearest neighbor value; then a full-frequency band torque vector table is constructed to ensure that there are accurate amplitude and phase data corresponding to each frequency point; then time-domain waveform reconstruction is performed, and the sine torque components of each frequency point are superimposed and calculated along the time axis. That is, taking 1 ms as the time step, the sum of the initial values of all sine waves is calculated at t = 0 ms, such as -0.2 Nm, and the superposition of all sine waves after 1 ms of evolution is calculated at t = 1 ms, such as +0.05 Nm, and continuously iterated to generate a 1000 ms damping torque sequence; finally, spectrum calibration is performed. The synthesized sequence is analyzed by the fast Fourier transform, compared with the original wind spectrum curve, and the amplitude is finely adjusted and the phase is corrected for the frequency bands with a deviation > 3% until the main peak frequency deviation < 0.5 Hz and the harmonic distortion rate < 1%, and a damping torque sequence consistent with the wind disturbance spectrum characteristics is output.

[0048] Then, record the key feature points of the frequency response characteristic curve of the synthesized damping torque sequence through step 1034, such as the amplitude value of the main frequency peak 35 dB@35 Hz, the phase lag angle 155°, and the full-band response trend; at the same time, read the currently effective filter parameter set of the magnetorheological damping pan-tilt, such as the center frequency 40 Hz, the bandwidth ±5 Hz, the gain 1.2, and the phase compensation 15°; then perform multi-dimensional comparison in the three-coordinate analyzer. When performing amplitude response comparison, detect the amplitude value 35 dB of the damping sequence at 35 Hz and the actual response 33 dB of the pan-tilt filter at this frequency, that is, the difference +2 dB, to generate an amplitude deviation matrix; when performing phase response comparison, detect the phase difference between the damping sequence phase 155° and the preset 150° lag of the pan-tilt at the same frequency point 35 Hz to form a phase deviation spectrogram; finally, integrate the two sets of data through the intelligent calibration algorithm to extract the core deviation amounts, such as the main frequency amplitude difference +2 dB, the main frequency phase difference +5°, and the bandwidth deviation -3 Hz, and comprehensively generate a structured parameter deviation amount, such as the amplitude compensation coefficient +0.1, the center frequency shift amount +3 Hz, the phase compensation supplement +5°, and the bandwidth expansion coefficient ×1.2.

[0049] Finally, through step 1035, according to the parameter deviation amounts, start a hierarchical correction protocol for the four core parameters of the amplitude compensation coefficient, the center frequency shift amount, the phase compensation supplement, and the bandwidth expansion coefficient; among them, input the amplitude compensation coefficient into the digital gain controller to trigger the gain coefficient to step up from the current value 1.0 in three levels to 1.15 to adjust the gain coefficient; then reposition the center frequency offset by the numerically controlled oscillator, for example, increase the original 40 Hz by 2 Hz offset and lock it at 42 Hz to adjust the cut-off frequency; then inject time compensation for the phase lag amount through the delay circuit, and adjust the capacitance array of the band-pass filter for bandwidth expansion; then perform hardware-in-the-loop verification, input the original wind disturbance spectrum on the simulated wind vibration platform, and monitor the fluctuation range of the pan-tilt attitude angle after correction. If the residual vibration energy > the threshold, start secondary fine-tuning; finally, generate an encrypted control instruction set with a timestamp, write it into the FPGA controller of the magnetorheological damping pan-tilt through the dual CAN bus, and at the same time back up the original parameters to the secure storage area to generate the corrected filter parameters.

[0050] For example, in the night inspection task of brand A drones in area B.

[0051] Specifically, the intelligent pod detects the spatial correlation weight of 0.88 between the heat source C of the main motor (coordinates (0.1, 0.2, 0.3)) and the heat source D of the infrared lens driver (coordinates (0.15, 0.25, 0.35)). First, an elastic model connecting the two heat sources is constructed, such as an 8.8 px orange connecting line, with a spacing of 18 mm < 20 cm threshold. After verifying that the motor temperature rise is 20 °C through heat conduction, the temperature rise of the driver is 12 °C within 3 seconds; when suddenly encountering a 6 - level gust of wind, the system converts the weight into a torque factor of 0.92, calculates the point torque amplitude of 0.15 Nm and the phase angle of 48° at 28 Hz; synthesizes the damping torque sequence; extracts the frequency response characteristics of the sequence 28 Hz @ 29.5 dB and the phase of 48° and compares them with the current parameters of the magneto - fluid gimbal to generate a parameter deviation of +1.3 dB in amplitude and -1.9 Hz in center frequency deviation; accordingly, the gimbal filtering parameters are corrected, the cut - off frequency is translated from 28 - 38 Hz to 26.1 - 36.1 Hz, and the gain coefficient is stepped from 1.1 to 1.23; finally, the gimbal jitter is reduced from ±1.2° to ±0.4°, ensuring the high - definition imaging ability of the pod under strong winds.

[0052] This solution constructs a physical model by quantifying the spatial correlation of heat sources, converts the thermal coupling intensity into anti - wind - disturbance control parameters; generates a damping torque sequence with precisely matched frequency spectrum based on the dynamic response mechanism to achieve a mechanical equivalent reproduction of external wind disturbances; intelligently compares the frequency response characteristics through a closed - loop feedback system to drive the magneto - fluid damping gimbal to dynamically adjust the cut - off frequency and gain coefficient; finally, an adaptive mapping from the spatial relationship of heat sources to vibration - reduction control is achieved, significantly improving the imaging stability and target tracking ability of the pod in complex wind - disturbance environments, while optimizing the system response timeliness and avoiding manual tuning deviation.

