A method and system for automatic identification, positioning and extinguishing of forest fires by helicopter

Abstract

The present application relates to the technical field of forest fire emergency disposal, in particular to a helicopter-borne forest fire automatic identification positioning and extinguishing method and system. The method comprises the following steps: identifying infrared suspected fire points according to meteorological conditions and a space-time weight matrix; analyzing open fire and smoke probability by using the spectral response difference of open fire and smoke in different wave bands; generating a high-confidence three-dimensional fire point list based on D-S evidence theory in combination with the infrared suspected fire points, the open fire and smoke probability, and the burning intensity and fire line strength; constructing a multi-objective optimization model, and solving a to-be-extinguished fire line and a regional priority list in combination with the multi-objective optimization model, fire field terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list; and generating or adjusting a flight route and an extinguishing instruction through a model predictive control framework based on a throwable window, the high-confidence three-dimensional fire point list, the to-be-extinguished fire line and regional priority list.

Description

Technical Field

[0001] This invention relates to the field of forest fire emergency response technology, specifically to a helicopter-borne automatic identification, positioning and extinguishing method and system for forest fires. Background Technology

[0002] Forest fires are characterized by their suddenness, rapid spread, and wide destructive range. Timely and accurate fire detection and scientific firefighting planning are crucial to minimizing fire losses. Existing forest fire monitoring technologies mainly include ground monitoring stations, satellite remote sensing, drone monitoring, and helicopter-borne monitoring. Among these, helicopter-borne monitoring plays a vital role in large-scale forest fire response due to its high mobility, wide coverage, and ability to approach the fire scene at close range. However, current technologies still have the following shortcomings:

[0003] Low detection accuracy of single sensors: Existing helicopter-borne monitoring relies on a single infrared sensor or visible light sensor. Infrared sensors are easily affected by sunlight spots, high-temperature rocks, etc., which can produce false fire points. Visible light sensors have significantly reduced detection capabilities in dense smoke or nighttime environments, making it impossible to achieve accurate fire point identification in complex fire environments.

[0004] Poor performance of multi-source data fusion: Some technologies attempt to fuse data from multiple sensors, but fail to address the spatiotemporal asynchrony of heterogeneous sensors (infrared, multi-band, lidar), and the fusion process does not consider the physical characteristics of fire (such as the coexistence of open flame and smoke), resulting in low confidence of the fusion results and a high risk of false detection and missed detection.

[0005] Insufficient fire location and terrain adaptation: Existing fire location methods are mostly based on two-dimensional plane coordinates, which cannot provide elevation information and do not combine fire site terrain characteristics (slope, aspect, roughness) for analysis, resulting in inaccurate judgment of subsequent fire extinguishing load deployment areas and posing safety risks.

[0006] Static mission planning leads to poor adaptability: Traditional firefighting mission planning is mostly based on fixed plans formulated according to the initial fire scene state, without considering the dynamic changes in fire spread, the real-time status of helicopters (remaining fuel, attitude) and real-time updates of weather conditions, resulting in low feasibility of the planning scheme and difficulty in coping with complex and ever-changing fire scene environments.

[0007] The fire suppression priority ranking is unreasonable: the existing priority ranking relies on a single indicator such as the area or distance of the fire point, without comprehensively considering the fire suppression cost, efficiency, the need for protection of key areas and the trend of fire spread, resulting in an imbalance in the allocation of fire suppression resources and untimely handling of key fire points.

[0008] Therefore, developing a real-time task planning method and system that can achieve high-precision fire detection, effective fusion of multi-source asynchronous data, precise three-dimensional positioning of fire points, and integration of dynamic fire scene and airborne status has become an urgent need in the field of forest fire emergency response. Summary of the Invention

[0009] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides a helicopter-borne automatic identification, location, and extinguishing method for forest fires, comprising the following steps:

[0010] Based on meteorological conditions and a spatiotemporal weight matrix, suspected infrared fire points are identified; the probability of open flames and smoke is analyzed by utilizing the differences in the spectral responses of open flames and smoke in different bands; combining the suspected infrared fire points, the probabilities of open flames and smoke, and the combustion intensity and fire line strength, a high-confidence three-dimensional fire point list is generated based on DS evidence theory; a multi-objective optimization model is constructed, and by combining the multi-objective optimization model, fire terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list, the fire line to be extinguished and the area priority list are solved; based on the deployable window, the high-confidence three-dimensional fire point list, the fire line to be extinguished, and the area priority list, flight paths and fire extinguishing commands are generated or adjusted through a model predictive control framework.

[0011] Optionally, the step of identifying suspected infrared fire points based on meteorological conditions and a spatiotemporal weight matrix includes the following steps:

[0012] Spatiotemporal weight matrix is ​​extracted from historical fire scene spatiotemporal dataset; meteorological condition correction coefficient is calculated, and the segmentation threshold is dynamically adjusted in combination with the spatiotemporal weight matrix and the meteorological condition correction coefficient; suspected fire points are initially extracted using the segmentation threshold, and false fire points are removed through morphological post-processing to obtain infrared suspected fire points.

[0013] Optionally, the step of analyzing the probability of open flame and smoke by utilizing the differences in the spectral responses of open flame and smoke in different wavebands includes the following steps:

[0014] The initial probability of open flame in the region of interest is analyzed based on the multi-band spectral characteristics of open flame, and the initial probability of smoke is analyzed using the spectral and texture characteristics of smoke. A fusion probability model of open flame and smoke is constructed, and the probabilities of open flame and smoke are obtained by combining the fusion probability model, the initial probability of open flame, and the initial probability of smoke.

[0015] Optionally, the step of combining the infrared suspected fire points, the probability of open flame and smoke, and the combustion intensity and fire line intensity to generate a high-confidence three-dimensional fire point list based on DS evidence theory includes the following steps:

[0016] Based on the suspected infrared fire points, the probabilities of open flame and smoke, and the intensity of combustion and fire line, evidence bodies and basic probability allocation functions are constructed respectively. Based on the evidence bodies and basic probability allocation functions, a high-confidence three-dimensional fire point list is generated through evidence conflict processing and synthesis.

[0017] Optionally, calculating the combustion intensity and fire line strength includes the following steps:

[0018] A correlation model between combustion intensity, fire line intensity, multispectral characteristics, and thermal radiation power is established using historical fire scene multispectral images. Combining real-time multispectral images and infrared thermal radiation data of the current fire scene, the combustion intensity and fire line intensity are obtained by fusion calculation of differentially normalized combustion index and thermal radiation power.

[0019] Optionally, the construction of the multi-objective optimization model, combined with the multi-objective optimization model, fire terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list, to solve for the fire line to be extinguished and the area priority list, includes the following steps:

[0020] Geometric constraints are constructed by combining fire terrain features and fire line spread characteristics. With the goals of minimizing firefighting costs and maximizing firefighting benefits, a multi-objective optimization model is constructed based on the geometric constraints. The multi-objective optimization model is solved based on the high-confidence three-dimensional fire point list to obtain the fire lines to be extinguished and the priority list of areas.

[0021] Optionally, the fire site terrain feature parameters are extracted, including the following steps:

[0022] The fire area point cloud is segmented, and then the irregular triangular mesh method is used for terrain modeling. Based on the terrain model, the fire area terrain feature parameters are extracted from two levels: the triangular face and the overall region.

[0023] Optionally, analyzing the fire spread characteristics includes the following steps:

[0024] Using real-time meteorological data and terrain features, a meteorological flow field of the fire area is constructed; based on the meteorological flow field of the fire area and the physical mechanism of forest fire spread, the fire line spread characteristics are analyzed by introducing a Rothermel spread model with an airflow velocity correction term.

[0025] Optionally, the step of generating or adjusting flight paths and fire suppression commands through a model prediction control framework based on the deployable window, the high-confidence three-dimensional fire point list, the fire suppression line and the area priority list includes the following steps:

[0026] An MPC prediction model is constructed by combining helicopter dynamics model, fire scene dynamics model, and fire extinguishing payload delivery model. The delivery window, the high-confidence three-dimensional fire point list, the fire line to be extinguished, and the area priority list are used as inputs to the MPC prediction model. Flight routes and fire extinguishing commands are generated or adjusted according to multi-dimensional collaborative optimization objectives and constraints.

[0027] This invention identifies suspected infrared fire points based on meteorological conditions and a spatiotemporal weight matrix; it analyzes the probability of open flames and smoke using multi-band spectral response differences; based on DS evidence theory, it integrates infrared suspected fire points, open flame and smoke probabilities, combustion intensity, and fire line intensity to generate a high-confidence three-dimensional fire point list; it constructs a multi-objective optimization model combining fireground topography and fire line spread characteristics to solve for a fire suppression priority list; and it dynamically generates or adjusts flight paths and fire suppression commands based on a model predictive control framework and constraints such as deployable windows. This invention specifically addresses the pain points of existing technologies, such as the susceptibility to false detections and missed detections with single sensors, low confidence in multi-source data fusion, insufficient two-dimensional fire point positioning and terrain adaptation, static and rigid task planning, and imbalanced fire suppression priorities. It achieves high-precision identification and three-dimensional positioning of fire points in complex fireground environments, improves the dynamic adaptability and operational safety of fire suppression task planning, optimizes resource allocation efficiency, ensures timely handling of key fire points, and significantly reduces the ecological and economic losses from forest fires.

