An air-ground integrated forest fire situation intelligent identification method

Abstract

This application provides an integrated ground-air intelligent forest fire identification method, comprising: determining the bright boundary of the ground fire source captured by the thermal imaging lens based on the degree of downwash airflow disturbance and the smoke outline integrity score; assessing the need for lowering the hovering position of the UAV for close-range reconnaissance; determining the lower limit of the corrected hovering height; analyzing the reliability level of smoke identification; combining the bright boundary of the ground fire source to establish the synergistic guarantee effect of precise thermal imaging positioning and visible light smoke identification; extracting the hovering height node that resists downwash airflow interference; updating the UAV flight altitude adjustment scheme based on the hovering height node that resists downwash airflow interference; continuously monitoring the evolution of smoke morphology in the visible light field of view; assessing the continuous and stable performance of the synergistic guarantee effect; and establishing the final fire verification conclusion.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an integrated ground-air intelligent identification method for forest fires. Background Technology

[0002] In the field of integrated ground-air forest fire intelligent identification, the collaborative monitoring of drones and satellite remote sensing has become an important technical means to improve the early warning capability of forest fires. An existing patent, "A Method for Forest Fire Detection and Prevention Based on Drones," publication number CN104143248B, discloses a technique that combines visible light images with thermal infrared images. It uses the thermal difference of high-brightness points in the image to determine forest fire risk and deploy fire extinguishing packs. However, this technique suffers from a problem: when the drone lowers its altitude for thermal imaging positioning, the downwash airflow disturbs the smoke column shape, causing the loss of smoke characteristics in the visible light channel. This field is directly related to the timeliness of fire detection, the accuracy of location, and the control of disaster losses, and is of crucial significance for protecting forest resources and the safety of people's lives and property.

[0003] Most current drone-based fire detection solutions tend to lower their flight altitude as much as possible after detecting a suspected fire, hoping to obtain higher-resolution ground thermal images through thermal imaging cameras and thus more accurately pinpoint the fire source. However, this conventional practice of lowering altitude often leads to a significant loss of another crucial piece of information in actual forest environments: the clear and discernible characteristics of smoke in the visible light channel. When the drone hovers too low, the downdraft generated by the high-speed rotation of the rotor strongly disturbs the airflow near the ground. The smoke column that would normally rise naturally from the fire source is quickly dispersed, and the originally concentrated and clearly defined smoke rapidly becomes a thin, irregular, drifting mist, or even dissipates completely within a short period. This phenomenon severely damages the outline shape, grayscale gradation, and contrast with the background of the smoke in the visible light image, rendering the criteria for fire confirmation based on smoke morphology ineffective.

[0004] While thermal imaging can identify high-temperature points, forest surfaces are often subject to interference from spontaneous combustion of dead branches and leaves, heat sources from animal activity, and sunlight reflection, making it difficult to eliminate false alarms using thermal imaging alone. Furthermore, smoke, as the most direct and specific external indicator of fire, becomes unreliable due to downwash interference, significantly reducing the reliability of the entire verification process. Lowering the altitude, intended to enhance the positioning accuracy of thermal imaging, directly weakens the effectiveness of the visible light smoke channel, creating a clear mutual constraint between the two.

[0005] How to ensure that thermal imaging images have sufficient ground resolution to reliably locate fire sources when drones approach to verify suspected fire points, while keeping the disturbance of the surface smoke column morphology by the rotor downwash airflow within an acceptable range and keeping the smoke characteristics in the visible light channel sufficiently intact, so that both thermal imaging positioning and smoke identification information channels can function simultaneously, has become a key problem that urgently needs to be solved in intelligent forest fire identification. Summary of the Invention

[0006] This invention provides an integrated ground-air intelligent forest fire identification method, comprising: The real-time hovering altitude of the UAV and the wind pressure feedback value collected by the air pressure sensing unit are obtained. The real-time hovering altitude is compared with a preset safe hovering altitude threshold to determine the initial lower limit of the hovering altitude. The wind pressure feedback value is compared with a preset surface wind speed warning threshold to obtain the surface wind speed estimation result. The initial hovering height limit and the estimated surface wind speed are fused with the image acquired by the visible light channel using particle filtering to evaluate the dissipation and damage performance and attenuation characteristics of the smoke boundary, and to determine the smoke outline integrity score and the degree of downwash airflow disturbance. Based on the degree of downwash airflow disturbance and the smoke contour integrity score, the bright boundary of the ground fire source captured by the thermal imaging lens is determined, the need to lower the hovering position of the UAV for close reconnaissance is assessed, and the lower limit of the corrected hovering height is determined. By using the corrected lower limit of hovering height and combining it with the real-time wind shear data obtained by the wind shear sensing unit, the boundary gradient value and grayscale attenuation feature of the smoke region in the visible light channel are extracted as smoke visual criteria. The smoke visual criteria are compared with the preset smoke identification reliability classification threshold to identify the smoke identification reliability level. The reliability level of the smoke identification was analyzed, and combined with the bright boundary of the surface fire source, the synergistic guarantee effect of precise positioning of thermal imaging and visible light smoke identification was established, and the hovering height node that resists the interference of downwash wind field was extracted. Based on the hovering altitude node that resists downwash wind field interference, update the UAV flight altitude adjustment scheme, continuously monitor the evolution of smoke morphology in the visible light field, evaluate the continuous and stable performance of collaborative protection effectiveness, and establish the final fire verification conclusion.

[0007] Preferably, the initial hovering height lower limit and the estimated surface wind speed are fused with the image acquired through the visible light channel using particle filtering to evaluate the dissipation and attenuation characteristics of the smoke boundary, determine the smoke outline integrity score and the degree of downwash airflow disturbance, including: The initial hovering height lower limit and the ground wind speed estimation result are obtained. The vertical influence range of the rotor downwash airflow is determined based on the initial hovering height lower limit. The horizontal diffusion amplitude of the downwash airflow is determined based on the ground wind speed estimation result. The vertical influence range and the horizontal diffusion amplitude are combined to obtain the airflow state description quantity. Acquire the current frame image of the visible light channel, perform grayscale processing on the current frame image to obtain a grayscale image, extract the contour boundary of the smoke region in the grayscale image by edge detection, mark the position coordinate sequence of the smoke boundary pixel points along the contour boundary, and obtain the smoke boundary distribution data. Based on the airflow state description and the smoke boundary distribution data, the state tracking of the smoke boundary pixels is performed using a particle filtering method. The airflow state description is used as the external perturbation input for the particle filtering, and the position offset of the smoke boundary pixels between adjacent frames is used as the observation input. The fused smoke boundary tracking result is obtained through particle weight update and resampling. Based on the fused smoke boundary tracking results, the brightness attenuation of smoke boundary pixels between adjacent frames is statistically analyzed. The frame-by-frame decreasing trend of grayscale values ​​of each boundary pixel is used as the attenuation feature. At the same time, the number of broken segments of the contour boundary and the proportion of broken segments to the total length of the contour are statistically analyzed. The smoke contour integrity score is determined based on the relationship between the proportion and the preset contour damage threshold. The degree of downwash airflow disturbance is determined based on the correspondence between the horizontal diffusion amplitude and the average offset of smoke boundary pixels.

