Fire source positioning system of UAV fire cannon based on infrared thermal imaging

Abstract

The present invention provides a fire source positioning system for a drone fire cannon based on infrared thermal imaging. The system includes a fire scene situation acquisition module that acquires thermal image data and high-precision point cloud data; a fire spread model construction module that generates a radiation intensity distribution map; a thermal image processing module that generates a structured fire source risk characteristic image; an LSTM time series model processing module that generates a smooth diffusion heat map to predict the spread direction and speed; a battle map generation module that achieves data visualization; and a drone flight path planning module that achieves comprehensive coverage of the fire cannon's strike path. To address the problem that existing technologies rely on static fire source data and have difficulty capturing the spatiotemporal dynamic evolution of fires, the present invention constructs a multi-dimensional fire spread model that integrates fire scene air flow vector projection analysis, dynamic superposition of thermal radiation background concentration, and LSTM time series modeling to accurately predict the fire spread path from low-risk areas to high-risk areas, providing immediate support for firefighting operations.

Description

Technical Field

[0001] The present invention relates to the field of fire fighting technology, and in particular to a fire source positioning system for a drone fire fighting cannon based on infrared thermal imaging. Background Art

[0002] Traditional fire cannon fire source positioning systems mainly rely on static fire source locations and single temperature data. This limitation makes it difficult to cope with the spatiotemporal dynamic changes of fire.

[0003] In complex environments, such as high-density urban buildings, chemical plants, and forest-city interfaces, the spread of fire is affected by many factors, including wind speed, terrain, distribution of flammable materials, building density, and the characteristics of chemical substances.

[0004] Specifically, in densely populated urban areas, fires spread rapidly upward due to the chimney effect of tall buildings. Traditional systems are unable to capture these dynamic changes in real time, hindering the timely deployment of firefighting resources. In chemical plant fires, hidden fire sources and the risk of re-ignition in chemical storage tank areas are key issues that traditional systems struggle to identify, leading to uncontrolled fires and even secondary disasters. Furthermore, fires at the intersection of forest and city areas are significantly affected by wind speed and vegetation distribution. Traditional systems are unable to accurately predict the direction and speed of fire spread, thus delaying firefighting efforts.

[0005] To address this issue, the existing technology CN118552873A proposes an urban fire warning analysis method, focusing on: real-time acquisition of fire scene monitoring video images through cameras, image preprocessing, smoke detection using an improved YOLOv5s network model, and determining whether the number of consecutive smoke detections exceeds a threshold to trigger a fire warning.

[0006] However, another major problem of the traditional system is not considered: the limitation of a single data source. Relying solely on temperature data or the location of the fire source cannot fully reflect the complexity of the fire scene, especially under changing environmental conditions. This increases the uncertainty and risk of firefighting operations. Specifically, the existing technology CN118552873A, the method proposed relies only on video images collected by the camera, lacks the integration of multi-source data such as infrared thermal imaging of the fire scene and LiDAR point cloud data, and cannot fully capture the "dynamic development" of the fire scene. In addition, this method is mainly based on a single analysis and lacks the ability to dynamically predict the spread of fire. It is difficult to adapt to complex environments (such as high-density buildings, chemical plants and forest-city junctions), resulting in problems such as fire source positioning deviation, misjudgment of diffusion paths and delayed resource scheduling in actual applications.

[0007] In addition, the existing technology CN118230533B proposes an intelligent analysis system based on fire information. Through a fire data acquisition module, a data analysis and processing module, and a fire level assessment module, it monitors the fire information in the fire area in real time, analyzes the fire scale and spread risk, assesses the degree of fire danger, and formulates rescue measures based on the assessment results.

[0008] Similarly, existing technology CN118552873A relies primarily on explicit indicators such as temperature, smoke concentration, and fire source area obtained at fixed / static detection points, resulting in significant deficiencies in its ability to dynamically predict fire spread. Specifically, the system primarily relies on the real-time collection and analysis of static data (such as temperature, smoke concentration, wind speed, and flammable material content), lacking the ability to dynamically model fire spread. It does not consider the temporal trends of fire spread to simulate and predict the spread path, and is unable to effectively predict the nonlinear spread behavior of fire in complex environments. Consequently, it is unable to provide forward-looking fire spread predictions when responding to large-scale, complex fire scenarios, thereby impacting the timeliness and accuracy of rescue measures.

[0009] In summary, the limitations of traditional fire monitor fire source location systems in complex environments are primarily reflected in the following aspects: an inability to capture dynamic changes in fire activity in real time, difficulty identifying hidden fire sources and the risk of re-ignition, a lack of multi-factor prediction capabilities, and lags in resource scheduling. These issues are particularly prominent in modern fire prevention and control, and urgently need to be addressed through the introduction of high-precision real-time data fusion, dynamic prediction models, and intelligent firefighting strategies. Summary of the Invention

[0010] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a fire source positioning system for unmanned aerial vehicle fire cannons based on infrared thermal imaging to solve the problems raised in the above background technology.

[0011] In order to achieve the above-mentioned purpose, the present invention is implemented through the following technical solutions: a fire source positioning system for a UAV fire cannon based on infrared thermal imaging, comprising:

[0012] Fire scene situation acquisition module, which acquires thermal image data representing the target fire scene and high-precision point cloud data of its spatial information;

[0013] The fire spread model construction module quantifies the spatial relationship between the air flow direction of the fire scene and the data collection points based on the point cloud data, and selects upstream collection points that are consistent with the fire source spread trend. Based on the thermal image data, the module considers the attenuation effect of the fire source radiation intensity at the upstream collection points as it changes with distance, calculates the thermal radiation background concentration of each upstream collection point, dynamically solves the fire source radiation intensity through gradient descent optimization, integrates the preset parameters of multiple diffusion model libraries, constructs a radiation intensity matrix, and generates a radiation intensity distribution map with geographic coordinates. This module uses the changes in radiation intensity to characterize the fire spread path of the target fire scene and form a processable fire source image.

[0014] The thermal image processing module performs morphological processing on the fire source image, performs convolution operations using a structural kernel that simulates the fluctuation characteristics of the fire source, and extracts key fire source feature data from the fire source image, including the fire source edge and diffusion direction, through radiation energy gradient analysis. This quantifies the fire spread trend and finally selects a pixel point in the fire source key feature data as the analysis starting point. The radiation energy correlation between each pixel point and the starting point is calculated to classify the fire spread risk level. Binarization mapping is performed based on the risk level to generate a structured fire source risk feature image, clarifying the fire front location and risk distribution.