[0028] Secondly, to efficiently execute the helicopter-borne automatic identification, location, and extinguishing method for forest fires provided by this invention, this invention also provides a helicopter-borne automatic identification, location, and extinguishing system for forest fires, comprising: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory includes program instructions for executing the helicopter-borne automatic identification, location, and extinguishing method for forest fires described in the first aspect of this invention. The helicopter-borne automatic identification, location, and extinguishing system for forest fires of this invention has a compact structure and stable performance, and can stably execute the helicopter-borne automatic identification, location, and extinguishing method for forest fires provided by this invention, further enhancing the overall applicability and practical application capability of this invention. Attached Figure Description

[0029] Figure 1 A flowchart of a helicopter-borne automatic identification, positioning and extinguishing method for forest fires provided in an embodiment of the present invention;

[0030] Figure 2 This is a framework diagram of a helicopter-borne automatic forest fire identification, positioning, and extinguishing system provided in an embodiment of the present invention. Detailed Implementation

[0031] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0032] Please see Figure 1 To address the aforementioned problems, this invention provides a helicopter-borne automatic identification, positioning, and extinguishing method for forest fires, such as... Figure 1 As shown, in one embodiment, the method includes the following steps:

[0033] S1. Identify suspected infrared fire points based on meteorological conditions and a spatiotemporal weight matrix.

[0034] In this embodiment, the step of identifying suspected infrared fire points based on meteorological conditions and a spatiotemporal weight matrix includes the following steps:

[0035] S11. Extract the spatiotemporal weight matrix from the historical fire scene spatiotemporal dataset.

[0036] First, a historical fire scene spatiotemporal dataset was collected. This dataset not only contains the time and location information of fires that occurred in the same area in recent years, but also associates it with detailed terrain type data (such as mountain, hill, valley, coniferous / broadleaf forest distribution areas, etc.). The time granularity is accurate to the hour, and the spatial granularity matches the pixel resolution of the infrared image.

[0037] Then, the sliding window method was used to count the frequency of fires at each pixel location under the same season, time period, and terrain type, and then the probability of fire occurrence was calculated. The higher the frequency, The closer the value is to 1.

[0038] Furthermore, Euclidean distance is used to calculate the distance between the current pixel (x,y) and all historical fire points. The shortest distance is used as the distance attenuation factor The closer the distance, the higher the risk of fire recurrence in the area, and the greater the weight should be. Therefore, an exponential function is used. To achieve distance attenuation, where The coefficients were obtained by linearly fitting the distance-recurrence probability data of historical fire recurrence cases.

[0039] Furthermore, the spatiotemporal weight matrix is ​​calculated, satisfying: ,in The value [0.6, 0.8] is the balance coefficient between spatial and temporal weights. Its core function is to adjust the contribution ratio of spatial proximity and temporal correlation in the overall weight.

[0040] S12. Calculate the meteorological condition correction coefficient, and dynamically adjust the segmentation threshold by combining the spatiotemporal weight matrix and the meteorological condition correction coefficient.

[0041] Infrared radiation is significantly attenuated by meteorological factors such as temperature, humidity, and wind speed during its transmission through the atmosphere, leading to a reduction in the radiation contrast between the fire point and the background. Therefore, it is necessary to construct an infrared radiation attenuation model to quantify this effect. The model is constructed based on classical infrared radiation transmission theory, combined with the operating band of a helicopter-borne infrared thermal imager, and correction coefficients are obtained through fitting indoor simulation experiments and field measurement data. Specifically, as the temperature T rises, the thermal motion of atmospheric molecules intensifies, increasing the scattering of infrared radiation and causing the fire point radiation signal to attenuate. As the relative humidity RH increases, the water vapor content in the air increases, and the absorption and scattering effect of water vapor molecules on infrared photons is significantly enhanced, further weakening the fire point radiation. As the wind speed V increases, the diffusion speed of the hot airflow generated by the fire point combustion accelerates, and the radiation energy in the fire point area is more concentrated in the diffusion direction, requiring an appropriate reduction in the threshold to enhance the fire point radiation characteristics.

[0042] In the embodiment, the correction coefficient satisfies: ,in The fitting coefficients are denoted as .

[0043] The grayscale histogram of infrared images typically exhibits a bimodal distribution with low grayscale peaks in the background and high grayscale peaks in the fire point. This invention employs the Otsu algorithm to calculate the global initial threshold. .

[0044] Furthermore, to achieve adaptive adjustment of the threshold, the spatiotemporal weight matrix is... and weather correction factor By incorporating local threshold calculation, a pixel-by-pixel adaptive local threshold is obtained. Meanwhile, to avoid missing the edges of weak radiation fire points, an adaptive low threshold is set. Ultimately, two pixel-by-pixel threshold matrices (T1 matrix and T2 matrix) are obtained, providing accurate judgment criteria for subsequent segmentation.

[0045] S13. Using the segmentation threshold, suspected fire points are initially extracted, and false fire points are removed through morphological post-processing to obtain infrared suspected fire points.

[0046] The segmentation is performed by traversing pixels one by one, in the order from the top left corner to the bottom right corner of the image, processing each pixel (x, y) row by row and column by column: first, the grayscale value of the pixel is read. , and then with and Compare them. When When identified as a highly suspected fire point, this area exhibits significant fire point radiation characteristics and is likely the core area of ​​a true fire point, marked as label 2; when When identified as a weak suspected fire point—this type of area may be the edge of a fire point or a low-intensity fire point, it is marked as label 1; when When the condition is met, it is determined to be a background area and marked as label 0.

[0047] After segmentation, a preliminary mask for suspected fire points is generated. This mask is a single-channel image with the same size as the original infrared image, and the pixel values ​​correspond to the aforementioned labels (0, 1, 2), which can intuitively reflect the distribution location and intensity level of suspected fire points. At the same time, the boundary coordinates (coordinates of the upper left and lower right corners of the smallest bounding rectangle) of strong and weak suspected fire point areas are recorded, providing accurate regional positioning information for subsequent morphological post-processing and multi-band identification.

[0048] Furthermore, morphological operations are used to perform morphological transformations on suspected fire point areas using structuring elements to eliminate false targets that do not conform to the morphological characteristics of fire points. First, an opening operation (erosion followed by dilation) is performed: the erosion operation uses a 3×3 circular structuring element (based on the characteristic that noise points are mostly small isolated points, a circular structuring element can uniformly remove noise in all directions), traversing the suspected fire point area. When a background pixel (label 0) exists in the pixels covered by the structuring element, the center pixel is marked as background, thereby removing isolated noise points with an area less than 5 pixels (such as small light spots caused by direct sunlight or strong reflections on rock surfaces). The dilation operation uses the same structuring element as the erosion operation to restore the edge contraction of the fire point area caused by erosion, ensuring the integrity of the area of ​​the real fire point. Then, a closing operation (dilation followed by erosion) is performed: the dilation operation first fills the small holes in the fire point area (these holes are mostly unburned areas in the fire point or breaks caused by image noise), and then the erosion operation restores the original outline of the fire point area, avoiding excessive expansion of the area caused by filling the holes. Finally, circularity screening is performed: First, a connected component labeling algorithm (such as the eight-neighbor connected component labeling method) is used to extract all suspected fire point connected components. The area S (total number of pixels within the connected component) and perimeter L (length of the edge pixels of the connected component, calculated using chain code to improve accuracy) of each component are then determined using the formula... Circularity is calculated because fire points, influenced by thermal convection during combustion, typically exhibit an approximately circular or elliptical shape, with a high circularity C value (generally ≥0.3). False fire points (such as elongated rock shadows or irregular light spots) have lower C values. Therefore, connected regions with C < 0.3 are eliminated, and regions with C ≥ 0.3 are retained as purified infrared suspected fire points.

[0049] S2. Analyze the probability of open flame and smoke by utilizing the differences in the spectral response of open flame and smoke in different wavebands.

[0050] The method of analyzing the probability of open flame and smoke by utilizing the differences in the spectral responses of open flame and smoke in different wavebands includes the following steps:

[0051] S21. Analyze the initial probability of open flame in the region of interest based on the multi-band spectral characteristics of open flame, and analyze the initial probability of smoke using smoke spectral and texture characteristics.

[0052] First, based on the coordinates of the minimum bounding rectangle of the suspected infrared fire point area, the ROI (Region of Interest) is cropped in the visible and near-infrared images. To avoid missing the spectral information of the fire point edge, the cropping is extended outward by 5 pixels (based on the infrared image pixel resolution of 0.5m, the extension range is 2.5m, which conforms to the radiation diffusion range of the fire point edge). At the same time, the cropped ROI is preprocessed with grayscale normalization to map the pixel values ​​to [0,255], eliminating the influence of exposure differences in imaging across different spectral bands.

[0053] Furthermore, the spectral characteristics of open flames show that the radiation intensity in the near-infrared band is significantly higher than that in the visible light band, while the background, such as vegetation and rocks, is the opposite. Spectral features are constructed based on this difference.

[0054] First, the average brightness of each ROI is calculated: an adaptive weighted average calculation method is used, with pixel grayscale values ​​as weights and the window size adaptively adjusted to the ROI size, to obtain the average brightness of the visible light band. and the average brightness in the near-infrared band To enhance the distinction between the open flame and the background, the band ratio was calculated. Simultaneously calculate the signal-to-noise ratio in the near-infrared band. ,in The average grayscale value of the top 30% of pixels within the ROI (identified as potential fire signal areas). The average value of the background area with a width of 10 pixels around the ROI. The standard deviation of grayscale in the background area. At that time, the reliability of the fire signal is relatively high.

[0055] Furthermore, an open flame detection function is constructed. ,in This indicates the initial probability of an open flame. Let be a ratio normalization function based on the Sigmoid function, satisfying: , The gain coefficient is obtained by fitting open flame samples. , The band ratio threshold is set to 3.5, which is the critical ratio between open flame and background, to ensure that σ(R) rapidly approaches 1 when R > 3.5; For the signal-to-noise ratio threshold function, when When τ=1, when hour .