[0008] Preferably, the step of acquiring the real-time hovering altitude of the UAV and the wind pressure feedback value collected by the air pressure sensing unit, comparing the real-time hovering altitude with a preset safe hovering altitude threshold to determine the initial lower limit of the hovering altitude, and comparing the wind pressure feedback value with a preset surface wind speed warning threshold to obtain the surface wind speed estimation result includes: The drone's real-time hovering altitude data and the wind pressure feedback value output by the air pressure sensing unit are obtained. The real-time hovering altitude is compared with a preset safe hovering altitude threshold, and the initial lower limit of hovering altitude is determined based on the comparison result. The rotor speed corresponding to the initial hovering height lower limit is determined by a preset altitude-speed mapping function, and the preset surface wind speed warning threshold is adjusted according to the rotor speed. The adjusted preset surface wind speed warning threshold is compared with the wind pressure feedback value, and the intensity of surface airflow activity is judged based on the comparison result to obtain the surface wind speed estimation result.

[0009] Preferably, the step of determining the bright boundary of the surface fire source captured by the thermal imaging lens based on the degree of downwash airflow disturbance and the smoke outline integrity score, assessing the need to lower the hovering position of the UAV for close-range reconnaissance, and determining the lower limit of the corrected hovering altitude includes: The degree of turbulence of the downwash airflow and the integrity score of the smoke outline are obtained. The current frame infrared image captured by the thermal imaging lens is retrieved. The bright area of ​​the ground fire source is identified in the infrared image. The coverage range and edge sharpness of the bright boundary of the ground fire source are determined according to the brightness gradient change and temperature difference contrast intensity of the high temperature area. The hovering position adjustment requirement is assessed based on the edge sharpness of the bright boundary of the surface fire source and the smoke outline integrity score. The assessment results determine whether there is a need to adjust the current hovering position. Based on the lowering requirement and the degree of airflow disturbance, the height is corrected on the basis of the initial lower limit of hovering height to obtain the corrected lower limit of hovering height.

[0010] Preferably, the step of retrieving the current frame infrared image captured by the thermal imaging lens, identifying the bright area of ​​the ground fire source in the infrared image, and determining the coverage range and edge sharpness of the bright boundary of the ground fire source based on the brightness gradient change and temperature difference contrast intensity of the high-temperature area includes: The degree of turbulence of the downwash airflow and the integrity score of the smoke outline are retrieved to obtain an infrared image in the current field of view of the thermal imaging lens. In the infrared image, a set of pixels whose temperature values ​​exceed a preset fire source temperature threshold are selected and marked as a high-temperature region to obtain the pixel position distribution of the high-temperature region. Based on the pixel position distribution in the high-temperature region, the temperature difference between adjacent pixels is calculated along the horizontal and vertical directions respectively. The change range of the temperature difference in each direction is used as the brightness gradient. The brightness gradient is corrected according to the degree of disturbance to obtain the corrected brightness gradient change characteristics. Based on the corrected brightness gradient change characteristics, edge pixels of the fire source are extracted at the edge of the high-temperature region. The temperature difference between the edge pixels of the fire source and the adjacent low-temperature background pixels is calculated as the temperature difference contrast intensity. The coverage range and edge sharpness of the high-brightness boundary of the fire source on the ground are determined based on the temperature difference contrast intensity distribution.

[0011] Preferably, by using the corrected lower limit of hovering height, combined with real-time wind shear data obtained by the wind shear sensing unit, the boundary gradient value and grayscale attenuation features of the smoke region in the visible light channel are extracted as smoke visual criteria. The smoke visual criteria are compared with a preset smoke identification reliability grading threshold to identify the smoke identification reliability level, including: The corrected lower limit of hovering height is obtained, real-time wind shear data is read from the wind shear sensing unit, and the degree of airflow turbulence is determined by comparing the real-time wind shear data with the preset wind shear threshold. Based on the degree of airflow turbulence, the current frame image acquired by the visible light channel is retrieved, and the smoke region is located in the current frame image. The gray-level difference between adjacent pixels is extracted along the edge of the smoke region as a boundary gradient value. The boundary gradient value is compensated according to the degree of airflow turbulence. The decrease magnitude and deceleration rate of gray-level values ​​from the edge to the center of the smoke region are statistically analyzed as gray-level attenuation features. The boundary gradient value and the gray-level attenuation features are combined to form a visual criterion for smoke. The smoke recognition reliability level is identified by comparing the smoke visual criteria with a preset smoke recognition reliability grading threshold.

[0012] Preferably, the reliability level of the smoke identification is analyzed, and combined with the high-brightness boundary of the surface fire source, the synergistic guarantee effect of precise thermal imaging positioning and visible light smoke identification is established, and the hovering height node that resists downwash wind field interference is extracted, including: The reliability level of smoke identification and the coverage range and edge sharpness of the bright boundary of the ground fire source are obtained. Based on the reliability level of smoke identification, it is determined whether the smoke characteristics of the visible light channel meet the identification requirements. Based on the edge sharpness, it is determined whether the fire source positioning of the thermal imaging channel meets the accuracy requirements, and a dual-channel collaborative state description is obtained. Based on the dual-channel collaborative state description, it is determined whether the collaborative guarantee effect of thermal imaging precise positioning and visible light smoke identification meets the standard, and the collaborative guarantee effect judgment result is obtained. Based on the collaborative support effectiveness determination result and the revised lower limit of hovering height, when the collaborative support effectiveness meets the standard, the current hovering height is recorded as the hovering height node for resisting downwash wind field interference. When the collaborative support effectiveness does not meet the standard, a preset height increment is added successively based on the revised lower limit of hovering height and the collaborative support effectiveness is re-evaluated. The hovering height node for resisting downwash wind field interference is determined based on the re-evaluation result.

[0013] Preferably, based on the hovering altitude node that resists downwash wind field interference, the UAV flight altitude adjustment scheme is updated, and the evolution of smoke morphology in the visible light field is continuously monitored to assess the sustained and stable performance of collaborative support effectiveness, and to establish the final fire verification conclusion, including: The node height value of the hovering height node is used as the basis for updating the UAV flight altitude adjustment scheme and written into the flight control command to update the hovering height reference. Visible light channel images are continuously acquired at the hovering height reference position. The contour integrity and grayscale distribution of the smoke region in the image are monitored and the changes in the smoke morphology evolution are recorded in continuous time frames to obtain a smoke morphology evolution monitoring record. Based on the smoke morphology evolution monitoring records, the integrity of the smoke outline and the magnitude of grayscale distribution changes in consecutive multi-frame images are determined. Based on the determination results, the continuous and stable performance of the collaborative protection effectiveness is evaluated. Combined with the location information of the bright boundary of the surface fire source, the final fire verification conclusion is established.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an integrated ground-air intelligent forest fire identification method. Addressing the complex needs of optimizing drone hovering altitude, identifying smoke boundaries, and accurately locating fire sources in forest fire scenarios, it proposes a logically interconnected solution. This invention acquires real-time drone hovering altitude and wind pressure feedback values, compares them with preset thresholds, and combines visible light image particle filtering fusion to assess the integrity of the smoke outline and downwash airflow disturbance, thereby correcting the lower limit of hovering altitude. Simultaneously, it utilizes thermal imaging to analyze the bright boundaries of fire sources and temperature differences, coordinating the reliability level of visible light smoke identification to establish hovering altitude nodes resistant to wind interference. Finally, it updates the flight altitude adjustment scheme and continuously monitors the evolution of the fire. The core innovation of this invention lies in achieving accurate identification and stable reconnaissance of smoke and fire sources through multi-source data fusion and dynamic altitude adjustment, significantly improving the accuracy and security of forest fire verification. Attached Figure Description

[0015] Figure 1 This is a flowchart of an integrated ground-air intelligent forest fire identification method according to the present invention.