[0015] The LSTM time series model processing module uses a gating mechanism to update the state, calculates the probability of fire spreading across levels based on the fire source risk characteristic image, and generates a smooth diffusion heat map to predict the spread direction and speed;

[0016] A battle map generation module that overlays diffusion heat maps onto actual geographic coordinates to achieve data visualization;

[0017] The UAV flight path planning module combines the fire area and diffusion probability to set the objective function and achieve comprehensive coverage of the fire cannon's strike path.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. To address the problem that existing technologies rely on static fire source data and have difficulty capturing the dynamic spatiotemporal evolution of fire, the present invention constructs a multi-dimensional fire spread model, integrates fire scene air flow vector projection analysis, dynamic superposition of thermal radiation background concentration, and LSTM time series modeling, to achieve accurate prediction of the fire spread path from low-risk areas to high-risk areas. Specifically, the present invention constructs a fire scene fluid perception unit to calculate the spatial projection factors of fire scene air flow vectors and point cloud collection points in real time. It dynamically screens the upstream fire source impact area based on the thermal radiation attenuation law and quantifies the multi-dimensional radiation contribution of the fire source to the spatial grid. Compared with the traditional fixed threshold model, the present invention realizes the dynamic superposition calculation of fire background concentration, solves the problem of misjudgment caused by sudden changes in fire scene airflow, and provides real-time data support for fire spread path prediction;

[0020] 2. Considering the risk that existing thermal imaging systems are susceptible to smoke interference, this invention uses a morphological processing unit to design a sine-Gaussian mixed structure kernel to simulate the fluctuation characteristics of fire sources and enhance local dynamic features. It also uses a dynamic threshold to binarize and segment active areas. Furthermore, based on the calculation of the multi-directional radiation energy change rate (longitude, latitude, and primary / secondary diffusion directions), it extracts the gradient characteristics of the fire front edge. By assigning weights to the radiation energy correlation, it divides the fire scene into three levels of high, medium, and low risk. This makes the system highly accurate in identifying hidden fire sources.

[0021] 3. Through real-time data processing and LSTM time series modeling, the present invention utilizes the infrared thermal imager and LiDAR equipment carried by drones to collect the thermal radiation distribution and three-dimensional spatial data of the fire scene in real time. The LSTM time series model is used to capture the dynamic changes of the fire. Combined with interpolation and smoothing to process the diffusion probability distribution, the system can quickly update the fire spread heat map, providing immediate support for firefighting operations.

[0022] 4. The present invention uses drones to collect thermal imaging and LiDAR point cloud data at a high frequency of 30Hz, dynamically annotates fire source levels, spread vector arrows, and safe channel flow guidance, and combines RTK-GPS centimeter-level positioning with particle systems to simulate fire monitor trajectories to generate a battle map containing spatial coordinates, timestamps, and temperature gradients, thereby greatly shortening firefighting response time. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0024] Figure 1 This is a schematic diagram of the structural framework of a fire source positioning system for a UAV fire-fighting cannon proposed in one embodiment of the present invention;

[0025] Figure 2 This is a visualization diagram of the radiation intensity matrix established to quantify the spatial impact of the fire source on each point cloud data collection point in the target fire scene, as proposed in one embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of the comprehensive radiation intensity distribution generated by analyzing the gradient change of the fire source radiation intensity column in the fire active area image proposed in one embodiment of the present invention;

[0027] Figure 4 Based on Figure 2 Schematic diagram of radiation impact values represented in the proposed radiation intensity matrix Figure 1 ;

[0028] Figure 5 Based on Figure 2 Schematic diagram of radiation impact values represented in the proposed radiation intensity matrix Figure 2 ;

[0029] Figure 6 This is a schematic diagram of using a path planning algorithm to ensure a safe and efficient flight strike path for a UAV, as proposed in one embodiment of the present invention;

[0030] Figure 7 This is a schematic diagram of performing morphological processing on a fire source image according to an embodiment of the present invention;

[0031] Figure 8 In one embodiment of the present invention, a schematic diagram of a battle map generation module is provided to intuitively display the direction of fire spread and potential risk areas. Figure 1 ;

[0032] Figure 9 In one embodiment of the present invention, a schematic diagram of a battle map generation module is provided to intuitively display the direction of fire spread and potential risk areas. Figure 2 . DETAILED DESCRIPTION

[0033] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0034] The present invention will be further described in detail below with reference to the accompanying drawings, but this does not limit the present invention.

[0035] like Figure 1 As shown in FIG. 1 , as one embodiment of the present invention, a fire source location system for a drone firefighting cannon based on infrared thermal imaging is proposed, comprising a data transmission control module, a fire scene situation acquisition module, a fire spread model construction module, a thermal image processing module, a time series model processing module, and a drone flight path planning module, all electrically connected to the data transmission control module. The data transmission control module utilizes 4G and / or 5G and / or satellite communication technologies to ensure stable and real-time data transmission.

[0036] In one embodiment of the present invention, a fire scene situation acquisition module, based on an infrared thermal imager and a laser radar (LiDAR) device carried by a drone, performs real-time scanning and situation data collection on a target fire scene. The situation data includes acquired thermal image data representing the thermal radiation distribution of the target fire scene and high-precision point cloud data representing the spatial information of the target fire scene. It can be understood that the infrared thermal imager detects the thermal radiation distribution of the fire scene in real time and generates a thermal image to provide basic data for the automatic identification of the fire source. The thermal radiation distribution is determined by the radiation intensity. In the field of infrared thermal imaging, the radiation intensity is generally used to describe the thermal radiation characteristics of the surface of an object. The radiation intensity can provide information about the temperature distribution of the surface of the object. Generally, high radiation intensity usually indicates a higher surface temperature of the object, while low radiation intensity indicates a lower temperature. The laser radar (LiDAR) performs a comprehensive scan of the fire scene and generates high-precision point cloud data for the subsequent construction of a digital elevation battle map and a fire spread model, thereby achieving comprehensive monitoring of the fire scene situation and providing precise support for subsequent fire source identification, path planning, and optimization of fire extinguishing strategies.

[0037] In one embodiment of the present invention, a fire spread model construction module is based on point cloud data, by quantifying the spatial impact of the fire source on each point cloud data collection point in the target fire scene, and taking into account the radiation intensity and fluctuation characteristics of the fire source. The fluctuation characteristic refers to the impact of the air flow in the fire scene on the spread or diffusion of the fire. That is, this module needs to evaluate the fluctuation characteristics of the fire source affected by the air flow in the fire scene, and then associate the point cloud data with the radiation intensity of the fire source to establish a comprehensive evaluation of the radiation intensity matrix including the full-dimensional influence relationship between the fire source, point cloud data collection points and fluctuation characteristics, so as to screen out the point cloud data collection point areas that are preferentially affected by the fire.