[0056] The spectral characteristics of smoke show that its scattering and absorption of blue light are significantly stronger than those of green light, while its texture characteristics are characterized by irregular diffusion and poor grayscale uniformity. Based on this, dual features are extracted.

[0057] First, spectral feature extraction is performed: considering the influence of illumination variation on reflectance calculation, radiometric correction is first applied to the visible light ROI (using a cosine correction model to eliminate the influence of illumination angle), and then the reflectance of the blue light band (400-500nm) is calculated. and reflectivity in the green light band (500-600nm) Construct smoke spectral characteristic parameters Pure smoke area <0.8, while the background vegetation area >1.0, industrial fumes (such as cooking smoke) Between 0.8 and 1.0, the smoke and background can be initially distinguished.

[0058] Texture feature extraction was then performed: Gray-level Co-occurrence Matrix (GLCM) analysis was employed, and GLCM parameters were optimized to address slight motion blur in helicopter-borne images: the stride was set to 2 pixels (balancing texture detail and anti-blurring capability), angles were set at 0°, 45°, 90°, and 135° (to comprehensively capture the diffusion texture of smoke), and gray levels were compressed to 16 levels (reducing computational load while retaining key texture information). Three core metrics—contrast, entropy, and correlation—were extracted from the GLCM in the four directions: contrast reflects the clarity of the texture (low contrast in smoke areas, typically <50), entropy reflects the irregularity of the texture (high entropy in smoke areas, typically >4.0), and correlation reflects the continuity of the texture (low correlation in smoke areas, typically <0.3). To avoid redundancy in multi-dimensional texture metrics, Principal Component Analysis (PCA) was used for dimensionality reduction: first, the covariance matrix of the three metrics was calculated, and the principal component with the largest eigenvalue (contribution rate ≥85%) was extracted as a single texture feature value. , The larger the value, the more the area matches the texture characteristics of smoke.

[0059] S22. Construct a fusion probability model of open flame and smoke, and combine the fusion probability model, the initial probability of open flame and the initial probability of smoke to obtain the probability of open flame and smoke.

[0060] First, a smoke probability function is constructed based on spectral and texture features. λ is the weighting coefficient of spectral and texture features. The optimal value λ=0.6 was determined by 5-fold cross-validation, which means that spectral features contribute more to smoke recognition. Normalize the spectral parameters The smaller, The larger the size, the higher the probability of smoke. The dimensions have been reduced and normalized to the [0,1] interval using PCA to ensure that the weights of the two can be directly superimposed.

[0061] Subsequently, a probability model for the fusion of open flame and smoke was constructed. Open flame combustion is inevitably accompanied by smoke generation, while smoke may be generated by non-fire factors (such as cooking smoke or industrial emissions). Therefore, the probability was corrected by the coexistence relationship.

[0062] Furthermore, the fusion model is specifically as follows: ,in, This represents the final probability of open flame, and γ is the coexistence correction coefficient. Based on historical data statistics of the open flame-smoke coexistence probability, γ is taken as 0.8. Let be the probability of open flame within the smoke area. Considering that the near-infrared band has a certain penetration through smoke, it is obtained by correcting the near-infrared signal-to-noise ratio attenuation model of the smoke area, satisfying: , The smoke attenuation coefficient is obtained by fitting experimental data on smoke concentration and near-infrared transmittance. d is the average thickness of the smoke area (estimated by the gray mean of the ROI), ensuring that the probability of open flame can still be accurately assessed even under smoke obscuration.

[0063] Smoke probability in open flame areas ,satisfy: Where 0.8 is the minimum probability threshold for open flame accompanied by smoke, to avoid the logical contradiction of open flame without smoke; when At that time, it was believed that there was no open flame. That is, the probability calculated directly using smoke features.

[0064] S3. Combining the infrared suspected fire points, the probability of open flame and smoke, and the intensity of combustion and fire line, a high-confidence three-dimensional fire point list is generated based on the DS evidence theory.

[0065] The calculation of the combustion intensity and fire line strength includes the following steps:

[0066] S311. Establish a correlation model between combustion intensity, fire line intensity, multispectral characteristics, and thermal radiation power using historical fire scene multispectral images.

[0067] First, samples meeting the criteria were selected from the historical fire multispectral image dataset: they must include paired images of the fire before (7-15 days before the fire to ensure stable vegetation) and the fire after (3-7 days after the fire was extinguished to avoid secondary changes in the debris after burning), and the images must not be severely obscured by clouds or fog and must not have obvious motion blur.

[0068] Then, the differential normalized flammability index dNBR is calculated. First, the normalized flammability index before the fire is calculated separately. and post-fire normalized flammability index Then through The differential index is obtained; the larger the dNBR value, the more severe the combustion damage.

[0069] Further extraction of thermal radiation power The fire point temperature T and fire point area A corresponding to the ROI are extracted from historical infrared images and substituted into the Stefan-Boltzmann law. Calculate, among which ε is the Stefan-Boltzmann constant, and ε is the emissivity of burning vegetation. For mixed coniferous and broadleaf forests, ε = 0.92; for coniferous forests, ε = 0.94; and for broadleaf forests, ε = 0.9.

[0070] The association model construction phase: First, the dNBR of historical samples is analyzed. Pearson correlation analysis was performed between the marked fire line intensity I (heat release rate per unit length of fire line, calculated from the field measured fire intensity and fire line length in historical data) and outlier samples with correlation coefficient r < 0.6; then, the least squares method of multiple linear regression was used to fit the correlation model, satisfying: ,in The fitting coefficients are used as the final model accuracy to evaluate the model's generalization ability through 5-fold cross-validation.

[0071] The fire intensity level threshold is determined by historical fire loss data: the degree of vegetation burning, the difficulty of firefighting, and the scale of loss corresponding to different fire line intensities are statistically analyzed. In the example, the threshold range is determined as follows: I < 1000 kW / m is light (surface burning of vegetation, low difficulty of firefighting), 1000 ≤ I < 5000 kW / m is moderate (deep burning of vegetation, moderate difficulty of firefighting), and I ≥ 5000 kW / m is severe (large-area continuous burning, high difficulty of firefighting).

[0072] S312. Combining real-time multispectral images and infrared thermal radiation data of the current fire scene, the combustion intensity and fire line strength are obtained by fusion calculation of differential normalized combustion index and thermal radiation power.

[0073] First, by combining real-time multispectral imagery and infrared thermal radiation data of the current fire scene, the current normalized combustion index is calculated. Then calculate the current difference exponent. Prioritize using images from the same spatial location in historical imagery. If there are no historical pre-fire images of the area, then the average image under the same vegetation type and topographic conditions will be used. Next, extract the thermal radiation characteristics of the current fire point and calculate the grayscale mean from the open flame ROI of the current infrared image. A weighted average is used, with pixel grayscale values ​​as weights, to highlight the contribution of high-temperature areas, and this is substituted into the temperature conversion formula. ,in The grayscale-to-temperature conversion coefficient was obtained by fitting historical infrared images with on-site temperature measurement data. The ambient reference temperature is used; the area is based on the three-dimensional area of ​​the open flame region. Furthermore, terrain slope correction was also considered. α is the slope of the fire, which is then substituted into the formula for thermal radiation power. ε is dynamically adjusted based on the current vegetation type at the fire site.

[0074] Furthermore, subsequently and Substitute the values ​​into the constructed correlation model to calculate the fire line intensity at the current fire scene. To improve calculation accuracy, a terrain correction factor is introduced. Rq represents the terrain roughness. Make corrections to obtain the final firepower intensity. The more rugged the terrain, the slightly higher the fire intensity assessment value, which is consistent with the actual difficulty of combustion and diffusion.

[0075] Determination of combustion intensity level: The intensity level of the current fire point is determined by comparing it with a defined threshold range (mild, moderate, or severe). For large-area continuous fire points, a sliding window method is used to calculate the local intensity. To avoid a single intensity value failing to reflect the combustion differences within a region, when the standard deviation of intensity within a window exceeds 1000 kW / m, it is marked as a mixed intensity region, and the maximum and minimum intensity values ​​within the window are output. Fire point region association stage: Based on the three-dimensional coordinates of the fire point, a spatial nearest neighbor matching algorithm is used to correlate the calculated combustion intensity level with... The system is precisely correlated with the corresponding fire point area. If multiple fire point areas overlap, the maximum intensity value and highest intensity level within the overlapping area are taken as the assessment result for that area. Finally, a list of assessment results with location information is output, with each record including: three-dimensional coordinates of the fire point (latitude, longitude, and elevation), and fire line intensity. Combustion intensity rating, assessment reliability (by...) and The reliability is determined by the correlation coefficient: r ≥ 0.7 indicates high reliability, 0.5 ≤ r < 0.7 indicates medium reliability, and r < 0.5 indicates low reliability (and should be marked as requiring special verification).

[0076] In this embodiment, the step of generating a high-confidence three-dimensional fire point list based on DS evidence theory by combining the infrared suspected fire points, the probability of open flame and smoke, and the combustion intensity and fire line strength includes the following steps:

[0077] S321. Construct evidence bodies and basic probability allocation functions based on the infrared suspected fire points, the probability of open flame and smoke, and the intensity of combustion and fire line, respectively.

[0078] Specifically, three interrelated evidence bodies are constructed: Evidence body E1 (infrared suspected fire point evidence), whose core data includes the infrared suspected fire point area, grayscale mean, area, and circularity characteristics, reflecting the thermal radiation and morphological characteristics of the fire point; Evidence body E2 (open flame / smoke probability evidence), whose data comes from the open flame probability... And smoke probability Quantify the degree of spectral and textural matching of fire points; Evidence body E3 (combustion intensity / fire line strength evidence), based on combustion intensity level and fire line strength. And assess credibility, characterizing the combustion intensity characteristics of the fire point. To ensure consistency in subsequent evidence synthesis, a unified identification framework is set for each piece of evidence. ,in, Defined as a state that is detected and determined to be a real fire by sensors. Defined as a state that is clearly determined to be non-fire (such as a false fire point or non-fire smoke). Defined as a state where the sensor's detection features are not significant and cannot be clearly determined.