[0053] 104. Map the temperature gradient vector to the corresponding energy transfer path, and use the filtering parameter to calibrate the interference intensity between two adjacent energy transfer paths. Optionally, step 104 may specifically include: 1041. Use each vector element in the temperature gradient vector as a starting point, perform path tracking along the vector direction to generate multiple energy transfer paths. 1042. Calculate the overlapping area between two adjacent energy transfer paths, and quantify the overlapping area as the initial interference intensity. 1043. Use the cut - off frequency and gain coefficient in the filtering parameter to scale - adjust the initial interference intensity to determine the calibrated interference intensity, where the process of scale - adjustment includes: using the cut - off frequency to filter out high - frequency interference components, and using the gain coefficient to amplify or attenuate the amplitude of the initial interference intensity.

[0054] In the above solution, the temperature gradient vector refers to the physical quantity that describes the direction and magnitude of the temperature change on the heat source surface, including two-dimensional information of angle (radiation direction) and intensity (rate of change). The energy transfer path refers to the straight line path starting from the center of the heat source and extending along the direction of the temperature gradient vector, representing the main heat diffusion channel. The initial interference intensity refers to the quantified value of the thermal energy superposition in the intersection area of adjacent energy paths, reflecting the degree of energy crosstalk between paths. The cut-off frequency refers to the highest frequency threshold (unit: Hz) allowed to pass through the filtering system, used to block high-frequency interference. The gain coefficient refers to the signal amplitude regulation factor (dimensionless), used to amplify or attenuate the interference intensity. The calibrated interference intensity refers to the quantified interference value after being filtered by the cut-off frequency and adjusted by the gain coefficient, representing the true and effective thermal interference. The overlapping area is determined based on the minimum distance between the spatial points and the included angle of the path directions.

[0055] In the embodiment of the present application, first, a set of temperature gradient vectors of each heat source surface is obtained through step 1041, that is, including the direction angle and intensity value, and the starting coordinate point of the vector is used as the path reference point; then, a straight-line tracking is performed along the vector direction, and the ray marching algorithm is used to start from the reference point, and the step size of each advancement is equal to the characteristic size of the heat source material. During the advancement process, the rate of temperature change is monitored, and the tracking stops when the temperature gradient decays to 30% of the initial value; then, the sequence of path trajectory points is recorded in the three-dimensional space coordinate, and the complex heat source is automatically branched, for example, the irregular heat sink heat source is split into 3 secondary paths in the vector direction; finally, all paths are integrated to generate multiple energy transfer paths.

[0056] Secondly, according to the energy transfer paths generated in 1041 through step 1042, the spatial scanning algorithm is used to detect all adjacent path pairs with a distance less than the average size of the heat source; then, the intersection point coordinates of the qualified path pairs are calculated, that is, the intersection point is determined by solving the analytical solution of the two-path straight-line equations, and when there is no intersection point, the nearest projection point is calculated; subsequently, an overlapping area model with the intersection point as the ellipse core is constructed, the length of the major axis of the ellipse is set to 0.5 times the distance between the two paths, the minor axis is 0.2 times the distance, and the main axis direction is parallel to the bisector of the included angle between the two paths; then, all temperature data points within the ellipse area are extracted from the dynamic temperature field map, and the Gaussian weighted integral algorithm is used to calculate the thermal field energy: the temperature value of each pixel point is multiplied by the coordinate weight coefficient, and the total thermal energy value of the area is obtained by accumulation; finally, through the thermodynamics-electrics conversion coefficient, this thermal energy value is mapped to the dimensionless initial interference intensity.

[0057] Finally, obtain the initial interference intensity value calculated in step 1042 through step 1043, and at the same time read the cut-off frequency and gain coefficient in the currently effective filtering parameters of the magneto-rheological damping pan-tilt; then perform high-frequency interference identification according to the spatial angle characteristics of the two energy transfer paths. If the path angle is greater than 60 degrees, it is marked as a high-frequency interference component. For example, the intersection angle of path A at 45° and path B at 120° is 75° > 60°; then perform high-frequency filtering based on the cut-off frequency. When the cut-off frequency < 30 Hz, apply 100% suppression to the marked high-frequency interference components; when 30 Hz ≤ cut-off frequency ≤ 50 Hz, adopt linear proportional suppression. For example, when the cut-off frequency = 40 Hz, retain 70% of the intensity; when the cut-off frequency > 50 Hz, completely retain the high-frequency interference components; then apply the gain coefficient for amplitude adjustment. When the interference intensity is lower than the noise threshold, use the 1.5th power of the gain coefficient for intensity compensation to suppress the attenuation of weak signals. When it is higher than the safety threshold, use the 0.8th power of the gain coefficient for attenuation protection to avoid overload; finally, perform weight reorganization according to the thermal conductivity of the path material to generate the calibrated interference intensity value.

