Inspection unmanned aerial vehicle fire emergency system based on deep learning

Abstract

The invention relates to the field of data analysis, in particular to an inspection unmanned aerial vehicle fire emergency system based on deep learning, and the system comprises the steps: collecting forest image data and forest environment data; determining a fire source position, identifying tree plant distribution characteristics of an area where the fire source position is located, and analyzing a spreading trend characterization value for a fire behavior; predicting a diffusion coverage area of the fire behavior according to the spreading trend characterization value in combination with the wind direction data, and determining spreading and spreading characterization parameters based on topographic features of the diffusion coverage area so as to determine a diffusion alarm level of the diffusion coverage area; determining terrain steep amplitudes corresponding to a plurality of feasible paths based on the predetermined distance of the diffusion coverage area so as to determine an available fire-fighting path; according to the system and the method, the unmanned aerial vehicle inspection efficiency can be improved, the forest fire behavior can be fed back in real time, the fire behavior diffusion trend can be predicted, and the pertinence and effectiveness of emergency response can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a patrol drone fire emergency system based on deep learning. Background Art

[0002] Traditional forest fire inspections mainly rely on manual patrols and watchtower monitoring. Manual patrols are inefficient, have limited coverage, and are difficult to carry out in complex terrain and severe weather conditions. Although watchtower monitoring can cover a certain range, there are blind spots in monitoring, and it is impossible to grasp the dynamics of forest fires in real time and comprehensively.

[0003] Drones have the advantages of strong mobility, high flexibility, and the ability to quickly reach designated areas, and can play an important role in forest firefighting. Equipped with high-definition cameras, infrared thermal imagers and other equipment, drones can conduct all-round, real-time monitoring of forests from the air, and can promptly detect fire hazards and fire sources, overcoming the limitations of traditional monitoring methods. At the same time, drones can also conduct reconnaissance at the fire scene, providing ground firefighters with detailed information on the fire scene, such as the size of the fire, the direction of spread, topography, etc., to help firefighters develop scientific and reasonable firefighting plans.

[0004] As an important technology in the field of artificial intelligence, deep learning has powerful feature extraction and pattern recognition capabilities. In forest fire inspections, deep learning can analyze and process image and video data collected by drones, automatically identify fire characteristics such as flames and smoke, and improve the accuracy and timeliness of fire monitoring. In addition, deep learning can also predict the development trend of fires and provide a scientific basis for firefighting decisions. For example, by learning from historical fire data and real-time monitoring data, a fire spread model can be established to predict the spread direction and speed of fires under different meteorological and terrain conditions, helping fire departments to prepare preventive and response measures in advance.

[0005] With the continuous development of drone technology and deep learning technology, the forest fire emergency system that combines the two can give full play to the monitoring advantages of drones and the intelligent analysis capabilities of deep learning to achieve early detection, accurate warning, scientific decision-making and efficient firefighting of forest fires.

[0006] Chinese patent application publication number: CN119396184A, discloses a forest fire prevention UAV automatic inspection system based on meteorological factors, including: a meteorological data acquisition device, an UAV airport and an edge computing device, a remote data analysis and flight management system and a UAV; the meteorological data acquisition device is used to collect meteorological factor data of different meteorological conditions, special holidays or different seasons in real time; the UAV airport and the edge computing device are used to receive the meteorological factor data and clean and process it to obtain the processed data, and control the UAV to cruise the inspection area; the remote data analysis and flight management system is used to analyze the processed data, judge the forest fire risk meteorological level in different situations, formulate different UAV cruise missions, and send the UAV cruise mission; the UAV automatically performs the cruise mission according to the control command; the system can automatically increase the frequency of UAV forest fire prevention inspections according to special nodes to meet actual inspection needs, making forest fire prevention more efficient and intelligent.

[0007] However, there are still the following problems in the prior art: In the process of using drones to patrol forest fires, patrol and warn only based on the area where the fire occurs in real time, without considering the impact of factors such as meteorological conditions, forest topography and tree distribution on the spread of fire. Furthermore, the patrol and warning of other areas affected by the spread of fire are ignored, making the inspection not comprehensive enough and inefficient. Summary of the invention

[0008] To this end, the present invention provides a patrol drone fire emergency system based on deep learning, which is used to overcome the problem in the prior art that, in the process of forest fire patrol using drones, patrol and alarm are only based on the area where the fire occurs in real time, without considering the impact of factors such as meteorological conditions, forest topography and tree distribution on the spread of the fire, and further, ignoring the patrol and alarm of other areas affected by the spread of the fire, resulting in incomplete inspection and low efficiency.

[0009] To achieve the above objectives, the present invention provides a patrol drone fire emergency system based on deep learning, which includes: A fire situation collection module, which includes an image collection unit for acquiring forest image data and an environment collection unit for acquiring forest environment data; A fire source analysis module, which is connected to the fire situation collection module, is used to determine the location of the fire source, identify the tree distribution characteristics in the area where the fire source is located, and analyze the fire spread trend characterization value; a diffusion analysis module connected to the fire source analysis module, for predicting the diffusion coverage area of ​​the fire according to the diffusion trend characterization value combined with wind direction data, and determining the diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area to determine the diffusion alarm level of the diffusion coverage area; A path matching module, which is connected to the fire situation collection module and the diffusion analysis module respectively, and is used to determine the terrain steepness corresponding to a number of feasible paths based on the predetermined distance of the diffusion coverage area, so as to determine the available fire fighting path; An alarm emergency module, connected to the diffusion analysis module, for issuing an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module; The tree distribution characteristics include the accumulation density of vegetation in the area and the average spacing between trees, and the terrain characteristics include the depth of valleys and the verticality of slopes.