[0016] Figure 2 This is a schematic diagram of an integrated ground-air intelligent forest fire identification method according to the present invention.

[0017] Figure 3 This is another schematic diagram of an integrated ground-air intelligent forest fire identification method according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0019] like Figures 1-3 This embodiment of the integrated ground-air forest fire intelligent identification method may specifically include: S101. Obtain the real-time hovering altitude of the UAV and the wind pressure feedback value collected by the air pressure sensing unit. By comparing the real-time hovering altitude with the preset safe hovering altitude threshold and the wind pressure feedback value with the preset surface wind speed warning threshold, the initial hovering altitude lower limit and the surface wind speed estimation result are extracted.

[0020] The system acquires real-time hovering altitude data of the UAV and wind pressure feedback values ​​output by the air pressure sensing unit. The real-time hovering altitude is compared with a preset safe hovering altitude threshold. If the real-time hovering altitude is lower than the preset safe hovering altitude threshold, the preset safe hovering altitude threshold is used as the initial lower limit of the hovering altitude. If the real-time hovering altitude is higher than or equal to the preset safe hovering altitude threshold, the current real-time hovering altitude is used as the initial lower limit of the hovering altitude. A preset altitude-speed mapping function is used to determine the rotor speed corresponding to the initial lower limit of the hovering altitude, and a preset surface wind speed warning threshold is adjusted accordingly for comparison with the wind pressure feedback value. If the wind pressure feedback value exceeds the adjusted preset surface wind speed warning threshold, it is determined that there is strong airflow activity on the surface. This is then assessed using a formula based on Bernoulli's principle. The surface wind speed estimation result is obtained by calculation, where v is the estimated wind speed, P is the wind pressure feedback value, and ρ is the air density constant.

[0021] In one implementation, the UAV collects current hovering altitude data in real time through an onboard altitude sensor. This altitude data is output in the form of vertical distance relative to the ground. Simultaneously, the air pressure sensing unit collects wind pressure feedback values ​​in the area below the rotor. These wind pressure feedback values ​​reflect the dynamic pressure intensity formed near the ground by the downwash airflow generated by the rotor rotation.

[0022] Specifically, when comparing the real-time hovering height with the preset safe hovering height threshold, if the real-time hovering height is lower than the threshold, it indicates that the drone is too close to the ground. In this case, the preset safe hovering height threshold is directly used as the initial lower limit of the hovering height. If the real-time hovering height is higher than or equal to the threshold, the current height is maintained as the initial lower limit of the hovering height. This judgment logic is used to constrain the minimum safe flight boundary of the drone when checking the fire situation.

[0023] Specifically, the altitude-rotor speed mapping function describes the correspondence between the hovering altitude of the UAV and the rotor speed required to maintain hovering. When hovering in low-altitude areas, the UAV needs a higher rotor speed to resist the airflow turbulence caused by the ground effect. This mapping relationship can be expressed as n=n0×(1+k / h), where n is the rotor speed corresponding to the current altitude, n0 is the reference speed at the standard altitude, h is the lower limit of the initial hovering altitude, and k is the ground effect correction coefficient. The rotor speed determined in this way is used to correct the ground wind speed warning threshold. The corrected warning threshold Pth'=Pth×(n / n0)², where Pth is the original preset ground wind speed warning threshold and Pth' is the corrected threshold.

[0024] It should be noted that surface wind speed estimation is based on Bernoulli's principle, where the wind pressure feedback value is proportional to the square of the airflow velocity. Under standard atmospheric conditions, the air density constant is approximately 1.29 kg / m³ (the standard value is often taken as 1.225 kg / m³). Dividing the wind pressure feedback value by this density constant yields the square of the velocity value, and then taking the square root gives the surface wind speed estimation result. This result is used to characterize the potential disturbance intensity of the rotor downwash airflow on the surface smoke column.

[0025] S102. The initial hovering height lower limit and the estimated surface wind speed are fused with the image acquired in the visible light channel using particle filtering to evaluate the dissipation and damage of the smoke boundary in the image and its attenuation characteristics, and to determine the smoke outline integrity score and the degree of downwash airflow disturbance.

[0026] The initial hovering height lower limit and the estimated surface wind speed are obtained. The vertical influence range of the rotor downwash airflow is determined based on the initial hovering height lower limit, and the horizontal diffusion amplitude of the downwash airflow is determined based on the estimated surface wind speed. The vertical influence range and the horizontal diffusion amplitude are combined to obtain the airflow state description. The current frame image in the visible light channel is acquired and converted to grayscale to obtain a grayscale image. The contour boundary of the smoke region is extracted in the grayscale image using edge detection. The position coordinate sequence of the smoke boundary pixels is marked along the contour boundary to obtain smoke boundary distribution data. Based on the airflow state description and the smoke boundary distribution data, a particle filter method is used to track the state of the smoke boundary pixels. The airflow state description is used as the external perturbation input for the particle filter, and the position offset of the smoke boundary pixels between adjacent frames is used as the observation input. The fused smoke boundary tracking result is obtained through particle weight update and resampling. Based on the fused smoke boundary tracking results, the brightness attenuation of smoke boundary pixels between adjacent frames is statistically analyzed. The frame-by-frame decreasing trend of grayscale values ​​of each boundary pixel is used as the attenuation feature. At the same time, the number of broken segments of the contour boundary and the proportion of broken segments to the total length of the contour are statistically analyzed. If the proportion exceeds a preset contour damage threshold, the smoke contour integrity score is determined to be low. If the proportion is lower than the preset contour damage threshold, the smoke contour integrity score is determined to be high. Meanwhile, the degree of downwash airflow disturbance is determined based on the correspondence between the horizontal diffusion amplitude and the average offset of smoke boundary pixels.

[0027] In one implementation, when the UAV hovers to check a suspected fire point, it determines the vertical influence range of the rotor downwash airflow based on the initial lower limit of the hovering altitude. This vertical influence range characterizes the distance the downwash airflow travels from the rotor plane down to the ground surface. The lower the hovering altitude, the greater the airflow pressure within the vertical influence range.

[0028] Specifically, the surface wind speed estimation results reflect the velocity and intensity of the downwash airflow spreading outwards after reaching the surface. Based on this, the horizontal diffusion range of the downwash airflow is determined. This horizontal diffusion range describes the radius of influence of the airflow in the horizontal direction on the surface. Combining the vertical influence range with the horizontal diffusion range forms an airflow state descriptive quantity. This descriptive quantity, in two-dimensional numerical form, characterizes the comprehensive effect of the UAV's current hovering attitude on the surface airflow field. Visible light image acquisition is performed by the visible light camera onboard the UAV. After acquiring the current frame image containing the smoke area, the color image is converted to grayscale. The pixel values ​​of the red, green, and blue channels are converted into single-channel grayscale values ​​using a weighted average method to obtain a grayscale image.