[0058] For example, in the inspection operation of model A UAV, the gradient vector of the heat source of its intelligent pod detecting the motor is [45°, 15 °C / cm]), and the gradient vector of the heat source of the controller is [120°, 8 °C / cm].

[0059] Specifically, first generate three paths of the motor (in the directions of 45° / 30° / 60°), and two paths of the controller (in the directions of 120° / 150°); detect the intersection of the 45° and 120° paths at the coordinate (0.22, 0.31), calculate the area of the elliptical overlapping area as 0.8 cm², the average temperature as 72 °C, and the initial interference intensity as 57.6 °C·cm²; then read the current filtering parameters of the pan-tilt (cut-off frequency 35 Hz, gain 1.25). Since the intersection angle of 75° > 60°, retain 80% of the high-frequency components. After gain adjustment, the output calibrated intensity is 1.25×0.8×57.6×0.85 ≈ 49 °C·cm², providing an accurate interference quantification basis for thermal management decision-making.

[0060] This method realizes the deep integration of the physical characteristics of the temperature field and signal processing technology, converts the temperature gradient vector into an operable energy transfer path; accurately captures the path-crossing thermal interference through geometric modeling; dynamically calibrates the interference intensity using the filtering parameters of the pan-tilt to effectively distinguish true thermal coupling from high-frequency noise; ultimately improves the quantification analysis ability of the pod system for heat conduction interference, provides a physically consistent decision benchmark for optimizing thermal management strategies, and significantly enhances the adaptability of the pod in complex thermal environments.

[0061] 105. Extract the heat diffusion characteristics from the temperature gradient vector, and generate a threat propagation path according to the calibrated interference intensity and the heat diffusion characteristics; Optionally, step 105 may specifically include: 1051. Perform time series decomposition on the temperature gradient vectors in the dynamic temperature field map of each heat source, extract multiple features such as the change trajectory, fluctuation pattern, and cumulative change amount of the temperature gradient direction at consecutive time points, and respectively form a direction change sequence, an intensity fluctuation sequence, and a heat diffusion energy value. Specifically, according to the change trajectory of the temperature gradient direction at consecutive time points, a direction change sequence is formed; according to the fluctuation pattern of the temperature gradient magnitude at consecutive time points, an intensity fluctuation sequence is formed; and according to the cumulative change amount of the temperature gradient vector within a preset time window, a heat diffusion energy value is formed. 1052. Combine the direction change sequence, the intensity fluctuation sequence, and the heat diffusion energy value into a heat diffusion feature vector for each heat source. 1053. Construct an interference intensity table between heat sources according to the calibrated interference intensity, where each table entry stores the interference intensity value between a pair of heat sources, and calculate the feature similarity of the heat diffusion feature vectors of any two heat sources. The similarity is determined by fusing the trajectory repeatability and fluctuation consistency of the two heat sources. 1054. Multiply the feature similarity by the interference intensity value of the corresponding heat source pair in the interference intensity table to obtain the propagation correlation weight between heat sources. 1055. Take the heat source with the largest heat diffusion energy value as the starting node, select the adjacent heat source with the highest propagation correlation weight with the current heat source from the unvisited heat sources, add the selected heat source to the path sequence, and update it as the current heat source. Repeat the above steps until the propagation correlation weight is lower than the preset threshold or there are no selectable adjacent heat sources, and output the path sequence as the threat propagation path.

[0062] In the above solution, the temperature gradient vector refers to a physical quantity that characterizes the direction and magnitude of the temperature change rate on the surface of the heat source, including the angle (change direction) and intensity (change amount). The heat diffusion feature refers to three key features extracted from the temperature gradient vector, namely, the direction change sequence (trajectory time series change pattern), the intensity fluctuation sequence (gradient amplitude time series evolution), and the heat diffusion energy value (gradient change cumulative effect). The interference intensity table refers to a two-dimensional matrix that stores the calibrated interference intensity between heat source pairs, quantifying the cross-influence of heat sources. The feature similarity refers to a similarity index calculated by fusing the trajectory coincidence degree and fluctuation synchronization. The propagation correlation weight refers to the product of the feature similarity and the interference intensity, determining the priority of heat threat conduction. The threat propagation path refers to a conduction link sequence generated along the highest correlation weight starting from the maximum heat source energy node. The trajectory repeatability and fluctuation consistency refer to performing trajectory matching on the direction change sequences of two heat sources to calculate the trajectory coincidence degree, and performing phase alignment on the intensity fluctuation sequences of two heat sources to calculate the fluctuation consistency. In the application embodiment, first, time series decomposition is performed on the temperature gradient vectors in each frame of the spectrum through step 1051, that is, the vector data from the 1st second to the 10th second is taken to form a vector sequence on the time axis. The angular change value of the vector directions in two adjacent seconds is calculated, and the change values in 9 time steps are continuously recorded to form a direction change sequence, so as to extract multiple features such as the change trajectory, fluctuation pattern, and cumulative change amount of the temperature gradient direction at consecutive time points; then the temporal change of the gradient direction is analyzed, and the vector direction offset between adjacent frames is traced. For example, 102° at the 1st second → 105° at the 2nd second → offset +3°, generating a direction change sequence containing 9 time steps, and synchronously recording the direction fluctuation entropy value. A high entropy value indicates direction disorder; at the same time, the evolution law of the gradient intensity is extracted, calculating the standard deviation of the fluctuation amplitude reflected by the gradient of each frame, and the valley difference of the peak capture mutation intensity between adjacent frames, forming a two-parameter intensity fluctuation sequence; finally, within a 10-second time window, based on the linear accumulation of the product of the gradient intensity vector modulus and the temperature change rate, that is, integrating the gradient intensity per second, a heat diffusion energy value is generated.