[0079] In this embodiment, to address the issue of time differences in the acquisition data of various sensors, a high-precision spatiotemporal collaborative alignment mechanism needs to be constructed. This mechanism relies on the millisecond-level timestamps of GPS / IMU to achieve time synchronization of cross-sensor data, while also combining the three-dimensional coordinates of the fire point to complete precise spatial correlation.

[0080] Furthermore, for each piece of evidence, based on the physical meaning and statistical regularity of its detection features, a basic probability allocation (BPA) function m(·) with a clear mapping relationship is constructed. The core function of this function is to transform the quantitative features detected by the sensor into a probability distribution. (Real fire) (Non-fire) The degree of trust in the three propositions (uncertain) provides a standardized trust quantification input for subsequent evidence synthesis. The specific construction process is as follows:

[0081] Evidence Item E1 (Infrared Suspected Fire Point Evidence): The core evidence of this item is the thermal radiation and morphological characteristics of the fire point captured by the infrared sensor, namely the average grayscale value of the suspected infrared fire point. and area Both are core physical attributes of fire points, and some false fire points have been removed through morphological post-processing, resulting in high feature reliability. To achieve accurate feature-to-trust level mapping, a BPA construction strategy based on weighted fusion of dual features is adopted:

[0082] (1) Gray-scale mean normalization function The Sigmoid function is chosen as the core mapping form, and the specific expression is as follows: ,in, The gain coefficient is obtained by fitting historical infrared fire point samples (covering weak fire, strong fire, and false fire points) to ensure... When the grayscale threshold of the fire point is exceeded, the σ value rapidly approaches 1; The grayscale threshold between the infrared fire point and the background is T1(x,y) mean. The function converts the continuous grayscale mean into a confidence component in the range [0,1]. The higher the grayscale (the stronger the thermal radiation of the fire point), the more sufficient the confidence basis for the Fire proposition.

[0083] (2) Area threshold function : Using a piecewise function form, when hour, ;when hour, This function is used to filter out trust interference caused by tiny false fire points, strengthening the support of fire point size for trust levels.

[0084] (3) Final BPA allocation: This indicates the overall level of confidence in the infrared signature for a real fire. , where m1(Uncertain)=0.1.

[0085] Evidence Subject E2 (Open Flame / Smoke Probability Evidence): The data source for this evidence subject is the probability of open flame. And smoke probability The core logic is to optimize trust allocation by utilizing the physical coexistence relationship between open flame and smoke (the probability of open flame being accompanied by smoke is >95%, but smoke is not necessarily accompanied by open flame) to avoid logical contradictions caused by independent identification.

[0086] (1) Fire proposition confidence level: We directly use the open flame probability result because this probability has integrated multi-band spectral and texture features and has been corrected for open flame-smoke coexistence. It can accurately quantify the confidence level of the region as a real open flame without additional weighting adjustment.

[0087] (2) Confidence level of non-fire propositions: using The corrected formula is introduced. The attenuation term is primarily due to the high coexistence of smoke and fire. Smoke areas cannot be directly classified as non-fire zones; their non-fire confidence level needs to be reduced. The attenuation coefficient of 0.5 is based on this coexistence probability statistics, ensuring a reasonable reduction in the non-fire confidence level of smoke areas and avoiding misjudgment.

[0088] (3) Confidence level of the Uncertain proposition: This uncertainty stems directly from the ambiguity of the smoke region. Smoke may be generated by fire or by non-fire factors (such as cooking smoke or industrial emissions). The uncertainty of its attribution is positively correlated with the smoke probability. Therefore, allocating 50% of the smoke probability as the uncertainty confidence level not only reflects the ambiguity of the smoke region but also logically echoes the attenuation term of the Non-fire proposition.

[0089] Evidence Type E3 (Evidence of Combustion Intensity / Flame Line Strength): The core evidence for this type of evidence is the flame line strength. .

[0090] (1) Fire proposition confidence: The mapping relationship is constructed using the Sigmoid function, satisfying: ,in, The gain coefficient is obtained by fitting historical fire intensity and fire authenticity samples to ensure that the confidence level jumps rapidly when the fire intensity crosses the critical value. This is the critical threshold for fire intensity, set as the boundary between light and moderate combustion (I0 = 1000 kW / m). Physically, it represents the minimum fire intensity required to extinguish a real fire. When m3 (Fire) rapidly approaches 1, it indicates that a high-intensity fire line corresponds to a high fire confidence level; when At that time, m3 (Fire) decreases as the intensity decreases, reflecting the uncertainty of the weak combustion zone.

[0091] (2) Confidence level of non-fire propositions: The fire intensity assessment results are derived directly from the fire proposition's confidence level and uncertainty. The fire intensity assessment results can only distinguish the combustion intensity and cannot completely exclude non-fire high-temperature areas (such as volcanic eruptions and industrial high-temperature sources). Therefore, a small amount of uncertainty needs to be reserved, rather than allocating all remaining confidence level to non-fire.

[0092] (3) Uncertain proposition confidence level: set m3(Uncertain)=0.05.

[0093] Through the construction of refined BPA functions, the detection features of infrared, multi-band, and combustion intensity are accurately transformed into standardized trust assignments. The form and parameters of each function are deeply bound to the physical characteristics of the fire point and the detection patterns of the sensor. This not only ensures the scientific nature of the trust assignment but also provides high-quality input data for subsequent evidence conflict handling and synthesis. The final output is a set of BPA functions (m1, m2, m3) that correspond one-to-one with the three pieces of evidence.

[0094] S322. Based on the evidence body and the basic probability allocation function, a high-confidence three-dimensional fire point list is generated through evidence conflict processing and synthesis.

[0095] First, calculate the conflict coefficient K between each piece of evidence to quantitatively describe the different pieces of evidence for the same set of propositions. The difference in trust allocation, with values ​​ranging from K∈[0,1], indicates a greater degree of trust conflict between the evidence (K=0 represents complete consistency, K=1 represents complete conflict, meaning the trust level of each piece of evidence is completely allocated to mutually exclusive propositions). The specific formula for calculating the conflict coefficient K is as follows: Where A, B, and C are propositions of the three evidence bodies E1, E2, and E3 in the recognition framework Θ, and must satisfy the following conditions: That is, the three propositions are mutually exclusive in pairs.

[0096] Furthermore, to ensure the rationality of the conflict level classification, the classification threshold was obtained by calibrating multi-source evidence samples measured at the fire scene: when K>0.6, directly using the DS synthesis rule will lead to evidence paradox in the synthesis result error. For example, if two highly reliable pieces of evidence support Fire and Non-fire respectively, direct synthesis may lead to an Uncertain erroneous conclusion.

[0097] Based on the classification results of the conflict coefficient, a differentiated synthesis strategy is adopted:

[0098] 1) High-conflict scenarios (K > 0.6): The BPA function is corrected using a reliability-weighted average method before synthesis. Weights are set based on the assessment credibility of each piece of evidence (e.g., 0.4 for high credibility of E3's combustion intensity assessment, 0.3 for medium credibility of E1, and 0.3 for medium credibility of E2). The reliability of the evidence is quantified by credibility measures, reducing the weight of conflicting evidence. The specific correction formula is as follows: , where X∈Θ, ω1+ω2+ω3=1, ω1, ω2, ω3 are the weights of each piece of evidence. After the correction, the credibility of conflicting evidence is reasonably weakened, so as to avoid its dominant interference with the synthesis result.

[0099] 2) Low-conflict scenarios (K≤0.6): The classic DS synthesis rule is directly used for evidence fusion. This rule achieves synergistic enhancement of multiple evidences through trust degree orthogonality.

[0100] Furthermore, based on measured samples of different fire conditions (a mixture of weak fires, strong fires, and false fires), the confidence threshold is finely set. In the example, the screening threshold is... Then, the selection criteria are clearly defined: the confidence level of the Fire proposition in the synthesized BPA function. If the fire point is directly identified as a high-confidence fire point, a secondary verification mechanism is added to further improve the reliability of the screening. If m(Fire)≥0.8 but the confidence level of the non-fire proposition m(Non-fire)≥0.1, the original evidence data corresponding to the fire point (such as the mean value of infrared grayscale, the probability of open flame in multiple bands, and the fire line intensity assessment results) needs to be reviewed to exclude misjudgments caused by evidence conflict correction deviations. After the review confirms that there are no abnormalities, the fire point will be included in the high-confidence list.

[0101] Based on this, it is also necessary to achieve accurate correlation and integration of multi-source data. The core information extracted includes: 1) Three-dimensional spatial information, namely the latitude and longitude (B, L, H) in the WGS84 coordinate system, along with the three-dimensional coordinates of the minimum bounding rectangle of the fire point area, to provide a spatial range basis for subsequent route planning; 2) Fire intensity information, covering the combustion intensity level (light / moderate / severe / mixed intensity) and the final fire line intensity. And assess the credibility level (high / medium / low) to provide a quantitative basis for prioritizing firefighting efforts; 3) Multi-sensor feature corroborating information, including infrared grayscale mean, circularity features, and open flame probability. And smoke probability 4) Associated attribute information, marking the evidence fusion identifier and deployable window ID corresponding to the fire point, to achieve rapid matching of fire point information and deployment constraints.