[0038] In specific implementation, the fire spread model construction module includes:

[0039] The fire scene fluid sensing unit quantifies the spatial impact of the fire source on the point cloud data collection points based on real-time fire scene air flow data to evaluate the impact of fire scene air flow on fire spread. The process is as follows:

[0040] First, for any temperature measurement point i and point cloud data collection point j, calculate its projection factor θ in the direction of the fire air flow ij , to determine the consistency of the directions of the two: , where the fire air flow vector Indicates the current fire scene air flow direction, the spatial vector from the temperature measurement point to the point cloud data collection point Indicates the spatial relationship between the two. Represents the vector length, which is used to measure the size of the vector. If the dot product result is positive, it means that the two vectors are in roughly the same direction. If it is negative, it means that the directions are roughly opposite. If it is zero, it means that the two vectors are perpendicular. In this embodiment, the projection factor θij Between -1 and 1, when θ ij When it is close to 1, it means that the possibility of fire spreading is greater; when θ ij When it is close to -1, it means that the direction is roughly opposite and the possibility of fire spreading is small; when θ ij When it is close to 0, it means that the impact of fire spread is small.

[0041] Secondly, the area where the point cloud data collection points with similar directions and distances within the preset threshold are located is selected as the upstream fire area. In specific implementation, the maximum distance threshold R is set based on the thermal radiation intensity of the temperature measurement point and the fire environment conditions. max , to limit the maximum impact distance between the temperature measurement point and the point cloud data collection point, set the distance between the temperature measurement point and the point cloud data collection point <R max And the projection factor θ ij When the temperature is > cos45°, the point cloud data collection point j is marked as the upstream point cloud data collection point of the temperature measurement point i, and the area where the collection point is located is recorded as the upstream fire area, thereby ensuring that the subsequent fire spread model can accurately identify the area that the fire may first affect, thereby providing a data basis for subsequent fire spread prediction and drone fire cannon fire extinguishing strategy.

[0042] Finally, for each point cloud data collection point j, the temperature measurement point set Ω in the upstream fire area is screened out. ij , and calculate the background concentration of fire heat radiation B in this area j It is understandable that the background concentration of fire heat radiation B j It is used to reflect the comprehensive impact of multiple temperature measurement points on the point cloud data collection point, which is determined by the radiation intensity E of each fire source Si. i Generate by superposition according to the diffusion law: It can be understood that the fire source Si is the actual source of thermal radiation in the fire scene, and its position and radiation intensity determine the fluctuation characteristics and impact range of the fire, while the temperature measuring point i is the point in the fire scene used to monitor the thermal radiation intensity. It can be a high-temperature area in the thermal image or other points with thermal radiation characteristics. Each temperature measuring point i contains its coordinates (xi, yi, hi) and thermal radiation intensity. In specific implementation, the radiation intensity and position of the fire source Si can be inferred through the data of the temperature measuring point i. The thermal radiation intensity of the temperature measuring point i reflects the heat release of the fire source Si. Generally, the stronger the thermal radiation of the fire source Si, the higher the thermal radiation intensity of the surrounding temperature measuring points i. By monitoring the dynamic changes of the temperature measuring points, the position of the fire source Si can be located. In the above formula, is the initial thermal radiation intensity of the fire source Si at the point cloud data collection point j, is the temperature measurement point set Ω of point cloud data collection point j in the upstream fire area ijThe size of represents the number of temperature measurement points in the set. is the horizontal distance between the fire source Si and the point cloud data collection point j, is the vertical distance between the fire source Si and the point cloud data collection point j, is the horizontal attenuation and / or diffusion coefficient of the radiation intensity of the current fire source Si, which is determined through experiments or numerical simulations to more accurately reflect the actual fire conditions. is the current diffusion coefficient in the vertical direction, and exp is an exponential function, which is used to calculate the attenuation of the radiation intensity of the fire source Si at the current fire scene with distance.

[0043] In one embodiment of the present invention, the fire spread model building module further includes: a fire source radiation intensity calculation unit, based on the obtained fire thermal radiation background concentration B j , the radiation intensity E of the fire source Si at each point cloud data collection point j in the upstream fire area is obtained through iterative optimization dynamic solution i , in order to accurately quantify the thermal radiation impact of the fire source on the surrounding areas except the upstream fire area, and provide a data basis for fire spread prediction and high-risk area demarcation. It should be noted that the purpose of this unit is to achieve dynamic calculation of fire source radiation intensity, solve the prediction deviation problem caused by traditional models ignoring the background concentration of thermal radiation and multi-source coupling effects, and provide core parameter support for dynamic grid division of the fire scene and precise dispatch of firefighting resources. The process is as follows:

[0044] First, based on the fire heat radiation background concentration B output by the fire scene fluid sensing unit j (Reflecting the comprehensive impact of multiple fire sources), through the minimum predicted thermal radiation concentration With the measured The difference between the two is used to construct the target optimization function to solve the optimal fire source radiation intensity. : , where is the sparse constraint coefficient, which is used to suppress noise interference and prevent overfitting. M is the number of temperature measurement points i, and N is the number of point cloud data collection points.

[0045] Secondly, based on the fire thermal radiation background concentration B at the temperature measurement point i j Maximum value, initial radiation intensity of the assigned fire source Si , and use gradient descent iteration to calculate the minimum predicted thermal radiation concentration and residuals , by controlling the step size of each iteration, the radiation intensity E is updated i ,when When it is less than the set threshold, the iteration is terminated and the optimized radiation intensity of all fire sources Si is output. The minimum predicted thermal radiation concentration refers to the minimum predicted value of the background thermal radiation intensity calculated by the model, which is used to set the threshold or as a benchmark value to help identify the impact of the fire source on each collection point.

[0046] In one embodiment of the present invention, the fire spread model construction module also includes: a radiation intensity matrix construction unit, which binds the fire source radiation intensity solution result with the spatial coordinates, establishes a comprehensive evaluation of the radiation intensity matrix including the fire source, point cloud data collection points and the full-dimensional influence relationship of the fluctuation characteristics, directly quantifies the thermal radiation impact of the fire source Si on the upstream fire and surrounding areas, and screens out the point cloud data collection point areas that are preferentially affected by the fire, so as to provide a planning basis for the subsequent UAV fire cannon attack path. Figure 2 As shown, in the radiation intensity matrix, the fire impact intensity is represented by gradient colors, and the impact value is mapped using transparency. The brighter the fire, the greater the impact. The process is as follows:

[0047] First, the convection diffusion coefficient σ preset by the K diffusion model library built by the system (such as CFD, GIS-based fire spread model) d (k) , the number of temperature measurement points M and the number of point cloud data collection points N, define the radiation intensity matrix dimension, and then construct the radiation intensity matrix C∈R M×N×K , where the element C of the radiation intensity matrix C is ijk It represents the radiation impact value of the fire source Si on the point cloud data collection point j under the k-th diffusion model library: , where are the diffusion coefficients of the kth diffusion model library in the x, y, and z directions, d j ,d i are the coordinates of point cloud data collection point j and fire source Si, d j -d i is the distance difference between the cloud data collection point j and the fire source Si in all directions.