[0063] Secondly, through step 1052, the three independent features generated in step 1051, namely the direction change sequence, the intensity fluctuation sequence, and the heat diffusion energy value, are integrated. They are concatenated end to end in the order of dimensions, that is, the direction sequence is placed in the first 9 positions of the vector, the intensity sequence is in the 10th - 11th positions, and the energy value is fixed at the 12th position, forming a 12-dimensional original feature vector; then principal component analysis dimensionality reduction optimization is performed, calculating the eigenvectors of the covariance matrix and retaining the first 5 principal components with a cumulative contribution rate exceeding 95%, such as the principal component PC1 of the direction temporal pattern, the principal component PC2 of the intensity fluctuation, etc., compressing the 12-dimensional vector into a 5-dimensional core feature; finally, the heat source spatial position label is added to generate an 8-dimensional standardized feature vector. The whole process strictly follows the engineering path of "feature splicing → dimensionality compression → spatial anchoring" to ensure that the feature vector contains the dynamic temporal characteristics, energy intensity attributes, and physical position information of heat diffusion.

[0064] Next, according to the calibrated interference intensity value output in step 1043, a heat source interference intensity matrix table is constructed through step 1053, that is, a lower triangular matrix with the heat source ID as the row-column index, and each table entry stores the interference intensity value of a unique heat source pair. For example, the interference intensity value between the main motor and the infrared camera module is 0.78, forming the basis of the relationship network of internal pod thermal interference. Secondly, for all heat source pairs, the feature similarity is calculated through a double similarity fusion algorithm. The first step analyzes the trajectory repeatability of the direction change sequence, and the dynamic time warping technology is used to compare the sequence forms. For example, the waveform matching degree between the main motor sequence [+3°, +5°, -2°...] and the infrared camera sequence [+2°, +1°, +3°...] is 0.85. The second step evaluates the fluctuation consistency of the intensity fluctuation sequence, and the synchronous fluctuation mode is detected through the covariance index. For example, both heat sources have an intensity sudden increase of >15% at second 5, and it is marked as high covariance. Finally, the trajectory repeatability weight of 0.6 and the fluctuation consistency weight of 0.4 are linearly fused to output the feature similarity in the range of 0.0 - 1.0.

[0065] Then, through step 1054, a two-factor coupling calculation is performed according to the heat source pair feature similarity calculated in step 1053 and the corresponding item in the interference intensity table, that is, the feature similarity and the interference intensity value are multiplied scalarly. For example, 0.878 × 0.78 = 0.685, to obtain the propagation correlation weight. Finally, through step 1055, the heat source with the maximum heat diffusion energy value, such as the main motor with 185.3 energy units, is used as the starting node of the path and added to the threat path sequence. Then, all adjacent nodes with propagation correlation weights to the current heat source are retrieved from the set of unvisited heat sources, and the heat source with the highest weight is selected. For example, the weight of the main motor → infrared camera is 0.806. Then, double verification is performed. If the weight ≥ the preset threshold, the heat source is added to the path sequence and updated as the current node. Taking the updated node as the base point, its unvisited adjacent heat sources are retrieved repeatedly, and the expansion stops when the weight < the threshold. Finally, when there are no reachable adjacent nodes or all weights are lower than the threshold, the search is terminated, and an ordered threat propagation path sequence is output.

[0066] For example, in the intelligent pod system when performing power inspection tasks.

[0067] Specifically, first, extract the sequential temperature gradient data of three core heat sources (main motor module C, infrared camera module D, and image processor E) to form a direction change sequence and an intensity fluctuation pattern, and calculate the heat diffusion energy values of each heat source; secondly, construct the feature similarity of module C-D as 0.878. For example, the comprehensive direction trajectory matching degree is 0.85 and the fluctuation consistency is 0.92, and generate a propagation weight of 0.685 by combining the pre-stored interference intensity table; then, starting from the maximum energy source module C, retrieve the propagation weights of adjacent heat sources, and select the module D with the highest weight to join the path; then, taking module D as the current node, detect that the weight of module E is 0.339 and does not reach the threshold, and terminate the path expansion; finally, output the threat propagation path [main motor C → infrared camera D], drive the pod cooling system to enhance the heat dissipation power of module D by 20% directionally, and at the same time reduce the working load of module C by 15%, effectively blocking the conduction link of the heat threat inside the system and ensuring the continuous and stable execution of the inspection task.

[0068] This method realizes the deep coupling of thermophysical characteristics and topological analysis, accurately captures the essence of heat conduction through multi-dimensional heat diffusion feature extraction; innovatively integrates interference intensity and feature similarity to generate propagation weights; reveals the heat threat conduction path based on the energy-driven path search algorithm; finally constructs a threat propagation model with strong interpretability and clear physical meaning, provides reliable decision-making support for the thermal safety management of unmanned aerial vehicles, and greatly improves the system's forward-looking prediction ability for thermal failure risks.