[0102] Furthermore, a high-confidence 3D fire point list is constructed using a standardized data format. Each high-confidence fire point is treated as an independent entry, sorted by fixed fields: fire point ID, 3D coordinates, intensity information, feature evidence, and correlation attributes. The list is also deduplicated; if different evidence fusion results correspond to fire points in the same spatial location, the entry with the highest fire confidence and most complete information is retained, while duplicates are removed. The final output is a high-confidence 3D fire point list. This list not only integrates fire point spatial location, combustion intensity, and sensor characteristics, but also establishes a connection with subsequent deployment windows and priority sorting steps through correlation attributes, providing accurate and reliable core input data for firefighting mission planning.

[0103] S4. Construct a multi-objective optimization model, and combine the multi-objective optimization model, fire terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list to solve for the fire line to be extinguished and the area priority list.

[0104] By leveraging the high-precision distance measurement capabilities of laser point clouds, two-dimensional suspected fire points / open flame areas are mapped to three-dimensional space, enabling the location of the fire points' geographical coordinates (latitude, longitude, and elevation). Simultaneously, by extracting terrain features from the point clouds, terrain parameters such as slope, aspect, and roughness of the fire site are analyzed, providing terrain constraints for subsequent fire spread prediction and fire extinguishing load deployment analysis.

[0105] In this embodiment, the extraction of the fire site terrain feature parameters includes the following steps:

[0106] S411. Segment the point cloud of the fire area and then use the irregular triangular mesh method to model the terrain.

[0107] First, the laser point cloud and image data are spatiotemporally aligned, and then the three-dimensional coordinates of the fire point are used. A circular area with a radius of 50m is initially defined centered on the area of ​​the open flame detected by infrared sensors. (For large fires), the radius will be adaptively expanded to... The square root corresponds to 1.5 times, ensuring coverage of the potential impact range of fire spread.

[0108] Then, ground points and non-ground points (vegetation, buildings, etc.) are separated from the laser point cloud within the defined area by ground point filtering (using the RANSAC random sampling consensus algorithm). Subsequently, the irregular triangular mesh (TIN) method is used for terrain modeling. The obtained TIN terrain model provides a high-quality three-dimensional terrain carrier for the accurate calculation of subsequent terrain parameters such as slope and aspect.

[0109] S412. Based on the terrain model, extract the terrain feature parameters of the fire site from two levels: triangular facets and the overall region.

[0110] Based on the TIN terrain model, terrain feature parameters are extracted from both micro (triangular facets) and macro (regional as a whole) levels, providing core constraints for fire spread prediction and fire extinguishing load deployment.

[0111] Microscopic parameter calculation (slope α, aspect β): For each triangular facet in the TIN model, firstly, solve for its normal vector n=(a,b,c) using the coordinates of the three vertices of the triangular facet. Then, calculate the angle between the normal vector n and the vertical unit vector k=(0,0,1), which is the slope α. The steeper the slope, the faster the fire spreads and the more difficult it is to extinguish. The slope aspect β is the angle between the direction of maximum slope of the triangular face and true north, obtained by solving for the horizontal components (dx, dy) of the slope direction. The angle between the slope and the wind direction directly affects the direction in which a fire spreads.

[0112] Macroscopic parameter calculation (terrain roughness Rq, terrain relief H): Terrain roughness reflects the ruggedness of the ground surface. The calculation first extracts the elevation data of all ground points within the fire area. Solve for the average elevation of the region N is the number of ground points, and then the formula is used. Calculations show that the larger the Rq value, the more rugged the terrain, and the more limited the helicopter's flight maneuverability; the terrain relief H is the maximum elevation within the region. With minimum elevation The difference, i.e. It is used to determine whether a fire is located in a complex terrain area such as a canyon or steep slope.

[0113] In another embodiment, analyzing the fire spread characteristics includes the following steps:

[0114] S421. Construct the meteorological flow field of the fire area using real-time meteorological data and terrain features.

[0115] Considering the computational constraints of real-time helicopter processing, a simplified incompressible Navier-Stokes equation (ignoring higher-order terms of viscous forces) is used to construct a basic flow field model to simulate the horizontal airflow motion and vertical convection effects of the atmosphere. The model assumes that the airflow is a steady-state flow field (approximately steady-state within a time scale of ≤2 hours), and the spatial resolution matches the three-dimensional positioning accuracy of the fire point to ensure that the flow field data can accurately cover the core area of ​​the fire.

[0116] The input parameters for basic flow field modeling include near-surface wind speed V, wind direction θ, air temperature T, and air pressure P provided by real-time weather stations. The initial airflow velocity and pressure fields are obtained by solving the continuity and momentum equations. To improve the adaptability of the flow field to the actual fire environment, three-dimensional dynamic correction is performed using extracted terrain feature parameters.

[0117] Firstly, there's slope correction. Airflow along the slope is accelerated by the component of gravity, and the correction coefficient... α is the slope, when α = 30° The airflow speed increased by 50%, and this coefficient was obtained through fitting experiments in a mountain wind tunnel.

[0118] Second, slope aspect correction: the smaller the angle between the slope aspect and the wind direction, the weaker the airflow obstruction, and the higher the correction coefficient. , β is the slope aspect, when θ=β, cos(θ-β)=1, When the airflow is smoothest, cos(θ-β)=-1 when θ and β are in opposite directions. The airflow is completely blocked;

[0119] Thirdly, roughness correction: the larger the terrain roughness Rq, the stronger the airflow turbulence effect and the more obvious the velocity attenuation. An exponential attenuation model is adopted. When Rq=10m The model parameters were calibrated using measured wind speed data from surfaces with different roughness.

[0120] The final corrected airflow velocity is The airflow direction is adjusted by vector synthesis (the resultant force direction of the downslope direction and the wind direction), and linear interpolation is used to supplement the flow field data in areas with abrupt topographic changes to avoid velocity discontinuities.

[0121] S422. Based on the meteorological flow field of the fire area and the physical mechanism of forest fire spread, the fire line spread characteristics are analyzed by introducing the Rothermel spread model with an airflow velocity correction term.

[0122] Fuel parameters are a core factor influencing the rate of fire spread and need to be dynamically matched with terrain type and real-time environment. First, a fuel parameter database is constructed, based on historical forest resource survey data, categorized and stored according to terrain type (steep slope, gentle slope, valley, ridge), fuel type (mixed coniferous and broadleaf forest, coniferous forest, broadleaf forest, shrubland), and baseline parameters (fuel moisture content M0, fuel load W0, fuel ignition point). For example, steep slope areas are mostly drought-resistant coniferous forests, with a baseline moisture content M0=15% and a carrying capacity W0=5kg / m². 2 Ignition point .

[0123] Based on the terrain type corresponding to the three-dimensional coordinates of the fire point, baseline parameters are retrieved from the database, and then dynamically corrected using real-time meteorological data: fuel moisture content. ,in The temperature and humidity correction coefficient is 0.02, obtained through fitting. T0 is the ambient reference temperature (25℃), and RH is the real-time relative humidity. The fuel load W = W0·(1+0.01·H), where H is the terrain undulation. The greater the undulation, the more uneven the fuel distribution, and the greater the load correction.

[0124] The initial conditions for the ignition point need to be accurately quantified to determine the initial state of the fire, including three core parameters: first, the area S of the open flame region, calculated using the scan line filling method based on the morphologically post-processed ignition point mask, and converted into the actual area by combining the pixel resolution of the infrared image; second, the ignition point temperature. It is obtained by converting the grayscale value of infrared image to temperature through calibration. The calibration formula is as follows: ,in The first is the gray-to-temperature conversion coefficient, calibrated experimentally using a blackbody radiation source; T0 is the ambient reference temperature. The second is the initial fire intensity I0, calculated based on fuel parameters and ignition temperature. , Let be the heat of combustion of fuel (h = 18 MJ / kg for coniferous forests), A be the combustion area (0.8 times the area of ​​the open flame zone, considering incomplete combustion areas), and η be the combustion efficiency (0.6, based on historical fire field measurement data). Fuel parameters (M, W, ... ) and initial conditions for ignition (S, The parameters (I0) provide the core input for the accurate solution of the subsequent propagation model, effectively avoiding prediction bias caused by fixed parameters.

[0125] Furthermore, the base spread velocity V0 under no-airflow conditions is calculated using the classical formula of the Rothermel model: Where ξ is the heat transfer efficiency, ρ is the air density, c is the specific heat capacity of air, and δ is the heat penetration depth (taken as 0.05m). To incorporate the driving effect of airflow, an airflow velocity correction term is introduced, and the propagation velocity formula of the Rothermel propagation model is improved as follows: ,in The reference wind speed is 5 m / s, which is the critical wind speed affecting the spread of forest fires. 0.3 is the airflow influence coefficient, which is obtained by fitting the data through field combustion experiments under different wind speeds.

[0126] The direction of spread is determined using the wind-slope combined force model, and the formula is as follows: Where ω is the wind direction weighting coefficient, which is determined by the wind speed and the downslope air velocity corresponding to the slope. , For real-time wind speed, Downhill airflow velocity ( When the wind speed is relatively high ( When ω≥0.6, wind direction plays a dominant role; when the slope is large ( When ω≤0.4, the slope aspect plays a dominant role.

[0127] To balance prediction accuracy and real-time performance, the time integration method is used to predict the fire line position, with an integration step size of Δt = 10 min and a prediction duration of 2 hours (covering the time scale of a single helicopter firefighting mission). The fire line position is updated every Δt interval based on the current spread speed and direction. The level set method is used to track the leading edge of the fire line to ensure the smoothness of the fire line profile.