[0048] Second, for each element C of the radiation intensity matrix C ijk , superimpose the radiation impact values of the K diffusion model libraries to form a complete radiation intensity matrix C. Finally, all point cloud data collection points are sorted according to the integrated radiation impact values, such as from high to low, and a radiation impact threshold is selected. The point cloud data collection point areas with radiation impact values greater than the threshold are screened out in turn, and a radiation intensity distribution map is generated as a whole. It can be understood that this radiation intensity distribution map has geographic coordinates. At this point, the fire spread path of the target fire scene is characterized by the change in radiation intensity, and a processable fire source image can be formed.

[0049] Based on the above technical concept, it can be understood that by superimposing the radiation impact values of K diffusion model libraries, a complete radiation intensity matrix C∈R is formed. M×N×K , so that the matrix integrates the radiation effect of the fire source Si on each point cloud data collection point under different diffusion models. Each element C in the matrix ijk It represents the radiation impact value of the fire source Si on the point cloud data collection point j under the kth diffusion model, such as Figure 4-Figure 5 As shown, it contains information such as the location of the fire source, radiation intensity, and fluctuation characteristics. In subsequent processing, by projecting the three-dimensional matrix onto a two-dimensional plane, a radiation intensity distribution map that comprehensively represents the target fire source image data can be generated. For example, binary segmentation is performed based on this map to guide the adaptive focusing of the thermal image and the analysis of the fire scene characteristics, extract the outline of the high-risk area, ensure that the high-radiation area is given priority and clearly presented, and thus guide the attack path of the drone fire cannon. For ease of understanding, in this embodiment, the area screened out is the "priority area" determined by the radiation intensity matrix C. The purpose is to first process the areas in the fire scene that are more affected by the fire source. The active area is within the screened priority area, and the radiation impact value is higher than the first threshold T fir The area indicates the part with the strongest fire source impact. The inactive area is within the priority area selected, and the radiation impact value is lower than the first threshold T fir The area indicates the part with weaker influence of fire source, which aims to further refine the analysis of the spread trend of fire source and provide the basis for fire fighting strategy and resource scheduling.

[0050] In one embodiment of the present invention, the proposed infrared thermal imaging-based UAV fire cannon fire source positioning system, in which the thermal image processing module is used, aims to adjust the minimum focal length of the infrared thermal imager to ensure that the area of the point cloud data collection points with radiation impact values greater than the threshold is always within the imaging range. Through morphological processing, the diffusion characteristics of the fire source image in the area are extracted, and then based on the multi-directional radiation energy change rate calculation, a binary fire source risk feature image containing high, medium and low risk levels of the fire scene is generated, providing structured input data for subsequent fire analysis and fire extinguishing decisions.

[0051] In specific implementation, the thermal image processing module includes:

[0052] The imaging focus adjustment unit calculates the minimum focal length based on the position and range of the screened area to ensure that the areas where these important fire sources are located are within the imaging range. The process is as follows: obtain the coordinates of the fire source Si, the point cloud data collection point j, and the field of view angle θ of the infrared thermal imager FOV , calculate the minimum focal length f min , , where d min is the minimum distance between the fire source Si and the point cloud data collection point j, h maxIt is the maximum height that can be accommodated on the imaging plane of the infrared thermal imager to optimize the accuracy and efficiency of fire source Si monitoring.

[0053] In one embodiment of the present invention, the thermal image processing module further includes: a morphological processing unit for performing morphological processing on the fire source image of the screened area, such as Figure 7 As shown, the process is as follows:

[0054] First, the radiation influence value of each pixel (p, q) in the fire source image of the screened area and the weight of the area determined based on the actual fire conditions are obtained through the radiation intensity matrix C. It should be noted that the radiation influence value of each pixel directly reflects the thermal radiation intensity of the location. High radiation influence values usually correspond to fire sources or high-temperature areas, while low radiation influence values correspond to non-fire sources or low-temperature areas. The weighted average formula is used to combine the radiation influence value and the area weight to calculate the basic radiation influence mean E of the fire source image in the area. mean ; Based on the basic radiation impact mean E mean and adjustment factor r, set the first threshold T fir , T fir =E mean ×r, where the adjustment factor r is set according to the dynamic characteristics of the fire scene (such as wind speed, type of burning material), usually r∈[1.2,1.5]; the radiation impact value is compared with the first threshold T fir Compare, if the radiation impact value is greater than the first threshold T fir , then the filtered area is marked as the active area (output value is 1), otherwise it is marked as the inactive area (value is 0). It can be understood that through the above binarization process, such as Figure 3 As shown in the figure, the generated complex comprehensive radiation intensity distribution can be simplified into two types of areas: active and inactive areas, highlighting the main affected areas of the fire source Si, which is convenient for the subsequent rapid identification of the diffusion trend of the fire source Si.

[0055] Secondly, a structural kernel K is defined to simulate the fluctuation characteristics of the fire source Si. In this embodiment, the structural kernel K is an m×n matrix, and the element value k in the matrix is pq Calculated by the following formula: , where Z is the normalization constant to ensure that the sum of the element values of the structure kernel K is 1, α is the sinusoidal modulation amplitude (usually α = 0.3), which is used to simulate the fluctuation of the fire source, and λ x and λ xare the sinusoidal wavelengths in the x and y directions, respectively, which are related to the scale of fire diffusion. σ is the Gaussian attenuation coefficient (usually σ = 1.5), which controls the spatial attenuation range of the structure kernel. m is the number of rows of the structure kernel K, which is used to define the size of the structure kernel K in the vertical direction and affects the capture range of vertical features of the image during the convolution operation. n is the number of columns of the structure kernel K, which is used to define the size of the structure kernel in the horizontal direction and affects the capture range of horizontal features of the image during the convolution operation. p is the index of the structure kernel K in the row direction, and its value range is usually −a to a, where a is the radius of the convolution kernel. q is the index of the structure kernel K in the column direction, and its value range is usually −b to b, where b is the radius of the convolution kernel. It can be understood that through the sinusoidal modulation term, the structural kernel K can simulate the fluctuation characteristics of the fire source Si, which helps to capture the dynamic changes of the fire source in morphological processing. The Gaussian attenuation term ensures that the structural kernel K has a certain attenuation characteristic in space, so that the processing process can focus on local areas (active and inactive areas) while reducing the influence of distant pixels. Through the above operations, key features in the fire source image, such as edges and diffusion directions, can be extracted.