[0069] 106. According to the threat propagation path, dynamically plan the deflection angle of the intelligent pod so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

[0070] Optionally, step 106 may specifically include: 1061. Obtain the spatial coordinates of all heat sources in the threat propagation path to form a coordinate set, and calculate the geometric centroid of the coordinate set. The geometric centroid includes the centroid coordinates of multiple components; 1062. Obtain the current pointing angle of the intelligent pod, where the current pointing angle includes the horizontal yaw angle and the vertical pitch angle; 1063. Convert the centroid coordinates to the pod coordinate system through coordinate transformation, and calculate the horizontal yaw angle correction amount and the vertical pitch angle correction amount according to the relative spatial position between the converted centroid coordinates and the intelligent pod, and use them as the target pointing angle; 1064. Decompose the difference between the current pointing angle and the target pointing angle into angular velocity components, perform kinematic constraints on the angular velocity components according to the inertial parameters of the magnetohydrodynamic damping gimbal, output a continuous angle change sequence that meets the constraints, and control the intelligent pod to deflect according to the continuous angle change sequence so that the center of the optical field of view aligns with the geometric centroid coordinates.

[0071] In the above solution, the threat propagation path refers to the directional conduction channel of thermal energy among the internal components of the pod. This path is composed of a sequence of heat source nodes, such as [motor → lens driver → controller]), each node contains precise spatial coordinates, and the connecting edges between nodes represent the direction of heat threat conduction and carry the conduction risk level weight. The geometric centroid means that the geometric centroid includes the centroid X coordinate, centroid Y coordinate, and centroid Z coordinate. Among them, the arithmetic mean of the X components of all heat source coordinates in the coordinate set is taken to obtain the centroid X coordinate, the arithmetic mean of the Y components of all heat source coordinates in the coordinate set is taken to obtain the centroid Y coordinate, and the arithmetic mean of the Z components of all heat source coordinates in the coordinate set is taken to obtain the centroid Z coordinate. The current pointing angle refers to the spatial direction of the optical axis of the intelligent pod. The target pointing angle refers to the yaw and pitch angles that need to be corrected. The angular velocity component refers to the velocity projection of the difference between the target deflection angle and the current pointing angle in the yaw and pitch directions, which is restricted by the inertial parameters of the pan-tilt head. The inertial parameter refers to the motion constraint of the magneto-fluid pan-tilt head (maximum angular velocity ±100° / s, maximum angular acceleration ±500° / s²).

[0072] In the embodiment of the present application, first, all the spatial three-dimensional coordinates of all heat sources are extracted from the threat propagation path node by node through step 1061, such as the motor module position (120, 80, 100), the lens driver position (135, 95, 105), to form a complete coordinate data set; then the geometric centroid of this set is calculated, where the arithmetic mean operation is independently performed on the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate in the three-dimensional coordinate system of the pod, that is, the sum of all heat source X coordinate values is divided by the number of heat sources to obtain the centroid X component, and the centroid Y component and Z component are calculated in the same way; finally, the centroid coordinates are output.

[0073] Secondly, through step 1062, three-dimensional motion data is integrally collected for the main body of the intelligent pod. Among them, the magnetometer detects the geomagnetic direction to calculate the absolute horizontal yaw angle in the range of 0° to 360°, and the three-axis accelerometer measures the gravity direction to calculate the vertical pitch angle in the range of -90° to 90°; secondly, the gyroscope provides the angular rate data, and the dynamic motion error is compensated through the quaternion solution algorithm; finally, the data of the three sensors are fused in the data processing unit, and the Kalman filtering technology is used to eliminate the instantaneous jitter error, and the smooth and accurate current pointing angle is output, such as the horizontal yaw angle is accurate to ±0.1°, and the vertical pitch angle is ±0.05°.

[0074] Next, through step 1063, according to the pod installation attitude parameters, such as the horizontal inclination angle and height offset, the geometric centroid coordinates calculated in step 1061 are used to translate the origin of the centroid coordinates to the rotation center of the pan-tilt by applying a rigid body transformation matrix, so as to convert them to the pod body coordinate system; then, the horizontal projection deviation between the centroid point and the Z-axis is measured in the XOY plane of the pod coordinate system as the yaw correction reference, and the vertical projection deviation is measured in the YOZ plane as the pitch correction reference; then, based on the pinhole imaging model where the horizontal yaw angle correction amount is proportional to the arctangent ratio of the X-direction deviation distance to the focal length, and the vertical pitch angle correction amount is proportional to the arctangent ratio of the Y-direction deviation distance to the focal length, a geometric projection relationship is established to calculate the spatial relative relationship between the converted centroid point and the optical field center of the pod; finally, the target pointing angle is output with the original pointing angle + angle correction amount.