[0128] Finally, the core spread feature is extracted from the prediction results: spread direction ( ), spread rate ( The fire spread characteristics are obtained by analyzing the fire spread range (a polygonal area centered on the fire point, including latitude and longitude boundaries) and the length of the fire front (the arc length of the fire front). The fire spread characteristics are also output as snapshots of the fire situation at different times (t=10min, 20min, ..., 120min). The obtained fire spread characteristics accurately reflect the development trend of the fire in the short term and provide the core basis for judging the urgency of the fire for subsequent fire fighting priority ranking.

[0129] Furthermore, the construction of the multi-objective optimization model, combined with the multi-objective optimization model, fire terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list, to solve for the fire line to be extinguished and the area priority list, includes the following steps:

[0130] S431. Construct geometric constraints by combining fire terrain features and fire line spread characteristics.

[0131] Based on the actual operational needs of the fire scene, the impact of terrain on firefighting operations, and the dynamic characteristics of fire spread, multi-dimensional geometric constraints are constructed to ensure that the solution results of the subsequent optimization model are consistent with the actual feasibility of helicopter firefighting. The specific constraint design is as follows:

[0132] 1) Slope constraints: In accordance with forest fire fighting operation specifications and helicopter operation safety requirements, the slope constraints will be... The area is defined as a steep slope. In such areas, the terrain is steep, helicopter hovering stability is poor, significantly increasing the difficulty of operations for firefighters and equipment. Furthermore, the fire spreads faster due to gravity, requiring more time for preparation and operation. Therefore, constraints are established. Where t0 is the standard firefighting time in a flat terrain scenario, and sinα is used to quantify the gain effect of slope on firefighting time. The greater the slope, the closer the sinα value is to 1, and the more significant the increase in firefighting time. This constraint can avoid firefighting failure or safety risks caused by insufficient time reserved for steep slope operations.

[0133] 2) Fire Spread Direction Constraints: The core objective is to prioritize the safety of critical areas (residential areas, important facilities, etc.), which is achieved through spatial orientation determination. Specifically, this is achieved by calculating the angle between the fire spread direction vector and the vector from the fire point to the center of the critical area. When the angle is ≤30°, the fire spread direction is determined to be towards the critical area. In this case, the fire suppression priority weight of that area needs to be increased through constraints. The weight increase coefficient is based on the protection level of the critical area: 0.3 for Level 1 protection areas, 0.2 for Level 2, and 0.1 for Level 3. This ensures that fire suppression resources are prioritized for fire lines threatening critical areas during model solving, reducing the risk of damage to core protected targets from fire.

[0134] 3) Spatial Distance Constraints: From the perspective of optimizing operational efficiency, the order of extinguishing adjacent fire points should follow the principle of proximity to avoid fuel waste and time loss caused by long-distance round-trip flights. The extinguishing interval between adjacent fire points should be set to ≤ ,in The settings are based on the single-operation coverage radius (approximately 150m) and effective payload range of a medium-sized firefighting helicopter, ensuring continuous coverage of adjacent fire points while reducing redundant flight path planning; if the distance between adjacent fire points exceeds 300m, they need to be divided into different operation units, and the firefighting sequence needs to be planned separately.

[0135] 4) Capacity Constraint: Considering the type and quantity of fire extinguishing load, as well as the size of the fire points, the number of fire points that can be covered by a single load application is limited to ≤ This avoids incomplete extinguishing of individual fire points due to dispersed load distribution. The value is dynamically determined by the load type. For example, a 1-ton dry powder load can effectively cover 3-5 lightly burning fire spots or 1-2 moderately burning fire spots, while the coverage capacity of water-based loads is greatly affected by the terrain slope and needs to be adjusted according to the slope coefficient. This constraint can ensure the efficient use of fire extinguishing loads and avoid the problems of overloading or insufficient coverage in a single application.

[0136] Based on terrain feature parameters (slope, roughness) and fire spread characteristics (spread direction, fire point distribution), by quantifying the thresholds and correlation logic of the constraints, the final output is a set of geometric constraints that can directly support the solution of the optimization model, ensuring the practicality and security of the model results.

[0137] S432. With the goal of minimizing fire extinguishing costs and maximizing fire extinguishing benefits, a multi-objective optimization model is constructed based on the geometric constraints. The multi-objective optimization model is solved based on the high-confidence three-dimensional fire point list to obtain the fire lines to be extinguished and the priority list of areas.

[0138] Based on the core needs of forest fire fighting operations, a dual core objective function is defined: first, to maximize the benefit-cost ratio. The goal is to achieve efficient use of firefighting resources by quantifying the ratio of firefighting benefits to resource consumption, thus avoiding blind operations that are high-consumption and low-efficiency; secondly, to minimize firefighting time. Considering the time-sensitive nature of forest fire spread, shortening the firefighting time can significantly reduce the fire spread area and reduce ecological and economic losses. The two objectives work together to ensure a balance between the economy and timeliness of firefighting operations.

[0139] Furthermore, the constraint system is further refined and supplemented based on geometric constraints: ① The constraint on the number of fire points explicitly requires that the solution results must cover all high-confidence fire points, without omitting key fire areas, while avoiding invalid planning for low-confidence pseudo-fire points; ② The constraint on the protection of key areas is implemented by setting priority weights, assigning weight coefficients of 1.5, 1.2, and 1.0 to level 1, 2, and 3 protection areas respectively, forcing priority to ensure the allocation of fire-fighting resources in high-level key areas during the solution process, and ensuring the safety of key areas such as densely populated areas and core ecological areas.

[0140] To adapt to complex optimization scenarios with multiple objectives and constraints, the Non-Dominated Sorting Genetic Algorithm (NSGA-III) was selected for model solving. First, the order of fire suppression and resource allocation for the fire lines / areas to be extinguished were used as core optimization variables, with defined value ranges (e.g., the fire suppression order corresponds to the fire point number sequence, and the resource allocation does not exceed the helicopter's maximum payload). Then, the solution was advanced through three core operations of the genetic algorithm: the selection operation employed a tournament selection strategy, selecting individuals with higher fitness (i.e., planning schemes closer to the objective function) from the current population to enter the next generation, with a selection pressure of 0.8 to balance population diversity and convergence speed; the crossover operation used a partially matched crossover method, performing gene recombination on the fire suppression order variable, with a crossover probability of 0.6 to ensure the population's evolutionary capacity; the mutation operation was achieved by randomly swapping the values ​​of two optimization variables, with a mutation probability of 0.1 to avoid the algorithm getting trapped in local optima. After multiple generations of iterative evolution, a Pareto optimal solution set satisfying all constraints was obtained (any solution in the set cannot improve the performance of one objective without degrading the performance of the other).

[0141] To facilitate subsequent prioritization, the optimal solution for multiple objectives needs to be transformed into a single comprehensive index. Therefore, a comprehensive priority index formula is constructed. The weighting coefficients ω1=0.6 and ω2=0.4 are set based on the core requirements of firefighting operations. Through expert scoring combined with a large number of historical firefighting cases, the final output is a Pareto optimal solution set containing multiple feasible planning schemes, as well as the comprehensive priority index corresponding to each scheme, providing core data support for the subsequent generation of the priority list.

[0142] Based on the magnitude of the comprehensive priority index P, each fire line / area to be extinguished is sorted, with a higher P value indicating a higher priority. At the same time, considering the protection needs of critical areas, fire lines facing critical areas are given higher priority (P value increases by 0.2). Finally, a priority list of fire lines and areas to be extinguished is generated, including area location, priority level, and recommended order of extinguishing.

[0143] S5. Based on the deployable window, the high-confidence three-dimensional fire point list, the fire line to be extinguished and the area priority list, generate or adjust the flight path and fire extinguishing instructions through the model prediction control framework.

[0144] In this embodiment, the drop zone is analyzed by combining the real-time status of the helicopter (position, speed, attitude, and remaining fuel) and the drop requirements of the fire extinguishing payload (such as drop altitude, speed, and angle constraints). At the same time, the dynamic constraints of the helicopter (such as lift, thrust, and maneuverability) and safety rules (such as staying away from high-voltage lines and no-fly zones) are considered to calculate the flight envelope online and determine the drop window (time and space range) that meets the constraints, providing feasible drop constraints for subsequent mission planning.

[0145] First, an initial deployable area is selected. Based on the characteristics of the fire extinguishing load, the helicopter's flight status, and the fire scene's obstacle environment, a spatial range that meets the basic deployment requirements is quickly delineated, laying the foundation for subsequent refined constraint screening. Specifically, the threshold settings for the fire extinguishing load deployment parameters need to be combined with experimental data on the load type (dry powder / water-based / foam): minimum deployment height. Determined by the load spread radius and landing speed requirements, ensuring the load evenly covers the fire point without damaging ground vegetation; maximum height This is based on the drift error control of the load in the air.

[0146] Next, the elevation of the core fire area was determined by combining the three-dimensional topography of the fire site. The absolute elevation conversion formula is used. The relative deployment height range Convert to an absolute height range in a geodetic coordinate system to avoid insufficient or excessive deployment height due to terrain undulations.

[0147] Secondly, the ΔV value of the delivery speed range needs to match the flight stability level of the helicopter. For medium-sized firefighting helicopters, ΔV is usually set to 10km / h. This value is determined through flight attitude stability experiments to ensure that when the speed fluctuation is within this range, the helicopter attitude angle deviation will not affect the payload delivery attitude.

[0148] Finally, DEM obstacle information preprocessing requires data format standardization (converting to the WGS84 coordinate system, consistent with the helicopter positioning reference), followed by the use of a morphological dilation algorithm to expand obstacle areas such as high-voltage lines and buildings to compensate for boundary errors in obstacle identification; safety distance. Referring to the "Forest Aviation Firefighting Safety Regulations", an additional 10m of redundancy is added for high-risk obstacles such as high-voltage lines.