[0056] Again, the structure kernel K is used to perform convolution calculation on the image of the active fire area: , where a and b are the convolution kernel radii, respectively. Generally, a=(m-1) / 2, b=(n-1) / 2 are taken. is the pixel value of the corresponding position in the image of the active fire area. The pixel value is updated according to the convolution result. If C ij If the value of the pixel is greater than the set threshold, the value of the pixel is set to 1, otherwise it is set to 0. The above threshold is usually set to 50% of the total energy of the structural kernel, such as 0.5.

[0057] Finally, by calculating the cumulative radiation energy value of each column of the fire active area image, the gradient change formula is used , e ij is the radiation impact value of the pixel in the i-th column and j-th row, k is the number of rows, and the gradient change of the fire source radiation intensity column in the fire active area image is analyzed. Based on this change, the key features of the edge and diffusion direction of the suspected fire source area in the fire active area image are determined to form the fire source key feature data containing this key feature.

[0058] In one embodiment of the present invention, the thermal image processing module further includes a multi-directional radiation energy change rate calculation unit, which calculates the radiation energy change rate of the fire source's key characteristic data in longitude, latitude, and in a primary diffusion direction related to the fire source's fluctuating characteristics (derived from statistical analysis of the diffusion directions of each fire source in the radiation intensity matrix C) and a secondary diffusion direction (at a specific angle relative to the primary diffusion direction, which is related to the fire source's irregularity), thereby identifying changes in the fire source's diffusion intensity. Specifically, the radiation energy change rate is calculated by dividing the energy difference between the current pixel and the pixel to its right, the current pixel and the pixel below it, or the current pixel and the next pixel along the direction by the total energy.

[0059] In one embodiment of the present invention, the thermal image processing module further includes: a region division unit: first, performing pre-processing operations such as denoising and contrast enhancement on the captured key characteristic data of the fire source to improve image quality; second, selecting a pixel point in the key characteristic data of the fire source as an analysis starting point, and calculating the radiation energy correlation between each pixel point and the starting point; , it can be understood that the weight coefficient w i It is related to the distance of each pixel from the starting point and the weight of the area where it is located. Ei is the radiation influence value of the adjacent pixels, and m is the number of adjacent pixels considered when calculating the radiation energy correlation. Finally, the radiation energy statistics of the high-risk area in the historical fire data are taken to set the second threshold T sec , in proportion to T mid =k×T sec , determine the intermediate threshold T mid , k is the empirical coefficient, the value is 0.7, if p ≥ T sec The key characteristic data of the fire source are divided into the fire active core area, corresponding to the high radiation intensity and rapid spread risk level. If p≤<T sec In the key characteristic data of the fire source, it is divided into weak combustion or safe area, corresponding to the low risk level. If T mid <p<T sec It is divided into a stable combustion zone in the key characteristic data of the fire source, corresponding to the controllable spread medium risk level.

[0060] In one embodiment of the present invention, the thermal image processing module further includes: a fire source risk characteristic image generation unit: a binary mapping is performed according to the risk level. In this embodiment, pixels in high risk level areas are marked as V high-risk =1 (such as white), and the rest of the area is marked as V low-risk =0 (such as black), a binary fire source risk feature image containing high, medium, and low risk levels of the fire scene is generated, and the fire front position is marked.

[0061] Based on the above technical concepts, it is understandable that in existing fire monitoring and prediction technologies, traditional methods often rely on static fire source location and intensity data, which makes it difficult to capture the dynamic changes and spread trends of the fire. Since the spread of fire has a high degree of temporal and spatial complexity, a single static model cannot fully reflect the dynamic characteristics of the fire scene, resulting in insufficient accuracy and timeliness of the prediction results.

[0062] Therefore, in one embodiment of the present invention, the proposed LSTM time series model processing module in the fire source positioning system of the UAV fire cannon based on infrared thermal imaging aims to combine the binary fire source risk characteristic image (providing the static risk distribution of the fire scene) and the LSTM time series model (capturing the dynamic characteristics of the fire spread) to further predict the spread direction and speed of the fire from the low-risk area to the medium-risk area and then to the high-risk area, thereby achieving more accurate fire spread simulation and fire extinguishing path planning.

[0063] In specific implementation, the LSTM time series model processing module includes:

[0064] The image spatiotemporal alignment and fusion unit uses motion-compensated optical flow to eliminate inter-frame displacement of images sampled by the infrared thermal imager carried by the drone, reduce the influence of smoke and dust interference in the fire scene, achieve sub-pixel alignment, ensure that N frames of binary fire source risk feature images are accurately spatially corresponding, and perform spatiotemporal fusion of the N aligned frames to form a clear and stable fused image data. , where is the time attenuation weight, which reflects the contribution of the binary fire source risk feature image at different time t to the fusion result. The image closer to the current time t has a higher weight. is a Gaussian filter kernel, which is used to effectively suppress random noise and transient interference, making the fused image clearer and more stable, thereby reducing the impact of interference on fire source spread detection. fuesd (x, y) is the pixel value of the fused image data at position (x, y), λ is the attenuation coefficient, τ is the time step, I t (x,y) is the pixel value of the t-th frame image at position (x,y).

[0065] In one embodiment of the present invention, the LSTM time series model processing module further includes: a data set partitioning unit that divides the fused image data into a training subset, a test subset, and a residual assessment (RDA) subset to support the training, validation, and performance evaluation of the LSTM time series model. It should be noted that in this embodiment, the divided training and test subsets contain at least the following data related to the processing of binary fire source risk feature images:

[0066] Image ID information: used to uniquely identify each binary fire source risk characteristic image, facilitating data management and traceability; spatial location information: including the coordinate position of the fire source Si in the binary fire source risk characteristic image and the size of the fire source (such as area and perimeter), facilitating the determination of the specific location and scale of the fire source Si; infrared thermal imager parameter information: such as focal length, aperture, and exposure time; environmental information collected based on peripheral sensors: such as ambient temperature, humidity, and wind speed; the RDA subset contains the actual hazard level assessment results of the test subset, which is used to compare the test results of the test subset to evaluate the accuracy of the model.

[0067] In one embodiment of the present invention, the LSTM time series model processing module also includes: an LSTM time series model training unit, which adopts a dual-branch YOLO-Mask architecture to improve the generalization ability and adaptability of the model. In this embodiment, the detection branch uses DarkNet-53 to extract image depth features and outputs the fire source bounding box and its confidence. The segmentation branch uses the feature pyramid network (FPN) to generate pixel-level masks to accurately segment the fire source area, and improves the accuracy of detection and segmentation by optimizing the bounding box loss function and the Dice coefficient loss function.