[0075] Finally, through step 1064, calculate the difference between the target pointing angle and the current pointing angle, and decompose the horizontal yaw difference and the vertical pitch difference into horizontal angular velocity components and pitch angular velocity components respectively; then read the inertial parameters of the magnetorheological damping pan-tilt, including the maximum angular velocity limit amplitude (±100° / s) and angular acceleration constraint (±500° / s²), and perform kinematic constraints on the horizontal angular velocity components and pitch angular velocity components. When the horizontal angular velocity component > 100° / s, perform speed limiting, and at the same time constrain the acceleration value to avoid inertial shock; then generate a continuous angle change sequence based on equal division of the time axis, that is, one control point every 20 ms, and each point smoothly transitions through three stages of uniform acceleration-uniform velocity-uniform deceleration. On the premise of ensuring that both the speed and acceleration do not exceed the hardware limits of the pan-tilt, output a discrete trajectory point sequence that meets the motion constraints; finally, drive the intelligent pod to execute this sequence, such as starting point at 0 ms → time Angle A → time Angle B →... → end point, so that the optical field center continuously and stably aligns with the geometric centroid coordinates within the set time.

[0076] For example, when the UAV of model A performs an inspection of area B, the threat path is [motor(120, 80, 100) → camera(135, 95, 105)].

[0077] Specifically, first calculate the geometric centroid, and the result is (125.2, 90.5, 102.3); the current pointing of the pod is 45.2 and 12.8°; through coordinate transformation calculation, corrections of 0.72° and 0.44° are required; control the pan-tilt to smoothly rotate at a horizontal angular velocity of 7.2° / s and a pitch angular velocity of 4.4° / s within 200 ms, so that the optical field center accurately aligns with the thermal threat centroid position.

[0078] This method realizes the dynamic coupling of threat path space features and pod control, improves the alignment accuracy through the weighted centroid positioning of heat source energy; the motion control constraint ensures that the line of sight stably tracks the target; finally, it achieves continuous spatial synchronization between the optical field of view and the threat core area, significantly enhancing the high-definition monitoring ability of the heat conduction process.

[0079] The following is a complete embodiment of steps 101 - 106: In an urban night security mission, a patrol UAV equipped with an intelligent pod detected a cluster of 3 intruding UAVs in area A (coordinates X: 102.3, Y: 37.8). First, the pod infrared imaging module (384×288 resolution) captured multiple frames of thermal radiation images at 0.1 second / frame. Through optical flow method tracking and NMF decomposition, three independent heat sources (heat sources 1 - 3) were decoupled. Among them, the centroid displacement of heat source 1 was ≤2 pixels and the direction variance <15° within 5 consecutive frames, generating a dynamic temperature field map, that is, the temperature was 78℃ → 85℃ within 2 seconds, and the gradient was 1.2℃ / pixel.

[0080] Secondly, based on the temperature gradient vector, that is, the main direction of heat source 1 was 45° / amplitude 15℃ / cm, the graph neural network was used to analyze the spatial correlation, and the correlation weight of heat source 1→2 was output as 0.82. Because the gradient direction continuously pointed and the distance <15 cm, the significant thermal influence of the main motor (heat source 1) on the battery compartment (heat source 2) was identified.

[0081] Then, according to the correlation weight of 0.82, a spatial relationship model was constructed, the connection weight was converted into a torque generation factor of 0.92, the torque amplitude of the 28Hz wind disturbance point was calculated as 0.15 Nm / phase 48°, and a damping torque sequence with a synthetic spectrum matching degree >97% was generated; by comparing the original parameters of the magneto - fluid gimbal, the filter parameters were corrected to a cut - off frequency of 26.1 - 36.1Hz and a gain coefficient of 1.23, reducing the gimbal jitter from ±1.2° to ±0.4°. Step 104: Map the gradient vector of heat source 1 to an energy path in the 45° direction, which intersects with the 120° path of heat source 2 at (0.22, 0.31), and calculate the initial interference intensity of the elliptical overlapping area as 57.6℃·cm²; applying the corrected filter parameters (cut - off frequency 26.1Hz / gain 1.23), retaining 80% of the high - frequency components (the intersection angle 75°>60°), and outputting the calibrated interference intensity of 49℃·cm². Step 105: Extract the heat diffusion characteristics of heat source 1 (direction change sequence [+3°, +5°, - 2°...], intensity fluctuation standard deviation 4.2℃ / cm, heat diffusion energy value 185.3 units), calculate the feature similarity of heat source 1→2 as 0.878 (trajectory matching degree 0.85 + fluctuation consistency 0.92); multiply the feature similarity by the calibrated interference intensity to obtain the propagation correlation weight of 43.0, and generate a threat propagation path [heat source 1→heat source 2] starting from the heat source 1 with the maximum energy source (the weight of heat source 2→3 is 21.7 < threshold 30).

[0082] Finally, calculate the geometric centroid of the threat path (0.125, 0.225, 0.325), and solve the target correction amount (yaw +0.72° / pitch +0.44°) based on the current pointing angle of the pod (yaw 42.5° / pitch 10.1°); generate a 200ms smooth deflection sequence (horizontal angular velocity 7.2° / s, pitch angular velocity 4.4° / s) through kinematic constraints (angular velocity limit ±100° / s), and drive the center of the optical field of view to accurately align with the centroid (offset error <0.05°) to achieve continuous tracking of the heat threat conduction path.