[0149] During the screening process, the effective regions of height constraints, velocity constraints, and obstacle constraints are intersected by a spatial overlay analysis algorithm. The final output is a set of polygonal boundary coordinates (including latitude, longitude, and elevation information) of the initial deployable region, and the load adaptation type of each region is marked, providing accurate spatial constraint input for subsequent flight envelope calculation.

[0150] Furthermore, based on the six-degree-of-freedom dynamic equations of the helicopter, a full-dimensional constraint model covering lift, thrust, and maneuverability is constructed to ensure that the flight state meets the requirements of structural safety and control stability. This results in the acquisition of a set of flight states with all dynamic constraints, including the speed range, altitude range, and attitude angle range, providing dynamic boundaries for subsequent flight envelope calculations.

[0151] Furthermore, by integrating safety rules and dynamic constraints, a feasible flight state domain is solved through optimization algorithms to achieve dynamic adaptation of the flight envelope. First, safety rules are quantified and transformed: For no-fly zones, spatial inclusion of polygonal boundaries is used for judgment. Ray casting is used to detect whether the position corresponding to the flight state falls within the no-fly zone, and the no-fly zone boundary is pushed in real-time by the ground command system. The calculation of remaining fuel constraints requires combining shortest path planning, using the A* algorithm to solve for the shortest flight distance from the helicopter's current position to each drop zone, and combining this with a fuel consumption rate model to obtain the one-way fuel consumption. In addition to distance requirements, obstacle avoidance constraints include heading constraints, i.e., the angle between the flight heading and the obstacle edge must be ≥30° to prevent crosswinds from causing the helicopter to deviate from its flight path and collide with obstacles. Second, the constraint set of the flight envelope is constructed using a multi-dimensional constraint matrix, transforming dynamic constraints (lift, thrust, overload) and safety rule constraints (no-fly zone, fuel, obstacles) into inequality constraints of variables. The optimization objective function is set to maximize the probability of successful drop, which is obtained by weighting three factors: drop altitude error, speed error, and attitude stability. The final flight envelope is a three-dimensional feasible region of velocity, altitude, and attitude, which is represented by a grid discretization and the success probability of deployment is marked for each grid point, providing a refined flight state basis for subsequent deployment window matching.

[0152] By precisely matching flight status with spatial regions and combining this with time-constrained filtering, a spatiotemporal delivery window that is both feasible and timely is ultimately obtained. Firstly, the matching of flight status with the initial deployable area employs a dual-dimensional verification process: spatially, it determines the absolute altitude of each grid point within the flight envelope. Does it fall within the corresponding area? Within the range, does the speed meet the deployment speed requirements? In terms of attitude, considering the terrain slope β, the angle between the helicopter's pitch angle and the slope should be ≤15° to avoid excessive attitude affecting deployment accuracy when flying downhill.

[0153] During the matching process, a spatial indexing acceleration algorithm (such as R-tree indexing) is used to divide the initial deployable areas into hierarchical levels according to spatial location, quickly locate the candidate areas corresponding to each flight state, and output the deployable spatial subsets that meet the flight constraints (including area ID, boundary coordinates, and suitable flight state range). Secondly, the arrival time calculation is based on real-time path planning results, using a dynamically weighted A* algorithm. Terrain roughness Rq and wind speed V are used as path weights (the greater the roughness and the greater the headwind speed, the higher the weight), to solve for the shortest time from the helicopter's current position to the center of each deployable area. The time calculation needs to include the helicopter's acceleration / deceleration process. When the distance is short, it is calculated at a constant speed; when the distance is long, it is calculated using a three-stage motion model of acceleration-uniform speed-deceleration to avoid time errors. The upper limit of the mission time needs to be set in conjunction with the fire spread prediction results, usually set to 30 minutes, meaning only areas reachable within 30 minutes are retained to avoid the fire spreading due to late arrival. This threshold can be dynamically adjusted by the ground command system according to the urgency of the fire. Finally, time-constrained areas are filtered out, and the spatial range of the remaining areas is associated with the corresponding time range to generate a list of deployable windows. Each window includes: window ID, spatial polygon coordinates, a set of feasible flight states, earliest / latest arrival time, and probability of successful deployment, providing accurate spatiotemporal constraint inputs for subsequent dynamic task replanning. Simultaneously, a real-time window validity update mechanism is established. When new fire points are detected at the fire site, meteorological data changes abruptly, or helicopter status changes, matching and filtering are re-triggered to ensure the dynamic adaptability of the windows.

[0154] Furthermore, the step of generating or adjusting flight paths and fire suppression commands through a model prediction control framework based on the deployable window, the high-confidence three-dimensional fire point list, the fire suppression line and the area priority list includes the following steps:

[0155] S51. By combining helicopter dynamics model, fire dynamics model, and fire extinguishing load delivery model, an MPC prediction model is constructed.

[0156] Specifically, the helicopter dynamics model adopts a six-degree-of-freedom nonlinear dynamics model to accurately describe the motion characteristics of the helicopter in three-dimensional space. The core is to establish the dynamic change law of state parameters such as position, velocity (including three-axis velocity components) and attitude (including pitch angle, roll angle and heading angle) as a function of control commands.

[0157] The fire dynamic model, based on the Rothermel spread model, combines real-time meteorological data and fire monitoring data to achieve dynamic updates and short-term predictions of fire spread characteristics. Real-time meteorological data (wind speed and direction changes) is used to correct airflow velocity correction coefficients, thereby adjusting the spread rate; fire monitoring data is used to reset the model's initial boundary conditions. The model's core function is to update fire spread characteristics (including real-time corrections to spread rate and direction) and predict future dynamic evolution of the fire. The prediction output includes key parameters such as the fire front position, rate of change of burning area, and fire point temperature distribution at different times.

[0158] Fire Extinguishing Load Placement Model: To address the differences in the mechanisms of action of different types of fire extinguishing loads (dry powder, water-based, foam), a differentiated load placement suppression model is constructed. The core objective is to describe the suppression effect and evolution of the fire ignition state after load placement. The model input parameters include key parameters of the placement window (placement height, placement speed, placement location) and load parameters (placement mass, diffusion radius, and duration of action). The output is a quantitative indicator of the suppression effect, including the rate of decrease in fire ignition temperature and the coefficient for slowing the spread rate, while also defining the effective time window of the load's action.

[0159] The model parameters were calibrated using extensive historical deployment experimental data to ensure the accuracy of suppression effect predictions under different fire scenarios. Based on the coupling of the three sub-models, the core set of state variables for the MPC prediction model was defined. This set of variables achieves full-dimensional coverage of helicopter status, fire status, and payload status: helicopter status variables accurately locate the real-time spatial attitude and motion of the helicopter; fire status variables (fire point temperature, spread rate, and range) quantify the fire evolution trend, with the range represented by a set of three-dimensional polygon coordinates; payload status variables (remaining payload and deployment location) correlate resource consumption with operational effectiveness, with the deployment location represented by latitude and longitude coordinates.

[0160] The update frequency of each state variable is consistent with the data acquisition frequency of the corresponding sensor to ensure that the model can track dynamic changes in real time. By integrating real-time updated data from helicopters and the fire site, the final output is a complete MPC prediction model that can directly support subsequent rolling optimization.

[0161] S52. Using the deployable window, the high-confidence three-dimensional fire point list, the fire extinguishing line and the area priority list as inputs to the MPC prediction model, the flight path and fire extinguishing instructions are generated or adjusted according to the multi-dimensional collaborative optimization objectives and constraints.

[0162] First, construct a multi-dimensional collaborative optimization target system:

[0163] 1) Minimize route tracking error: The route tracking error is defined as the average three-dimensional Euclidean distance between the actual flight trajectory of the helicopter and the planned route. The goal is to control this average value within 5m.

[0164] 2) Minimize the error of fire extinguishing load delivery: Delivery error specifically refers to the horizontal distance between the actual landing point of the load and the center of the core area of ​​the target fire point. Different load types correspond to different error control thresholds (dry powder load ≤ 8m, water-based load ≤ 10m).

[0165] 3) Minimize mission completion time: The total time it takes for a helicopter to depart from its current location, complete all high-priority fire suppression missions, and return to a safe take-off and landing point is used as a quantitative indicator. Shortening mission completion time can effectively curb the spread of fires and reduce the ecological damage and economic losses caused by fires. Especially for high-intensity fires that spread rapidly, timeliness directly determines the success or failure of firefighting.

[0166] 4) Maximize remaining fuel: The goal is to ensure that the helicopter has at least 20% of its total fuel remaining after completing the designated firefighting mission. Sufficient remaining fuel provides room for handling emergencies, such as sudden weather changes, the need to temporarily deal with new fires, or helicopter malfunctions, ensuring a safe return or emergency avoidance, and significantly improving the safety of flight operations.

[0167] Furthermore, based on the characteristics of the fire scene environment, helicopter operational capabilities, and mission planning priorities, a comprehensive constraint system is constructed:

[0168] 1) Drop window constraint: This constraint is the core prerequisite for ensuring the effectiveness of payload delivery. It requires that the flight path must accurately pass through the drop space of each target fire point, and the delivery action must strictly fall within the drop time window.

[0169] 2) Helicopter dynamics constraints: Strictly adhere to the mechanical performance limits of the helicopter, specifically including overload constraints, speed constraints, and altitude constraints.

[0170] 3) Fire safety constraints: The core is to set a safe distance between the helicopter and the front of the fire line. This safe distance is determined based on experimental data on the range of fire heat radiation and the radius of influence of hot airflow.