[0068] The time series modeling unit extracts the time series features Tt such as the temperature mean, area change rate, and combustion product distribution of the low-risk area from the binary fire source risk feature image and updates the gate state: , where f t is the output of the forget gate, which determines the state of the cell C t-1 Which information is discarded in C t is the updated unit state, which combines the output of the forget gate and the input gate. ot is the output of the output gate, which is used to determine the final output value. h t is the hidden state, representing the output of the LSTM timing model unit, which is used to pass to the next time step, w f Corresponding to the forget gate weight, b f is the forget gate bias vector, σ is the sigmoid function, which is used to limit the output between 0 and 1. is element-wise multiplication, h t-1 is the hidden state at the previous moment;

[0069] Output the probability of fire spreading across risk levels , represents the possibility of fire spreading from low-risk areas to high-risk areas, where W p Used to represent the hidden state h t The influence of each time series feature on the diffusion probability, b p is the bias vector used to calculate the diffusion probability. When P tWhen the set threshold (taken as 0.65) is exceeded, an early warning is triggered, indicating that the fire is spreading from the low-risk area to the high-risk area.

[0070] In one embodiment of the present invention, the LSTM time series model processing module further includes: a fire spread prediction output unit, which generates a diffusion heat map according to the diffusion probability. In this embodiment, the diffusion heat map is generated by the following steps:

[0071] First, the diffusion probability P is calculated based on the LSTM time series model. t , modeling the spatial distribution of the possibility of fire spread; secondly, using interpolation algorithms (such as Gaussian interpolation) to smooth the probability distribution, generate a continuous diffusion probability grid, and then combine it to generate a diffusion heat map.

[0072] In one embodiment of the present invention, the proposed UAV firefighting cannon fire source positioning system based on infrared thermal imaging also includes:

[0073] The battle map generation module electrically connected to the data transmission control module is used to output the fire spread prediction unit, such as by color mapping, to identify high-risk areas (probability P t >0.65) are marked in red, and low-risk areas are represented in green or blue to intuitively show the direction of fire spread and potential risk areas. GIS visualization tools (such as QGIS or ArcGIS) or Python visualization libraries (such as Matplotlib or Seaborn) are used to overlay the diffusion heat map onto the actual geographic coordinates, and arrows are used to indicate the diffusion direction to form the final diffusion trend visualization operation map. It should be noted that in this embodiment, if Figure 6 、 Figure 8 As shown in the figure, the visual combat map generated by the drone's flight path is achieved through the following steps:

[0074] a. Use the infrared thermal imager on the drone to capture thermal images at a speed of 30 frames per second (the drone position is as follows Figure 9 At the same time, each frame of the image is bound to the RTK-GPS centimeter-level positioning information and precise timestamp, and is updated to the time display area on the system page according to the preset specified format. The setInterval function is called every 1 second to achieve real-time time updates;

[0075] b. In the window.onload event: Get <canvas>Elements and their 2D drawing context information, and define a series of combat map drawing functions, which at least include:

[0076] drawInfraredHeatMap is used to construct the infrared diffusion heat map. In the specific implementation, the infrared diffusion heat map data is assumed to be a two-dimensional array, where each element represents the HSV value of a pixel. The HSV value is converted to RGB value using the hsvToRgb function, and then a rectangle is filled to represent the pixel.

[0077] hsvToRgb, used to implement the conversion from HSV color space to RGB color space, used to convert the HSV data of the infrared diffusion heat map into drawable RGB colors;

[0078] drawLiDARPointCloud is used to construct LiDAR point cloud data. Assume that the LiDAR point cloud data is a two-dimensional array, where each element represents the grayscale value of a point. The point cloud data is drawn by filling rectangles with different grayscales.

[0079] drawFire is used to construct the fire source area, fire monitor and its firing range, and safe passage. In specific implementation, fire source elements are generated according to the input fire source location, size and temperature information. For fire sources with a temperature greater than 800℃, the simulation is as follows Figure 9 The display effect shown by mark B1 in the figure is simulated as follows for a fire source of 600-800℃ Figure 9 The display effect shown in the mark B2 is simulated as follows for a fire source < 600℃ Figure 9 The display effect shown in the mark B3 in the middle renders the fire level in real time, intuitively showing the degree of fire danger; generates fire monitor elements (such as a circular icon and a dotted circular icon indicating the launch range); generates a vector arrow representing the spread of the fire based on the input start and end coordinates, as well as the speed and temperature information, where the arrow length is calculated based on the speed and temperature gradient, and the color is set according to the temperature, following the mapping basis , generating a streamer guide for a safe channel (e.g. Figure 9 As shown by the mark D in the figure), in specific implementation, the streamer effect is represented by setting the dotted line style. In the formula, H represents the hue, which is a parameter in the color space and is used to describe the type of color (such as red, green, blue, etc.). The general hue range is usually 0 to 360, corresponding to the angle on the hue circle. T is the temperature, which is the input temperature value (in degrees Celsius). By creating a certain number of particles, the particle trajectory of the fire monitor attack path is generated (such as Figure 9 The particle positions are set to be updated continuously to achieve a dynamic impact trajectory effect.

[0080] c. Call the above series of combat map drawing functions to ensure that the firing range of the fire monitor is larger than the fire source area, that is, the fire monitor's attack coverage radius meets the radius of the fire source's circumscribed circle of 1.5.

[0081] In one embodiment of the present invention, the UAV fire source positioning system based on infrared thermal imaging is proposed, in which the UAV flight path planning module is used to generate a battle map and combine the fire area A with the fire source positioning system. j And the diffusion probability P of the fire area t , set the optimal resource scheduling objective function and use the path planning algorithm (such as Dijkstra algorithm) to ensure that the drone can achieve a safe and efficient flight strike path and achieve full coverage of the fire area. The objective function is set as follows: , constraints: , where a ij It represents the suppression efficiency or fire extinguishing effect of unit resource i on fire area j, and is used to reflect the fire extinguishing effect of different resources in different fire areas. j =α×A j ×P t , which is used to reflect the actual danger level and resource demand of the fire area j, and α is the adjustment coefficient, which is used to control the effect of the fire area and the diffusion probability on the demand d j The combined effect of ρ is the penalty factor used to balance the objective function, C i is the cost coefficient of unit resource i, X i is the amount of resource i per unit, R total is the total resource limit, m is the type or quantity of available resources, and n is the number of fire areas.

[0082] It should be noted that a ij It represents the fire extinguishing effect of using one unit of resource i on fire area j. The specific understanding is as follows:

[0083] Define resource types: Resource i can be fire extinguishing agents such as water, foam, and dry powder, or rescue forces such as firefighters, fire trucks, and drones;

[0084] Fire area characteristics: The characteristics of fire area j include flame temperature, fuel type, fire spread speed, etc.