[0083] Figure 2 The embodiment of the present application provides a schematic structural diagram of an intelligent pod control system for a drone, as Figure 2 shown. The system includes: A first generation module 21, configured to obtain multiple frames of thermal radiation images of the drone through an infrared imaging module in the intelligent pod, perform heat source decoupling processing on the multiple frames of thermal radiation images, and generate a dynamic temperature field map of each heat source; An analysis module 22, configured to analyze the spatial correlation between heat sources by using a graph neural network based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources; A correction module 23, configured to generate a damping torque matching the wind disturbance spectrum according to the spatial correlation between heat sources, and correct the filtering parameters of the magneto-rheological damping gimbal in the intelligent pod according to the frequency response characteristics of the damping torque; A calibration module 24, configured to map the temperature gradient vector to a corresponding energy transfer path, and calibrate the interference intensity between two adjacent energy transfer paths by using the filtering parameters; A second generation module 25, configured to extract heat diffusion features from the temperature gradient vector, and generate a threat propagation path according to the calibrated interference intensity and the heat diffusion features; A planning module 26, configured to dynamically plan the deflection angle of the intelligent pod according to the threat propagation path, so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

[0084] Figure 2 The above-mentioned intelligent pod control system for a drone can execute Figure 1 the intelligent pod control method for a drone described in the embodiment shown. The implementation principle and technical effects will not be elaborated here. For the above-mentioned intelligent pod control system for a drone, the specific manners in which each module and unit perform operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0085] In a possible design, Figure 2The intelligent pod control system of the drone in the illustrated embodiment can be implemented as a computing device, such as Figure 3 as shown. The computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, and among them, the one or more computer instructions are for the processing component 32 to call and execute.

[0086] The processing component 32 is used for the intelligent pod control method of the drone in the above Figure 1 illustrated embodiment.

[0087] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0088] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0089] Of course, the computing device may necessarily further include other components, such as an input / output interface, a display component, a communication component, etc.

[0090] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0091] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0092] Among them, the computing device may be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device may refer to a cloud server, and the above processing component, storage component, etc. may be basic server resources leased or purchased from the cloud computing platform.

[0093] The embodiment of the present application further provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the intelligent pod control method of the drone in the above Figure 1 illustrated embodiment.

[0094] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0096] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. An intelligent pod control method for an unmanned aerial vehicle, characterized in that Including: Obtaining multiple thermal radiation images of the drone through an infrared imaging module in the intelligent pod, performing heat source decoupling processing on the multiple thermal radiation images to generate a dynamic temperature field map of each heat source; Analyzing the spatial correlation between heat sources using a graph neural network based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources; Generating a damping torque matching the wind disturbance spectrum according to the spatial correlation between heat sources, and correcting the filtering parameters of the magneto-rheological damping gimbal in the intelligent pod according to the frequency response characteristics of the damping torque; Mapping the temperature gradient vector to a corresponding energy transfer path, and calibrating the interference intensity between two adjacent energy transfer paths using the filtering parameters; Extracting heat diffusion features from the temperature gradient vector, and generating a threat propagation path according to the calibrated interference intensity and the heat diffusion features; Dynamically programming the deflection angle of the intelligent pod according to the threat propagation path, so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

2. The method according to claim 1, wherein Extracting heat diffusion features from the temperature gradient vector, and generating a threat propagation path according to the calibrated interference intensity and the heat diffusion features, including: Performing time series decomposition on the temperature gradient vectors in the dynamic temperature field map of each heat source, extracting multiple features such as the change trajectory, fluctuation pattern, and cumulative change amount of the temperature gradient direction at consecutive time points, respectively forming a direction change sequence, an intensity fluctuation sequence, and a heat diffusion energy value; wherein, according to the change trajectory of the temperature gradient direction at consecutive time points, a direction change sequence is formed; according to the fluctuation pattern of the temperature gradient magnitude at consecutive time points, an intensity fluctuation sequence is formed; according to the cumulative change amount of the temperature gradient vector within a preset time window, a heat diffusion energy value is formed; Combining the direction change sequence, the intensity fluctuation sequence, and the heat diffusion energy value into a heat diffusion feature vector of each heat source; Constructing an interference intensity table between heat sources according to the calibrated interference intensity, where each table entry stores the interference intensity value between a pair of heat sources, and calculating the feature similarity of the heat diffusion feature vectors of any two heat sources, and the similarity is determined by fusing the trajectory repeatability and fluctuation consistency of the any two heat sources; Multiplying the feature similarity by the interference intensity value of the corresponding heat source pair in the interference intensity table to obtain the propagation correlation weight between heat sources; Taking the heat source with the largest heat diffusion energy value as the starting node, selecting the adjacent heat source with the highest propagation correlation weight with the current heat source from the unvisited heat sources, adding the selected heat source to the path sequence, and updating it as the current heat source, repeating the above steps until the propagation correlation weight is lower than a preset threshold or there are no selectable adjacent heat sources, and outputting the path sequence as the threat propagation path.

3. The method according to claim 1, wherein Dynamically programming the deflection angle of the intelligent pod according to the threat propagation path, so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path, including: Obtaining the spatial coordinates of all heat sources in the threat propagation path to form a coordinate set, and calculating the geometric centroid of the coordinate set, where the geometric centroid includes the centroid coordinates of multiple components; Obtaining the current pointing angle of the intelligent pod, where the current pointing angle includes a horizontal yaw angle and a vertical pitch angle; The centroid coordinates are transformed to the pod coordinate system through coordinate transformation. The horizontal yaw angle correction amount and the vertical pitch angle correction amount are calculated based on the transformed centroid coordinates and the spatial relative position of the intelligent pod, and are used as the target pointing angles. The difference between the current pointing angle and the target pointing angle is decomposed into angular velocity components. According to the inertial parameters of the magneto - rheological damping gimbal, kinematic constraints are imposed on the angular velocity components, and a continuous angle change sequence that satisfies the constraints is output. The intelligent pod is controlled to deflect according to the continuous angle change sequence so that the optical field of view center is aligned with the geometric centroid coordinates.