[0029] It should be noted that when extracting the contour boundary of the smoke region in the grayscale image, the Sobel operator is used to perform edge detection on the grayscale image. The Sobel operator calculates the gradient magnitude by calculating the grayscale gradient values ​​Gx and Gy of each pixel in the horizontal and vertical directions. Where Gx is the horizontal gradient, Gy is the vertical gradient, and G is the comprehensive amplitude, pixels whose G exceeds a preset edge threshold are marked as edge points. This threshold is dynamically set to 1.5 times the average gray value of the image. For example, when the average is 100, the threshold is 150. There is a significant gray value difference between the smoke area and the background canopy. The continuous edge points formed after edge detection constitute the contour boundary of the smoke. The row and column coordinates of each boundary pixel are recorded sequentially along the contour boundary to form the smoke boundary distribution data.

[0030] In one embodiment, the particle filtering method is a recursive Bayesian estimation method based on Monte Carlo sampling. Its core idea is to use a set of weighted random sampled particles to approximate the probability distribution of the target state. In the smoke boundary tracking scenario, each particle represents the position hypothesis of the smoke boundary pixel in the next moment. Particle filtering includes two alternately executed stages: prediction and update. In the prediction stage, the position of each particle is predicted according to the state transition model. In the update stage, the weight of each particle is adjusted according to the actual observation. The higher the weight, the more the position hypothesis represented by the particle matches the actual observation. After multiple iterations, the estimated value of the target state is obtained by weighted summation or resampling.

[0031] For example, when particle filtering is applied to smoke boundary tracking, the airflow state description is introduced as an external disturbance input into the prediction stage of particle filtering.

[0032] Specifically, the horizontal diffusion range in the airflow state descriptor determines the potential horizontal displacement range of smoke boundary pixels, while the vertical influence range determines the degree to which the smoke's upward movement is obstructed. During the prediction phase, when predicting the position of each particle, the airflow state descriptor is transformed into a probability distribution parameter for positional offset, causing the predicted position of the particle to shift accordingly towards the airflow diffusion direction. In the update phase, the actual positional offset of smoke boundary pixels between adjacent frames is used as the observation input to calculate the distance error between the predicted and actual observed positions of each particle, thereby adjusting the particle weights. Based on this particle filtering process, particles with lower weights are eliminated during the resampling phase, while particles with higher weights are copied. The set of particles retained after weight updates and resampling represents the smoke boundary position estimate after fusing airflow disturbance information. The position coordinates of each retained particle are weighted and averaged to obtain the fused smoke boundary tracking result.

[0033] Understandably, when scoring the contour integrity based on the fused smoke boundary tracking results, the number of broken line segments in the contour boundary and the sum of the pixel lengths of each broken segment are counted. The proportion of the total length of the broken segments to the original total length of the contour boundary is calculated. If this proportion exceeds a preset contour damage threshold, it indicates that the smoke contour has been significantly broken due to the downwash airflow, and the smoke contour integrity score is determined to be low. If this proportion is lower than the preset contour damage threshold, the smoke contour integrity score is determined to be high. The quantization of the attenuation feature is obtained by comparing the gray value changes of smoke boundary pixels frame by frame. The difference in gray value of the corresponding boundary pixels in frame t and frame t-1 is calculated, and the average of the gray value differences of all boundary pixels is taken. This average value is the attenuation feature value of the current frame. The larger the attenuation feature value, the faster the smoke boundary is dissipating. Furthermore, the determination of the downwash airflow disturbance level is based on the correspondence between the horizontal diffusion amplitude and the average offset of the smoke boundary pixels. The average offset is obtained by calculating the average of the position changes of all smoke boundary pixels between adjacent frames. When the horizontal diffusion amplitude is large and the average offset increases synchronously, it indicates that the downwash airflow has significantly disturbed the smoke morphology. Accordingly, the downwash airflow disturbance level is judged as high. When both the horizontal diffusion amplitude and the average offset are in a small range, the downwash airflow disturbance level is judged as low. This disturbance level characterizes the interference intensity of the rotor airflow on the smoke features of the visible light channel at the current hovering height.

[0034] S103. Determine the bright boundary of the ground fire source captured by the thermal imaging lens based on the degree of downwash airflow disturbance and the smoke outline integrity score, assess the need to lower the hovering position of the UAV for close reconnaissance, and determine the lower limit of the corrected hovering height.

[0035] The degree of downwash airflow disturbance and the smoke outline integrity score are obtained. The current frame infrared image captured by the thermal imaging lens is retrieved, and the bright areas of the surface fire source are identified in the infrared image. The coverage range and edge sharpness of the bright boundary of the surface fire source are determined by analyzing the brightness gradient change and temperature difference contrast intensity of the high-temperature area. Based on the edge sharpness of the bright boundary of the surface fire source and the smoke outline integrity score, the hovering position adjustment requirement is assessed. If the smoke outline integrity score is high and the edge sharpness exceeds a preset clarity threshold, the current hovering position is determined not to need adjustment. If the smoke outline integrity score is low or the edge sharpness is lower than the preset clarity threshold, the current hovering position needs adjustment. Based on the adjustment requirement and the degree of downwash airflow disturbance, the height is corrected based on the initial lower limit of the hovering height. If the disturbance level is high and there is a need for adjustment, the initial lower limit of the hovering height remains unchanged. If the disturbance level is low and there is a need for adjustment, the initial lower limit of the hovering height is adjusted downwards by a preset height adjustment amount to obtain the corrected lower limit of the hovering height.

[0036] In one implementation, the thermal imaging lens on the drone collects infrared images of the ground area in real time. Each pixel in the infrared image corresponds to a temperature value. By traversing all pixels and comparing them with a preset fire source temperature threshold, the set of pixels with temperature values ​​exceeding the threshold is marked as a bright area.

[0037] It should be noted that the extraction of the bright boundary of the ground fire source is based on the temperature transition characteristics between the bright area and the surrounding low-temperature background. The process involves scanning pixel by pixel along the outer edge of the bright area. When the temperature value between two adjacent pixels drops sharply, the location is marked as a boundary point. All boundary points are connected to form the bright boundary of the ground fire source. The edge sharpness index is defined as the average value of the temperature difference between the pixels inside and outside the boundary point. The larger the temperature difference, the more obvious the thermal contrast between the fire source area and the background, and the higher the edge sharpness.

[0038] Specifically, the quantification of edge sharpness is achieved by statistically analyzing the difference between the inner and outer pixel temperatures of all boundary points on the highlight boundary, and then taking the arithmetic mean of these differences. This average temperature difference serves as the numerical representation of edge sharpness.

[0039] In one embodiment, the assessment of the hovering position reduction requirement adopts a dual-condition joint determination method. When the smoke outline integrity score is high and the edge sharpness exceeds a preset clarity threshold, it indicates that the smoke features of the visible light channel are intact and the fire source positioning accuracy of the thermal imaging channel meets the requirements, and it is determined that the current hovering position does not need to be reduced. When the smoke outline integrity score is low or the edge sharpness is lower than the preset clarity threshold, it indicates that the information quality of at least one channel is insufficient, and it is determined that there is a need for reduction. Furthermore, the height correction process comprehensively considers the combination of the reduction requirement and the degree of downwash airflow disturbance. If the disturbance degree is high and there is a need for reduction, it means that continuing to reduce the height will exacerbate the destruction of smoke features. In this case, the initial hovering height lower limit is maintained unchanged to protect the visible light channel information.