[0171] 4) Priority constraints: The mandatory planning scheme prioritizes the operation sequence of high-priority fire-fighting areas. It requires helicopters to fly to the next highest priority area after completing the fire-fighting of the previous high-priority fire point, and not to skip high-priority fire points due to factors such as route convenience.

[0172] Finally, a closed-loop mechanism of short-term prediction, real-time optimization, and rolling execution is used to achieve precise adaptation to the dynamic fire environment and helicopter status. Specifically, combining the real-time requirements of helicopter firefighting operations with the onboard computing power capacity, the control cycle is determined through field testing and verification. Within each control cycle, the current full-state data of the helicopter (including precise position, three-axis speed, attitude angle, remaining fuel, and real-time engine power) and the dynamic state data of the fire scene (including the real-time spread rate of each fire point, the position of the fire front, the coordinates of newly added high-confidence fire points, and real-time wind speed and direction) are collected and fused as the input boundary conditions for the optimization problem.

[0173] Furthermore, the optimization problem to be solved has multiple objective functions: minimizing flight path tracking error, minimizing payload delivery error, minimizing mission completion time, and maximizing remaining fuel. This is transformed into a single-objective optimization through weighted summation (the weighting coefficients are dynamically adjusted based on fire suppression priority; for example, the delivery error weight corresponding to a high-priority fire point is increased by 30%). Simultaneously, it is strictly constrained by the spatiotemporal range of the delivery window, helicopter dynamic limits (overload, speed, and altitude constraints), and firefield safety distance constraints. By solving this optimization problem, the optimal control sequence for multiple future control cycles can be obtained. The flight control commands encompass precise control parameters such as throttle opening, elevator / aileron / rudder deflection angles, enabling refined control of flight attitude and speed. The fire suppression delivery commands specify the exact time points, payload delivery amounts, and precise delivery locations for each delivery action.

[0174] In other embodiments, to ensure the real-time performance and effectiveness of the control, a rolling execution mechanism that only executes the current cycle command is adopted. That is, only the optimal command of the current control cycle is sent to the helicopter control system. In the next control cycle, the latest helicopter status and fire data are collected again, the boundary conditions and constraint parameters of the optimization model are updated, and the above optimization process is repeated to form a continuously adapting closed-loop control.

[0175] Regarding the specific solution method, considering the convex optimization characteristics of the optimization problem with multiple objectives and constraints, the quadratic programming (QP) algorithm is selected for efficient solution. First, the multi-objective optimization problem is transformed into a single-objective quadratic programming problem through weighted summation (the weight matrix is ​​determined using the analytic hierarchy process, taking into account both task priority and operational safety). Then, the objective function is quadratically transformed, and various constraints are converted into linear inequality constraints (e.g., nonlinear helicopter dynamics constraints are approximated as linear constraints through local linearization). After solving using the quadratic programming algorithm, detailed parameters of the optimal flight path are obtained, including point-by-point latitude and longitude coordinates, corresponding elevations, flight speed planning for each segment of the path, and a corresponding set of firefighting commands. This achieves precise coordination between the flight path and firefighting actions, ultimately outputting real-time control commands (including control parameter thresholds) that can directly drive the helicopter to execute, and high-precision flight path planning results, ensuring the accuracy and real-time performance of firefighting operations.

[0176] Please see Figure 2 In this embodiment, to efficiently execute the helicopter-borne automatic forest fire identification, location, and extinguishing method provided by this invention, the invention also provides a helicopter-borne automatic forest fire identification, location, and extinguishing system, comprising: an input device, an output device, a processor, and a memory. The input device, output device, processor, and memory are interconnected. The memory contains program instructions used to execute the steps of the helicopter-borne automatic forest fire identification, location, and extinguishing method. The helicopter-borne automatic forest fire identification, location, and extinguishing system of this invention has a compact structure and stable performance, and can stably execute the helicopter-borne automatic forest fire identification, location, and extinguishing method of this invention, further enhancing the overall applicability and practical application capability of this invention.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A helicopter-borne automatic identification, positioning, and extinguishing method for forest fires, characterized in that, Includes the following steps: Based on meteorological conditions and a spatiotemporal weight matrix, identify suspected infrared fire points; By utilizing the differences in the spectral responses of open flames and smoke in different wavebands, the probabilities of open flames and smoke are analyzed. Combining the infrared suspected fire points, the probability of open flame and smoke, and the combustion intensity and fire line strength, a high-confidence three-dimensional fire point list is generated based on the DS evidence theory; A multi-objective optimization model is constructed, and the fire line and area priority list are solved by combining the multi-objective optimization model, fire terrain feature parameters, fire line spread characteristics and the high-confidence three-dimensional fire point list. Based on the deployable window, the high-confidence three-dimensional fire point list, the fire extinguishing line and the area priority list, the flight path and fire extinguishing instructions are generated or adjusted through the model prediction control framework. The process of identifying suspected infrared fire points based on meteorological conditions and a spatiotemporal weight matrix includes the following steps: Extract the spatiotemporal weight matrix from historical fire scene spatiotemporal datasets; Calculate the meteorological condition correction coefficient, and dynamically adjust the segmentation threshold by combining the spatiotemporal weight matrix and the meteorological condition correction coefficient; Calculate the spatiotemporal weight matrix ,satisfy: ， in The value [0.6, 0.8] represents the balance coefficient between spatial and temporal weights. Indicates the probability of a fire occurring. This represents the current pixel (x, y) and all historical fire points. The shortest distance is the distance attenuation factor. The coefficients are obtained by linearly fitting the distance-recurrence probability data of historical fire recurrence cases. Suspected fire points are initially extracted using the segmentation threshold, and false fire points are removed through morphological post-processing to obtain infrared suspected fire points; The process of combining the infrared suspected fire points, the probability of open flame and smoke, and the combustion intensity and fire line strength to generate a high-confidence three-dimensional fire point list based on DS evidence theory includes the following steps: Based on the infrared suspected fire points, the probability of open flame and smoke, and the intensity of combustion and fire line, respectively, construct evidence bodies and basic probability allocation functions; Based on the evidence body and the basic probability allocation function, a high-confidence three-dimensional fire point list is generated through evidence conflict processing and synthesis. The evidence includes infrared suspected fire point evidence, open flame and smoke probability evidence, and combustion intensity and fire line strength evidence. Calculate the conflict coefficient between different pieces of evidence to describe the differences in trust allocation among different pieces of evidence for the same set of propositions; The classification threshold is obtained by calibrating multi-source evidence samples measured at the fire scene. Based on the classification results and screening rules, a high-confidence three-dimensional fire point list is generated.

2. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 1, characterized in that, The method of analyzing the probability of open flame and smoke by utilizing the differences in the spectral responses of open flame and smoke in different wavebands includes the following steps: The initial probability of open flame in the region of interest is analyzed based on the multi-band spectral characteristics of open flame, and the initial probability of smoke is analyzed using the spectral and texture characteristics of smoke. A fusion probability model of open flame and smoke is constructed. By combining the fusion probability model, the initial probability of open flame, and the initial probability of smoke, the probabilities of open flame and smoke are obtained.

3. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 1, characterized in that, The calculation of the combustion intensity and fire line strength includes the following steps: A correlation model was established between combustion intensity, fire line intensity, multispectral characteristics, and thermal radiation power using historical fire scene multispectral images; By combining real-time multispectral images and infrared thermal radiation data of the current fire scene, the combustion intensity and fire line strength are obtained through the fusion calculation of differentially normalized combustion index and thermal radiation power.

4. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 1, characterized in that, The construction of a multi-objective optimization model, combined with fire terrain feature parameters, fire line spread characteristics, and the high-confidence three-dimensional fire point list, to solve for the fire line to be extinguished and the area priority list, includes the following steps: Geometric constraints are constructed by combining fire terrain features and fire line spread characteristics. With the objectives of minimizing fire extinguishing costs and maximizing fire extinguishing benefits, a multi-objective optimization model is constructed based on the geometric constraints. The multi-objective optimization model is then solved based on the high-confidence three-dimensional fire point list to obtain the fire lines to be extinguished and the priority list of areas.

5. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 4, characterized in that, Extracting the terrain feature parameters of the fire site includes the following steps: The fire zone point cloud is segmented, and then the terrain modeling is performed using the irregular triangular mesh method; Based on the terrain model, the terrain feature parameters of the fire site are extracted from two levels: triangular facets and the overall region.

6. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 4, characterized in that, The analysis of the fire spread characteristics includes the following steps: Meteorological flow fields in the fire area were constructed using real-time meteorological data and terrain features; Based on the meteorological flow field of the fire area and the physical mechanism of forest fire spread, the characteristics of fire line spread are analyzed by introducing the Rothermel spread model with an airflow velocity correction term.

7. The helicopter-borne automatic identification, positioning, and extinguishing method for forest fires according to claim 1, characterized in that, The process of generating or adjusting flight paths and fire suppression commands through a model-based predictive control framework, based on the deployable window, the high-confidence three-dimensional fire point list, the fire suppression line, and the area priority list, includes the following steps: An MPC prediction model was constructed by combining helicopter dynamics model, fire dynamics model, and fire extinguishing load delivery model. The MPC prediction model takes the deployable window, the high-confidence three-dimensional fire point list, the fire line to be extinguished, and the area priority list as inputs, and generates or adjusts flight paths and fire extinguishing commands based on multi-dimensional collaborative optimization objectives and constraints.

8. A helicopter-borne automatic identification and positioning fire extinguishing system for forest fires, characterized in that, The helicopter-borne automatic forest fire identification and extinguishing system includes: an input device, an output device, a processor, and a memory. The input device, output device, processor, and memory are interconnected. The memory includes program instructions, which are used to execute the helicopter-borne automatic forest fire identification and extinguishing method according to any one of claims 1-7.

Images