4. The method according to claim 1, wherein Multiple frames of thermal radiation images of the UAV are obtained through the infrared imaging module in the intelligent pod, and the multiple frames of thermal radiation images are subjected to heat source decoupling processing to generate a dynamic temperature field map of each heat source, including: Through the infrared imaging module of the intelligent pod, multiple frames of thermal radiation images of the UAV are captured at continuous time points, and each frame of thermal radiation image contains temperature values of multiple pixels. Based on the radiation intensity distribution and spatial position stability of the multiple frames of thermal radiation images, the heat source contour of each independent heat source is tracked in consecutive frames. Using the tracked heat source contour, the radiation contribution of each independent heat source in the multiple frames of thermal radiation images is determined. According to the radiation contribution, a unique identifier is assigned to each heat source, and the temperature distribution data of each heat source in each frame of the image is extracted. The temperature distribution data of each heat source is integrated according to the time series to generate a dynamic temperature field map of the heat source.

5. The method according to claim 1, characterized in that Based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources, a graph neural network is used to analyze the spatial correlation between the heat sources, including: The temperature gradient vector is extracted from the dynamic temperature field map of each heat source. The temperature gradient vector is input into the graph neural network as node features, and the graph neural network learns the dependency relationship between the node features. Through the iterative calculation of the graph neural network, the association weights between nodes are output, and the association weights represent the degree of mutual influence of temperature changes between heat sources. According to the association weights, a description of the spatial correlation between heat sources is generated.

6. The method according to claim 1, characterized in that, According to the spatial correlation between heat sources, a damping torque that matches the wind disturbance spectrum is generated, and the filtering parameters of the magneto - rheological damping gimbal in the intelligent pod are corrected according to the frequency response characteristics of the damping torque, including: Based on the spatial correlation between the heat sources, a spatial relationship model is constructed, where the spatial relationship model represents the position and mutual influence degree of each heat source as connection weights. The dynamic response of the wind disturbance spectrum at different frequencies is simulated through the spatial relationship model to convert the connection weights into torque generation factors, and the torque amplitudes and phases at multiple frequency points are calculated according to the torque generation factors. According to the torque amplitudes and phases, a damping torque sequence is synthesized, where the spectral characteristics of the damping torque sequence are consistent with the characteristics of the wind disturbance spectrum. The amplitude response and phase response in the frequency response characteristics of the extracted damping torque sequence are compared with the current filtering parameters of the magneto - rheological damping gimbal in the intelligent pod to generate a parameter deviation amount. According to the parameter deviation amount, the cut - off frequency and gain coefficient in the current filtering parameters are adjusted to generate corrected filtering parameters.

7. The method according to claim 1, characterized in that, Map the temperature gradient vector to a corresponding energy transfer path, and calibrate the interference intensity between two adjacent energy transfer paths by using the filtering parameter, including: Take each vector element in the temperature gradient vector as a starting point, and perform path tracking along the vector direction to generate multiple energy transfer paths; Calculate the overlapping area between two adjacent energy transfer paths, and quantify the overlapping area as the initial interference intensity; Use the cut-off frequency and gain coefficient in the filtering parameter to scale and adjust the initial interference intensity to determine the calibrated interference intensity, where the process of the scaling adjustment includes: filtering out high-frequency interference components by using the cut-off frequency, and amplifying or attenuating the amplitude of the initial interference intensity by using the gain coefficient.

8. An intelligent pod control system for an unmanned aerial vehicle, characterized in that, Include: A first generation module, configured to obtain multiple frames of thermal radiation images of the unmanned aerial vehicle through an infrared imaging module in the intelligent pod, perform heat source decoupling processing on the multiple frames of thermal radiation images, and generate a dynamic temperature field map of each heat source; An analysis module, configured to analyze the spatial correlation between heat sources by using a graph neural network based on the temperature gradient vectors of the dynamic temperature field maps of all the heat sources; A correction module, configured to generate a damping torque matching the wind disturbance spectrum according to the spatial correlation between heat sources, and correct the filtering parameter of the magneto-rheological damping gimbal in the intelligent pod according to the frequency response characteristics of the damping torque; A calibration module, configured to map the temperature gradient vector to a corresponding energy transfer path, and calibrate the interference intensity between two adjacent energy transfer paths by using the filtering parameter; A second generation module, configured to extract heat diffusion features from the temperature gradient vector, and generate a threat propagation path according to the calibrated interference intensity and the heat diffusion features; A planning module, configured to dynamically plan the deflection angle of the intelligent pod according to the threat propagation path, so that the center of the optical field of view is spatially synchronized with the geometric centroid of the threat propagation path.

9. A computing device, characterized in that, Include a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the method for controlling the intelligent pod of the unmanned aerial vehicle according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, Store a computer program, and when the computer program is executed by a computer, implement the method for controlling the intelligent pod of the unmanned aerial vehicle according to any one of claims 1 to 7.

Images