[0040] Understandably, if the disturbance level is low and there is a need for downward adjustment, it means that the current airflow has a weak interference with the smoke and there is still room for improvement in the thermal imaging channel. In this case, the initial hovering height lower limit is adjusted downward by the preset height adjustment amount to obtain the corrected hovering height lower limit. This corrected hovering height lower limit serves as the height control boundary for UAVs during close-range reconnaissance.

[0041] By retrieving the downwash airflow disturbance level and smoke outline integrity score, analyzing the brightness gradient changes and spatial distribution differences in the high-temperature area within the field of view of the thermal imaging lens, identifying the temperature difference contrast intensity between the edge of the surface fire source and the surrounding low-temperature background, and determining the coverage range and edge sharpness of the high-brightness boundary of the surface fire source.

[0042] The degree of downwash airflow disturbance and the smoke outline integrity score are retrieved to obtain an infrared image within the current field of view of the thermal imaging lens. A set of pixels whose temperature values ​​exceed a preset fire source temperature threshold is selected from the infrared image and marked as a high-temperature region, thus obtaining the pixel location distribution within the high-temperature region. Based on the pixel location distribution within the high-temperature region, the temperature difference between adjacent pixels is calculated along both the horizontal and vertical directions. The magnitude of the temperature difference in each direction is used as the brightness gradient. The numerical difference of the brightness gradient within the high-temperature region and at its edge is statistically analyzed. The influence of the disturbance degree on the attenuation of the brightness gradient is then corrected using the formula G'=G×(1-0.5×D), where G is the original brightness gradient, D is the disturbance degree (between 0 and 1), and 0.5 is the attenuation coefficient. This yields the corrected brightness gradient change characteristics and spatial distribution difference description. Based on the corrected brightness gradient change characteristics and spatial distribution differences, edge pixels of the fire source are extracted at the edge of the high-temperature region. The temperature values ​​of the fire source edge pixels and adjacent low-temperature background pixels are obtained, and the temperature difference between them is calculated as the temperature difference contrast intensity. The temperature difference contrast intensity is weighted and adjusted in conjunction with the contour integrity score to obtain the temperature difference contrast intensity distribution of the surface fire source edge. Based on the temperature difference contrast intensity distribution of the surface fire source edge, the area of ​​the bounding rectangle of the region enclosed by edge pixels whose temperature difference contrast intensity exceeds a preset temperature difference threshold is counted. The area of ​​the bounding rectangle is taken as the coverage range of the high-brightness boundary of the surface fire source, and the average value of the temperature difference contrast intensity distribution is taken as the edge sharpness.

[0043] In one implementation, when the UAV is performing a fire detection mission, it retrieves the downwash airflow disturbance level and smoke outline integrity score obtained from the previous processing stage. These two parameters reflect the intensity of the interference of the rotor airflow on the surface smoke morphology in the current hovering state and the degree of preservation of smoke features in the visible light channel.

[0044] Specifically, after the thermal imaging lens acquires an infrared image within the current field of view, it iterates through the temperature values ​​of all pixels in the image and compares them with a preset fire source temperature threshold. Pixels with temperature values ​​exceeding the threshold are selected and marked as high-temperature areas, which represent the location range of a significant heat source on the ground.

[0045] It should be noted that the brightness gradient is an indicator describing the degree of temperature change between adjacent pixels in a thermal imaging image. In an infrared image, the temperature difference between each pair of adjacent pixels is calculated along both the horizontal and vertical directions. The horizontal temperature difference of a pixel is the difference between the pixel's temperature and the temperature of its right-hand neighbor, while the vertical temperature difference is the difference between the pixel's temperature and the temperature of its bottom neighbor. The sum of the absolute values ​​of the horizontal and vertical temperature differences is taken as the brightness gradient value of that pixel. The brightness gradient value inside a high-temperature region is usually relatively gentle, while the brightness gradient value at the edge of a high-temperature region exhibits a sharp change.

[0046] In one embodiment, the correction of the brightness gradient by the degree of disturbance is reflected in the fact that when the disturbance level of the current washing airflow is high, the heat distribution of the surface heat source will be locally diffused due to the airflow disturbance, causing the originally clear temperature boundary to become blurred, and the brightness gradient value will decrease accordingly. By introducing the degree of disturbance as a correction factor, the original brightness gradient value is compensated and corrected. When the disturbance level is high, the correction amount of the brightness gradient is increased, and when the disturbance level is low, the correction amount is decreased, resulting in a description of the corrected brightness gradient change characteristics and spatial distribution differences. Based on the above-mentioned corrected brightness gradient change characteristics, pixels with brightness gradient values ​​exceeding a preset edge gradient threshold are identified at the edge of the high-temperature region. These pixels are extracted as fire source edge pixels, which constitute the thermal boundary between the fire source and the surrounding environment.

[0047] For example, the temperature difference contrast intensity is calculated for each edge pixel of the fire source. The temperature value of the edge pixel and the temperature value of the adjacent pixel in the low-temperature background region are obtained, and the temperature difference between the two is the temperature difference contrast intensity at the edge pixel. The weighted adjustment of the contour integrity score is achieved by defining the weighted adjustment formula of the temperature difference contrast intensity as ΔT'=ΔT×α, where ΔT is the original temperature difference contrast intensity, and α is the confidence weight coefficient determined based on the contour integrity score. When the contour integrity score is high, the value of α is close to 1, indicating a high confidence in the temperature difference contrast intensity. When the contour integrity score is low, the value of α decreases to reflect the decrease in the consistency between the visible light channel and the thermal imaging channel information. After weighted adjustment, the temperature difference contrast intensity distribution of the edge of the ground fire source is obtained. Further, the coverage area is determined based on the spatial distribution of the edge pixels of the fire source. All edge pixels with temperature difference contrast intensities exceeding a preset temperature difference threshold are counted, and the area of ​​the bounding rectangle of the region enclosed by these pixels is calculated. This bounding rectangle area represents the coverage area of ​​the bright boundary of the ground fire source.

[0048] Understandably, edge sharpness reflects the clarity of thermal contrast between the fire source area and the surrounding low-temperature background. By taking the arithmetic mean of the temperature difference contrast intensity values ​​of all fire source edge pixels, the average value of the temperature difference contrast intensity distribution is obtained. This average value serves as a quantitative indicator of edge sharpness. The higher the edge sharpness, the clearer and more distinguishable the fire source boundary.

[0049] Preferably, the coverage area and edge sharpness together constitute a characteristic description of the bright boundary of the ground fire source. The coverage area represents the spatial scale of the fire source, and the edge sharpness represents the accuracy of the fire source location. The combination of the two provides a criterion for further adjustment of the drone's hovering altitude through the thermal imaging channel.

[0050] S104. By using the corrected lower limit of hovering height and combining the real-time wind shear data obtained by the wind shear sensing unit, the boundary gradient value and grayscale attenuation feature of the smoke region in the visible light channel are extracted as smoke visual criteria. The smoke visual criteria are compared with the preset smoke identification reliability classification threshold to identify the smoke identification reliability level.