[0085] Example: Assume that the fire area j is a forest fire, the fire is fierce, and the flame height reaches 5 meters. There are two resources: Resource 1: Water, each liter of water can reduce the flame height by 0.1 meters, then α 1j =0.1 m / L; Resource 2: Fire truck, each fire truck can spray 100 liters of water per minute, then α 2j =100×0.1=10 m / min.

[0086] At the same time, when evaluating the fire area in the fire protection field, the parameter d is used. j Indicates the actual danger level and resource requirements of the fire area j. When it is implemented, A j Represents the area of the fire area j, usually in square meters (m 2 ) is quantified as a unit, for example, if the fire area j is a 100 square meter fire area, then A j =100m 2 Secondly, P t It represents the probability of fire spread corresponding to time step t. It is a dimensionless value ranging from 0 to 1. For example, based on a comprehensive assessment of factors such as the spread speed of the fire and wind speed, it is predicted that the probability of the fire spreading outward from area j in the next hour (time step t) is 30%, then P t =0.3. α is the adjustment coefficient, which is used to comprehensively consider the impact of fire area and diffusion probability on demand d j The degree of influence is generally determined based on actual experience or experimental data, and its value is also between 0 and 1. Assuming that in a certain fire assessment, α is determined to be 0.5, these quantified values are substituted into the formula to calculate d j . Taking the above example again, d j =α×A j ×P t =0.5×100m 2 ×0.3=15m 2 (In this example, the units are simplified for the sake of convenience. Specific implementation requires unit conversion or dimensionless processing.) j =15 reflects the actual danger level and resource requirements of fire area j. The larger the value, the more dangerous the fire area is and the more firefighting resources are needed, such as more firefighters, more fire extinguishing agents, and larger firefighting equipment.

[0087] Based on the above technical concept, it can be understood that the cost of unit resource i is mainly concentrated in consumables, as shown in Table 1, which is a basic corresponding data table of unit resources, input volume and total resources.

[0088] Table 1

[0089]

[0090] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.< / canvas>

Claims

1. The fire source positioning system of UAV fire cannon based on infrared thermal imaging is characterized by: include: Fire scene situation acquisition module, which acquires thermal image data representing the target fire scene and high-precision point cloud data of its spatial information; The fire spread model construction module quantifies the spatial relationship between the air flow direction of the fire scene and the data collection points based on the point cloud data, and selects upstream collection points that are consistent with the fire source spread trend. Based on the thermal image data, the module considers the attenuation effect of the fire source radiation intensity at the upstream collection points as it changes with distance, calculates the thermal radiation background concentration of each upstream collection point, dynamically solves the fire source radiation intensity through gradient descent optimization, integrates the preset parameters of multiple diffusion model libraries, constructs a radiation intensity matrix, and generates a radiation intensity distribution map with geographic coordinates. This module uses the changes in radiation intensity to characterize the fire spread path of the target fire scene and form a processable fire source image. The thermal image processing module performs morphological processing on the fire source image, performs convolution operations using a structural kernel that simulates the fluctuation characteristics of the fire source, and extracts key fire source feature data from the fire source image, including the fire source edge and diffusion direction, through radiation energy gradient analysis. This quantifies the fire spread trend and finally selects a pixel point in the fire source key feature data as the analysis starting point. The radiation energy correlation between each pixel point and the starting point is calculated to classify the fire spread risk level. Binarization mapping is performed based on the risk level to generate a structured fire source risk feature image, clarifying the fire front location and risk distribution. The LSTM time series model processing module uses a gating mechanism to update the state, calculates the probability of fire spreading across levels based on the fire source risk characteristic image, and generates a smooth diffusion heat map to predict the spread direction and speed; A battle map generation module that overlays diffusion heat maps onto actual geographic coordinates to achieve data visualization; The UAV flight path planning module combines the fire area and diffusion probability to set the objective function and achieve comprehensive coverage of the fire cannon's strike path.

2. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1 is characterized by: The fire spread model building blocks include: The fire scene fluid sensing unit quantifies the spatial impact of the fire source on the point cloud data collection points based on real-time fire scene air flow data. The process is as follows: First, for any temperature measurement point i and point cloud data collection point j, calculate the spatial vector projection factor of the two in the direction of the fire air flow to determine the consistency of their directions; Secondly, the area where the point cloud data collection points with similar directions and distances within the preset threshold are located is selected as the upstream fire area; Finally, for each point cloud data collection point j, the temperature measurement point set Ω in the upstream fire area is screened out. ij , combined with the attenuation effect of the fire source radiation intensity in the horizontal and vertical directions as the distance changes, the radiation impact of each fire source Si on the upstream collection point is dynamically superimposed to generate the fire thermal radiation background concentration B in this area j : , where is the initial thermal radiation intensity of the fire source Si at the point cloud data collection point j, is the temperature measurement point set Ω of point cloud data collection point j in the upstream fire area ij The size of is the horizontal distance between the fire source Si and the point cloud data collection point j, is the vertical distance between the fire source Si and the point cloud data collection point j, is the attenuation and / or diffusion coefficient of the radiation intensity of the current fire source Si in the horizontal direction, is the diffusion coefficient in the vertical direction, and exp is the exponential function.

3. The fire source positioning system for UAV firefighting cannon based on infrared thermal imaging according to claim 1 or 2, characterized in that: The fire spread model building module also includes: Fire source radiation intensity calculation unit, based on the obtained fire thermal radiation background concentration B j The fire radiation intensity E of the fire source Si at each point cloud data collection point j in the upstream fire area is obtained through iterative optimization dynamic solution i , the process is as follows: First, the minimum predicted thermal radiation concentration With the measured The difference between the two is used to construct the target optimization function to solve the optimal fire source radiation intensity. : , where is the sparse constraint coefficient, which is used to suppress noise interference and prevent overfitting. M is the number of temperature measurement points i, and N is the number of point cloud data collection points. Secondly, based on the fire thermal radiation background concentration B at the temperature measurement point i j Maximum value, initial radiation intensity of the assigned fire source Si , and use gradient descent iteration to calculate the minimum predicted thermal radiation concentration and residuals , by controlling the step size of each iteration, the radiation intensity E is updated i ,when When it is less than the set threshold, the iteration is terminated and the optimized radiation intensity of all fire sources Si is output. .

4. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 3 is characterized by: The fire spread model building module also includes: The radiation intensity matrix construction unit quantifies the radiation impact of the fire source on the upstream fire area and generates a radiation intensity distribution map representing the fire source image. The process is as follows: First, the convection diffusion coefficient σ preset by the K diffusion model library built by the system d (k) , the number of temperature measurement points M and the number of point cloud data collection points N, define the radiation intensity matrix dimension, and then construct the radiation intensity matrix C∈R M×N×K , where the element C of the radiation intensity matrix C is ijk It represents the radiation impact value of the fire source Si on the point cloud data collection point j under the k-th diffusion model library: , where are the diffusion coefficients of the kth diffusion model library in the x, y, and z directions, d j ,d i are the coordinates of point cloud data collection point j and fire source Si, d j -d i is the distance difference between the cloud data collection point j and the fire source Si in all directions; Second, for each element C of the radiation intensity matrix C ijk , superimpose the radiation impact values of K diffusion model libraries to form a complete radiation intensity matrix C; Finally, all point cloud data collection points are sorted according to the integrated radiation impact value, and a radiation impact threshold is selected. The point cloud data collection point areas with radiation impact values greater than the threshold are screened out in turn to generate a radiation intensity distribution map as a whole.

5. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1 is characterized in that: Thermal image processing module includes: The morphological processing unit performs morphological processing on the fire source image, and the process is as follows: First, the radiation intensity matrix C is used to obtain the radiation impact value of each pixel (p, q) in the fire source image and the weight of the area where it is located, and the basic radiation impact mean E of the fire source image in the area is calculated. mean ; Based on the basic radiation impact mean E mean and adjustment factor r, set the first threshold T fir , the radiation impact value and the first threshold T fir Compare, if the radiation impact value is greater than the first threshold T fir , then the area is marked as an active fire area, otherwise it is marked as an inactive fire area; Secondly, a structural kernel K is defined to simulate the wave characteristics of the fire source Si, and its element value k pq Calculated by the following formula: , where Z is the normalization constant, α is the sinusoidal modulation amplitude, and λ x and λ x are the sinusoidal wavelengths in the x and y directions respectively, σ is the Gaussian attenuation coefficient, m is the number of rows of the structure core K, n is the number of columns of the structure core K, p is the index of the structure core K in the row direction, and q is the index of the structure core K in the column direction; Again, the structure kernel K is used to perform convolution calculation on the image of the active fire area: , where a and b are the convolution kernel radii, is the pixel value of the corresponding position in the image of the active fire area. The pixel value is updated according to the convolution result. If C ij If the value is greater than the set threshold, the pixel value is set to 1, otherwise it is set to 0; Finally, by calculating the cumulative radiation energy value of each column of the fire active area image, the gradient change formula is used , analyze the gradient change of the fire source radiation intensity column in the fire active area image, and determine the key features of the edge and diffusion direction of the suspected fire source area in the fire active area image based on this change, and form the fire source key feature data containing this key feature, where, e ij is the radiation impact value of the pixel in the i-th column and j-th row, and k is the number of rows.

6. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1 is characterized by: The specific process of classifying fire spread risk levels is as follows: First, the radiation energy change rate of the key characteristic data of the fire source in longitude, latitude, and the main diffusion direction and secondary diffusion direction related to the fluctuation characteristics of the fire source is calculated to identify the change in the diffusion intensity of the fire source; Secondly, a pixel point is selected as the starting point for analysis in the key characteristic data of the fire source, and the radiation energy correlation degree p between each pixel point and the starting point is calculated. , where w i is the weight coefficient, Ei is the radiation influence value of the adjacent pixels, and m is the number of adjacent pixels considered when calculating the radiation energy correlation; Finally, take the radiation energy statistics of high-risk areas in the historical fire scene data and set the second threshold T sec , in proportion to T mid =k×T sec , determine the intermediate threshold T mid , k is the empirical coefficient, where If p ≥ T sec The key characteristic data of the fire source are divided into the core area of active fire, corresponding to the high radiation intensity and rapid spread risk level; If p≤T sec Then it is classified as weak combustion or safe area in the key characteristic data of the fire source, corresponding to the low risk level; If T mid <p<T sec It is divided into a stable combustion zone in the key characteristic data of the fire source, corresponding to the controllable spread medium risk level.

7. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 5 is characterized by: Before performing morphological processing on the fire source image, it is necessary to calculate the minimum focal length based on the position and range of the screened area to ensure that the fire source area is always within the imaging range. The process is as follows: Obtain the coordinates of the fire source Si, the point cloud data collection point j, and the field of view angle θ of the imaging device FOV , calculate the minimum focal length f min , , where d min is the minimum distance between the fire source Si and the point cloud data collection point j, h max It is the maximum height that can be accommodated on the imaging plane of the imaging equipment to optimize the accuracy and efficiency of fire source Si monitoring.

8. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1 is characterized by: The LSTM time series model processing module is implemented as follows: First, the inter-frame displacement of the images sampled by the imaging equipment carried by the UAV is eliminated by the motion-compensated optical flow method to ensure that the N frames of binary fire risk characteristic images are accurately spatially corresponding. The N frames of aligned images are fused in the spatiotemporal domain to form the fused image data. , where is the time decay weight, is the Gaussian filter kernel, I fuesd (x, y) is the pixel value of the fused image at position (x, y), λ is the attenuation coefficient, τ is the time step, I t (x,y) is the pixel value of the t-th frame image at position (x,y); Secondly, the fused image data is divided into training set, test set and evaluation set to construct a time series dataset with multi-dimensional attributes; Thirdly, a dual-branch network architecture of detection and segmentation is adopted to simultaneously extract the fire source bounding box features and pixel-level masks, optimizing the model's adaptability to complex fire scenes. Finally, the memory unit is dynamically updated through the LSTM gating mechanism to calculate the probability of fire spreading across risk levels; A diffusion heat map is generated based on the probability distribution, and after interpolation and smoothing, a threshold is combined with the warning to generate a continuous diffusion probability grid.

9. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1, characterized in that: The objective function set is the optimal resource scheduling objective function, wherein the objective function set based on the UAV flight path planning module is as follows: , constraints: , where a ij represents the suppression efficiency or fire extinguishing effect of unit resource i on fire area j, d j =α×A j ×P t , used to reflect the actual danger level and resource requirements of fire area j, A j is the fire area, P t is the diffusion probability, α is the adjustment coefficient, ρ is the penalty factor, C i is the cost coefficient of unit resource i, X i is the amount of resource i per unit, R total is the total resource limit, m is the type or quantity of available resources, and n is the number of fire areas.

10. The fire source positioning system for UAV fire cannon based on infrared thermal imaging according to claim 1, characterized in that: Use GIS visualization tools or Python visualization libraries to overlay the diffusion heat map onto the actual geographic coordinates, and use arrows to indicate the diffusion direction to form the final diffusion trend and achieve data visualization.

Images