[0051] The corrected lower limit of hovering height is obtained, and real-time wind shear data at the current moment is read from the wind shear sensing unit. The wind shear data represents the variation of wind speed and direction between different height layers. If the wind shear data exceeds a preset wind shear threshold, the current airflow turbulence level is determined to be high; if the wind shear data is below the preset wind shear threshold, the current airflow turbulence level is determined to be low. Based on the airflow turbulence level, the current frame image acquired by the visible light channel is retrieved, and the smoke region is located in the current frame image. The gray-level difference between adjacent pixels along the edge of the smoke region is extracted as the boundary gradient value. When the airflow turbulence level is high, the boundary gradient value is attenuated and compensated. The decrease magnitude and deceleration rate of the gray-level value are statistically analyzed pixel by pixel along the smoke region from the edge to the center as gray-level attenuation features. The boundary gradient value and gray-level attenuation features are combined to form a smoke visual criterion. The smoke recognition reliability level is determined to be high if the boundary gradient value exceeds the preset boundary clarity threshold and the grayscale attenuation feature meets the preset attenuation gradient range, and if the boundary gradient value is lower than the preset boundary clarity threshold or the grayscale attenuation feature deviates from the preset attenuation gradient range.

[0052] In one implementation, when the UAV performs fire verification at the lower limit of the corrected hovering altitude, the wind shear sensing unit uses a combination of ultrasonic waves and barometers to estimate the wind shear intensity at the current altitude layer in real time. This intensity value reflects the vertical change rate of wind speed between the altitude of the UAV and the ground surface. By comparing the estimated wind shear intensity value with a preset threshold (e.g., 0.1 (m / s) / m), the current airflow turbulence level is determined to be high or low.

[0053] It should be noted that the boundary gradient value is an indicator of the sharpness of the edge of the smoke region in a visible light image. After locating the smoke region, the edge of the smoke region is scanned pixel by pixel, and the gray-level difference between each edge pixel and its adjacent non-smoke region pixel is calculated. The larger the gray-level difference, the more obvious the contrast between the smoke edge and the background, and the clearer the boundary. The boundary gradient value is obtained by averaging the gray-level differences of all edge pixels. This value reflects the recognizability of the smoke outline in the visible light image.

[0054] Specifically, when the airflow turbulence level is determined to be high, it indicates that the downwash airflow has significantly disturbed the smoke morphology, the smoke edges become blurred due to airflow dispersion, and the original boundary gradient value will be low. In this case, attenuation compensation is performed on the boundary gradient value, and the compensation formula is G. c =G0+β×V w G c G0 represents the compensated boundary gradient value, and V represents the original boundary gradient value. w β represents the wind speed variation in the wind shear data, and β is a preset compensation coefficient. This compensation amount is positively correlated with the intensity of airflow turbulence to restore the true clarity of the smoke edge in the absence of airflow interference.

[0055] In one embodiment, the extraction of grayscale attenuation features starts from the edge pixels of the smoke region and statistically analyzes the change in grayscale value pixel by pixel along the direction pointing towards the center of the smoke. The grayscale distribution of the smoke region typically exhibits a gradual decrease from the edge to the center, with higher grayscale values ​​at the edges and lower grayscale values ​​at the center. The magnitude and rate of decrease in grayscale value during this decreasing process are used as grayscale attenuation features. Furthermore, the compensated boundary gradient value is combined with the grayscale attenuation features to form a smoke visual criterion. This criterion comprehensively reflects the edge sharpness and internal grayscale distribution pattern of smoke features in the visible light channel.

[0056] Understandably, the reliability level of smoke recognition is determined based on the joint evaluation of boundary gradient value and grayscale attenuation characteristics. When the boundary gradient value exceeds the preset boundary clarity threshold and the grayscale attenuation characteristics fall within the preset attenuation gradient range, the reliability level of smoke recognition is determined to be high. When the boundary gradient value is lower than the preset boundary clarity threshold or the grayscale attenuation characteristics deviate from the preset attenuation gradient range, the reliability level of smoke recognition is determined to be low.

[0057] S105. Analyze the reliability level of smoke identification, combine it with the bright boundary of the surface fire source, establish the synergistic guarantee effect of precise positioning of thermal imaging and visible light smoke identification, and extract the hovering height node that is resistant to downwash wind field interference.

[0058] The system obtains the smoke identification reliability level and the coverage area and edge sharpness of the bright boundary of the surface fire source. Based on the smoke identification reliability level, it determines whether the smoke characteristics of the visible light channel meet the identification requirements. Based on the edge sharpness, it determines whether the fire source positioning of the thermal imaging channel meets the accuracy requirements, thus obtaining a dual-channel collaborative state description. According to the dual-channel collaborative state description, if the smoke identification reliability level is high and the edge sharpness exceeds a preset sharpness threshold, the collaborative guarantee effect of thermal imaging accurate positioning and visible light smoke identification is determined to be satisfactory. If the smoke identification reliability level is low or the edge sharpness is lower than the preset sharpness threshold, the collaborative guarantee effect is determined to be unsatisfactory, thus obtaining a collaborative guarantee effect determination result. Based on the collaborative support effectiveness determination result and the revised lower limit of hovering height, when the collaborative support effectiveness meets the standard, the current hovering height is recorded as the hovering height node for resisting downwash wind field interference. When the collaborative support effectiveness does not meet the standard, the hovering height is increased by a preset height increment on the basis of the revised lower limit of hovering height and the collaborative support effectiveness is re-evaluated. If the cumulative increase in height reaches the preset maximum adjustment range and still does not meet the standard, the current height is recorded as a compromise. If the standard is met within the maximum adjustment range, the hovering height at the time of meeting the standard is recorded as the hovering height node for resisting downwash wind field interference.

[0059] In one implementation, when performing fire verification, the UAV relies on two information sources simultaneously: the visible light channel and the thermal imaging channel. The dual-channel collaborative state description is used to characterize whether the information quality of these two channels at the current hovering altitude simultaneously meets the basic requirements for fire identification.

[0060] Specifically, the visible light channel reflects the identifiability of smoke features through the smoke identification reliability level. When the smoke identification reliability level is high, it indicates that the boundary and grayscale features of the smoke are clearly visible in the image. The thermal imaging channel reflects the accuracy of fire source location through edge sharpness. When the edge sharpness exceeds the preset sharpness threshold, it indicates that the thermal contrast between the fire source and the background is clear. The information from the two channels corroborates each other, which can improve the accuracy of fire verification.

[0061] It should be noted that the assessment of the collaborative protection effectiveness adopts a dual-condition joint judgment method. Only when the smoke recognition reliability level of the visible light channel is high and the edge sharpness of the thermal imaging channel exceeds the preset sharpness threshold is the collaborative protection effectiveness deemed to be up to standard. This indicates that both channels are in good working condition. When the information quality of either channel does not meet the requirements, i.e., the smoke recognition reliability level is low or the edge sharpness is lower than the preset sharpness threshold, the collaborative protection effectiveness is deemed to be down to standard. This indicates that the information of one channel is damaged at the current hovering height.

[0062] In one embodiment, when the collaborative protection effectiveness is determined to be satisfactory, the current hovering altitude of the UAV is directly recorded as a hovering altitude node that resists downwash wind interference. This altitude node represents the position at which the UAV can safely approach while maintaining the quality of dual-channel information. Further, when the collaborative protection effectiveness is determined to be unsatisfactory, it indicates that the current hovering altitude is too low, causing excessive interference of the downwash airflow with the smoke morphology or insufficient thermal imaging resolution. In this case, a preset altitude increment is added to the corrected lower limit of the hovering altitude to raise the UAV to a new altitude position. Dual-channel images are then re-acquired at the new altitude position, and the collaborative protection effectiveness is evaluated. This process is repeated until the collaborative protection effectiveness is satisfactory.

[0063] Understandably, the hovering altitude node for resisting downwash wind interference is a safe altitude position determined after iterative verification. At this altitude, the interference of the downwash airflow generated by the UAV rotor on the surface smoke column shape is within an acceptable range, while the thermal imaging lens can obtain sufficiently clear information about the fire source boundary.

[0064] S106. Based on the hovering altitude node that resists downwash wind field interference, update the UAV flight altitude adjustment scheme, continuously monitor the evolution of smoke morphology in the visible light field, evaluate the continuous and stable performance of collaborative support effectiveness, and establish the final fire verification conclusion.

[0065] Based on the hovering altitude node that resists downwash wind interference, the altitude value of this node is used as the basis for updating the UAV flight altitude adjustment scheme. This value is written into the flight control command to update the hovering altitude reference for the current mission. Visible light channel images are continuously acquired at the hovering altitude reference position, and the integrity of the smoke region's outline and the changes in grayscale distribution within consecutive time frames are monitored to obtain a smoke morphology evolution monitoring record. Based on the smoke morphology evolution monitoring record, if the smoke outline remains intact in multiple consecutive frames and the grayscale distribution change is below a preset stability threshold, the collaborative protection effect is determined to be continuously stable. Combined with the location information of the bright boundary of the surface fire source, the final fire verification conclusion is confirmed as a fire. If the smoke outline shows continuous dissipation or the grayscale distribution change exceeds the preset stability threshold, the final fire verification conclusion is determined to be a suspected false alarm.

[0066] In one implementation, after determining a hovering altitude node that is resistant to downwash wind field interference, the UAV writes the altitude value of the node into the flight control command as the hovering altitude reference for the current fire verification task. The UAV maintains stable hovering at the altitude reference position and continues to perform the monitoring task.

[0067] Specifically, at the hovering height reference position, a visible light camera continuously acquires image frames at a preset sampling frequency. For each frame of the image, the contour boundary line and the gray-scale distribution features inside the smoke region are extracted. By comparing the shape changes of the smoke contour and the numerical fluctuations of the gray-scale distribution between adjacent frames, a smoke morphology evolution monitoring record is formed, which describes the stability of the smoke in the time dimension.

[0068] It should be noted that the final fire verification conclusion is based on the monitoring records of smoke morphology evolution and the location information of the bright boundary of the fire source on the ground. When the smoke outline remains intact in multiple consecutive frames of images and the gray-scale distribution change is lower than the preset stability threshold, the collaborative protection effect is determined to be stable. Combined with the fire source location information obtained by the thermal imaging channel, the final fire verification conclusion is established as fire confirmation. When the smoke outline shows continuous dissipation or the gray-scale distribution change exceeds the preset stability threshold, the final fire verification conclusion is established as suspected false alarm.

[0069] If the technical solution of this application involves the processing of personal information, the relevant products have established a sound user authorization mechanism: before collecting, using, or sharing personal information, the obligation to inform is fulfilled in accordance with the law, and the individual's voluntary and explicit consent is obtained; if sensitive personal information is involved, the user's separate and explicit consent is further obtained. Specific measures include, but are not limited to: setting up prominent prompts in the information collection area, or clearly displaying the processing rules (including the processor, purpose, method, information type, etc.) through electronic interfaces such as pop-ups, checkboxes, and active submissions, to ensure that users voluntarily authorize based on their knowledge. All personal information processing activities strictly comply with national laws and regulations, especially the relevant provisions of the "Personal Information Protection Law of the People's Republic of China," to effectively safeguard the legitimate rights and interests of personal information subjects.

[0070] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A ground-air integrated intelligent forest fire identification method, characterized in that, include: The real-time hovering altitude of the UAV and the wind pressure feedback value collected by the air pressure sensing unit are obtained. The real-time hovering altitude is compared with a preset safe hovering altitude threshold to determine the initial lower limit of the hovering altitude. The wind pressure feedback value is compared with a preset surface wind speed warning threshold to obtain the surface wind speed estimation result. The initial hovering height limit and the estimated surface wind speed are fused with the image acquired by the visible light channel using particle filtering to evaluate the dissipation and damage performance and attenuation characteristics of the smoke boundary, and to determine the smoke outline integrity score and the degree of downwash airflow disturbance. Based on the degree of downwash airflow disturbance and the smoke contour integrity score, the bright boundary of the ground fire source captured by the thermal imaging lens is determined, the need to lower the hovering position of the UAV for close reconnaissance is assessed, and the lower limit of the corrected hovering height is determined. By using the corrected lower limit of hovering height and combining it with the real-time wind shear data obtained by the wind shear sensing unit, the boundary gradient value and grayscale attenuation feature of the smoke region in the visible light channel are extracted as smoke visual criteria. The smoke visual criteria are compared with the preset smoke identification reliability classification threshold to identify the smoke identification reliability level. The reliability level of the smoke identification was analyzed, and combined with the bright boundary of the surface fire source, the synergistic guarantee effect of precise positioning of thermal imaging and visible light smoke identification was established, and the hovering height node that resists the interference of downwash wind field was extracted. Based on the hovering altitude node that resists downwash wind field interference, update the UAV flight altitude adjustment scheme, continuously monitor the evolution of smoke morphology in the visible light field, evaluate the continuous and stable performance of collaborative protection effectiveness, and establish the final fire verification conclusion.

2. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, The initial hovering height lower limit and the estimated surface wind speed are fused with the image acquired through the visible light channel using particle filtering to evaluate the dissipation and attenuation characteristics of the smoke boundary, determine the smoke outline integrity score and the degree of downwash airflow disturbance, including: The initial hovering height lower limit and the ground wind speed estimation result are obtained. The vertical influence range of the rotor downwash airflow is determined based on the initial hovering height lower limit. The horizontal diffusion amplitude of the downwash airflow is determined based on the ground wind speed estimation result. The vertical influence range and the horizontal diffusion amplitude are combined to obtain the airflow state description quantity. Acquire the current frame image of the visible light channel, perform grayscale processing on the current frame image to obtain a grayscale image, extract the contour boundary of the smoke region in the grayscale image by edge detection, mark the position coordinate sequence of the smoke boundary pixel points along the contour boundary, and obtain the smoke boundary distribution data. Based on the airflow state description and the smoke boundary distribution data, the state tracking of the smoke boundary pixels is performed using a particle filtering method. The airflow state description is used as the external perturbation input for the particle filtering, and the position offset of the smoke boundary pixels between adjacent frames is used as the observation input. The fused smoke boundary tracking result is obtained through particle weight update and resampling. Based on the fused smoke boundary tracking results, the brightness attenuation of smoke boundary pixels between adjacent frames is statistically analyzed. The frame-by-frame decreasing trend of grayscale values ​​of each boundary pixel is used as the attenuation feature. At the same time, the number of broken segments of the contour boundary and the proportion of broken segments to the total length of the contour are statistically analyzed. The smoke contour integrity score is determined based on the relationship between the proportion and the preset contour damage threshold. The degree of downwash airflow disturbance is determined based on the correspondence between the horizontal diffusion amplitude and the average offset of smoke boundary pixels.

3. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, The process of acquiring the real-time hovering altitude of the UAV and the wind pressure feedback value collected by the air pressure sensing unit, comparing the real-time hovering altitude with a preset safe hovering altitude threshold to determine the initial lower limit of the hovering altitude, and comparing the wind pressure feedback value with a preset surface wind speed warning threshold to obtain the surface wind speed estimation result includes: The drone's real-time hovering altitude data and the wind pressure feedback value output by the air pressure sensing unit are obtained. The real-time hovering altitude is compared with a preset safe hovering altitude threshold, and the initial lower limit of hovering altitude is determined based on the comparison result. The rotor speed corresponding to the initial hovering height lower limit is determined by a preset altitude-speed mapping function, and the preset surface wind speed warning threshold is adjusted according to the rotor speed. The adjusted preset surface wind speed warning threshold is compared with the wind pressure feedback value, and the intensity of surface airflow activity is judged based on the comparison result to obtain the surface wind speed estimation result.

4. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, The process of determining the bright boundary of the surface fire source captured by the thermal imaging lens based on the degree of downwash airflow disturbance and the smoke outline integrity score, assessing the need to lower the hovering position of the UAV for close-range reconnaissance, and determining the lower limit of the corrected hovering altitude includes: The degree of turbulence of the downwash airflow and the integrity score of the smoke outline are obtained. The current frame infrared image captured by the thermal imaging lens is retrieved. The bright area of ​​the ground fire source is identified in the infrared image. The coverage range and edge sharpness of the bright boundary of the ground fire source are determined according to the brightness gradient change and temperature difference contrast intensity of the high temperature area. The hovering position adjustment requirement is assessed based on the edge sharpness of the bright boundary of the surface fire source and the smoke outline integrity score. The assessment results determine whether there is a need to adjust the current hovering position. Based on the lowering requirement and the degree of airflow disturbance, the height is corrected on the basis of the initial lower limit of hovering height to obtain the corrected lower limit of hovering height.

5. The integrated ground-air forest fire intelligent identification method according to claim 4, characterized in that, The process involves retrieving the current frame infrared image captured by the thermal imaging lens, identifying bright areas of surface fire sources within the infrared image, and determining the coverage and edge sharpness of the bright areas of the surface fire sources based on the brightness gradient changes and temperature difference contrast intensity of the high-temperature areas. This includes: The degree of turbulence of the downwash airflow and the integrity score of the smoke outline are retrieved to obtain an infrared image in the current field of view of the thermal imaging lens. In the infrared image, a set of pixels whose temperature values ​​exceed a preset fire source temperature threshold are selected and marked as a high-temperature region to obtain the pixel position distribution of the high-temperature region. Based on the pixel position distribution in the high-temperature region, the temperature difference between adjacent pixels is calculated along the horizontal and vertical directions respectively. The change range of the temperature difference in each direction is used as the brightness gradient. The brightness gradient is corrected according to the degree of disturbance to obtain the corrected brightness gradient change characteristics. Based on the corrected brightness gradient change characteristics, edge pixels of the fire source are extracted at the edge of the high-temperature region. The temperature difference between the edge pixels of the fire source and the adjacent low-temperature background pixels is calculated as the temperature difference contrast intensity. The coverage range and edge sharpness of the high-brightness boundary of the fire source on the ground are determined based on the temperature difference contrast intensity distribution.

6. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, Using the corrected lower limit of hovering height, combined with real-time wind shear data obtained by the wind shear sensing unit, the boundary gradient value and grayscale attenuation features of the smoke region within the visible light channel are extracted as smoke visual criteria. These criteria are then compared with a preset smoke identification reliability grading threshold to identify the smoke identification reliability level, including: The corrected lower limit of hovering height is obtained, real-time wind shear data is read from the wind shear sensing unit, and the degree of airflow turbulence is determined by comparing the real-time wind shear data with the preset wind shear threshold. Based on the degree of airflow turbulence, the current frame image acquired by the visible light channel is retrieved, and the smoke region is located in the current frame image. The gray-level difference between adjacent pixels is extracted along the edge of the smoke region as a boundary gradient value. The boundary gradient value is compensated according to the degree of airflow turbulence. The decrease magnitude and deceleration rate of gray-level values ​​from the edge to the center of the smoke region are statistically analyzed as gray-level attenuation features. The boundary gradient value and the gray-level attenuation features are combined to form a visual criterion for smoke. The smoke recognition reliability level is identified by comparing the smoke visual criteria with a preset smoke recognition reliability grading threshold.

7. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, Analyzing the reliability level of the smoke identification, and combining it with the bright boundary of the surface fire source, the synergistic effect of precise thermal imaging positioning and visible light smoke identification is established. Hovering height nodes that resist downwash wind interference are extracted, including: The reliability level of smoke identification and the coverage range and edge sharpness of the bright boundary of the ground fire source are obtained. Based on the reliability level of smoke identification, it is determined whether the smoke characteristics of the visible light channel meet the identification requirements. Based on the edge sharpness, it is determined whether the fire source positioning of the thermal imaging channel meets the accuracy requirements, and a dual-channel collaborative state description is obtained. Based on the dual-channel collaborative state description, it is determined whether the collaborative guarantee effect of thermal imaging precise positioning and visible light smoke identification meets the standard, and the collaborative guarantee effect judgment result is obtained. Based on the collaborative support effectiveness determination result and the revised lower limit of hovering height, when the collaborative support effectiveness meets the standard, the current hovering height is recorded as the hovering height node for resisting downwash wind field interference. When the collaborative support effectiveness does not meet the standard, a preset height increment is added successively based on the revised lower limit of hovering height and the collaborative support effectiveness is re-evaluated. The hovering height node for resisting downwash wind field interference is determined based on the re-evaluation result.

8. The integrated ground-air forest fire intelligent identification method according to claim 1, characterized in that, Based on the hovering altitude node that resists downwash wind field interference, the UAV flight altitude adjustment scheme is updated, and the evolution of smoke morphology in the visible light field is continuously monitored to assess the sustained and stable performance of collaborative support effectiveness, and the final fire verification conclusion is established, including: The node height value of the hovering height node is used as the basis for updating the UAV flight altitude adjustment scheme and written into the flight control command to update the hovering height reference. Visible light channel images are continuously acquired at the hovering height reference position. The contour integrity and grayscale distribution of the smoke region in the image are monitored and the changes in the smoke morphology evolution are recorded in continuous time frames to obtain a smoke morphology evolution monitoring record. Based on the smoke morphology evolution monitoring records, the integrity of the smoke outline and the magnitude of grayscale distribution changes in consecutive multi-frame images are determined. Based on the determination results, the continuous and stable performance of the collaborative protection effectiveness is evaluated. Combined with the location information of the bright boundary of the surface fire source, the final fire verification conclusion is established.

Images