[0010] Furthermore, the fire situation analysis module is used to determine the location of the fire source, including Using the forest image data collected by the image acquisition unit to determine smoke distribution and smoke concentration changes, so as to locate the direction of the fire source; It is used to extract the heat distribution along the direction of the fire source, and determine the distribution position corresponding to the highest heat as the fire source position.

[0011] Furthermore, the fire situation analysis module is used to analyze the fire spreading trend representation value, including: The ratio of the accumulation density of vegetation in the area to the accumulation density threshold is used as the first spreading trend feature; The ratio of the average spacing threshold to the average spacing between trees in the area is used as the second spreading trend feature; The first spreading trend feature and the second spreading trend feature are weightedly summed to obtain the spreading trend representation value.

[0012] Furthermore, the diffusion analysis module is used to predict the diffusion coverage area of ​​the fire according to the spreading trend characterization value combined with wind direction data, including: To obtain the wind direction in the area where the fire source is located; To determine the spreading trend characterization values ​​corresponding to several areas along the wind direction starting from the area where the fire source is located; If there is any area whose corresponding spreading trend characterization value is greater than or equal to the spreading trend characterization threshold, the area is predicted as the spreading coverage area of ​​the fire.

[0013] Furthermore, the diffusion analysis module is used to determine the spreading diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area, including: The ratio of the valley depth to the valley depth threshold is used as the first spreading and diffusion feature; The ratio of the slope verticality to the slope verticality threshold is used as the second spreading feature; The sum of the first spreading diffusion characteristic and the second spreading diffusion characteristic is used as the spreading diffusion characterization parameter.

[0014] Furthermore, the diffusion analysis module is used to determine the diffusion alarm level of the diffusion coverage area, including: Preset the corresponding relationship between the diffusion alarm level and the preset diffusion characterization parameter range; Determining a preset spread diffusion characterization parameter interval to which the spread diffusion characterization parameter belongs; Setting the diffusion alarm level corresponding to the preset spreading diffusion characterization parameter interval as the diffusion alarm level of the diffusion coverage area; Among them, the diffusion alarm level corresponds one-to-one to the preset spread characterization parameter range.

[0015] Furthermore, the path matching module is used to determine that the feasible path includes: To determine a number of paths, and determine the path that meets the road condition benchmark conditions as a feasible path; The road condition reference condition includes that the road width is greater than a path width threshold and the curve radius is greater than a curve radius threshold.

[0016] Furthermore, the path matching module is used to determine the available fire fighting paths, including: If there is any feasible path whose terrain steepness is less than the terrain steepness threshold, the feasible path is determined as the available fire fighting path.

[0017] Furthermore, the predetermined distance is determined according to the range of the fire-fighting equipment.

[0018] Furthermore, the alarm emergency module is used to send an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module, including: Pre-set the corresponding relationship between the alarm signal and the diffusion alarm level; The alarm signal corresponds to the diffusion alarm level one by one, and the alarm signal includes the sounding of an alarm.

[0019] Compared with the prior art, the present invention is provided with a fire situation collection module, which includes an image collection unit for acquiring forest image data and an environment collection unit for collecting forest environment data; a fire source analysis module, which is used to determine the location of the fire source, identify the tree distribution characteristics of the area where the fire source location is located, and analyze the spread trend characterization value for the fire situation; a diffusion analysis module, which is used to predict the diffusion coverage area of ​​the fire situation based on the spread trend characterization value combined with wind direction data, and determine the diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area to determine the diffusion alarm level of the diffusion coverage area; a path matching module, which is used to determine the terrain steepness corresponding to several feasible paths based on a predetermined distance of the diffusion coverage area to determine the available fire fighting path; an alarm emergency module, which is used to send an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module. The present invention can improve the efficiency of drone inspections, provide real-time feedback on forest fires, predict the fire spread trend, and improve the pertinence and effectiveness of emergency response.

[0020] In particular, the present invention sets up a fire source analysis module, which determines the location of the fire source and combines it with the tree characteristics of the corresponding area to analyze the spread trend of the fire. In actual situations, dense vegetation distribution means a large number of combustibles per unit area. When a fire occurs, a large amount of vegetation can continuously provide fuel for combustion, making it easier for the fire to spread and burn more violently. At the same time, since the distance between densely distributed vegetation is relatively close, heat is more easily transferred between vegetation through conduction, radiation and convection, so that the fire can spread to the surrounding area faster, expanding the impact range of the fire. Similarly, if the distance between trees in the area where the fire source is located is too close, it will make it easier for the flame to spread directly from one tree to another, expanding the impact range of the fire in a short time. At the same time, the tree distribution The effect of thermal radiation generated in areas that are too close is more significant. Each burning tree will radiate a large amount of heat to the surrounding trees, causing the temperature of the adjacent trees to rise rapidly, greatly increasing the possibility of reaching the ignition point. This chain reaction of thermal radiation will cause more trees to catch fire one after another, further promoting the spread of the fire. Therefore, the present invention determines the fire spread trend characterization value through the vegetation accumulation density in the area where the fire source is located and the average spacing between trees to characterize the degree of influence of the distribution of trees and vegetation in the area where the fire source is located on the spread of the fire, and provides data support for the subsequent prediction of the spread coverage area of ​​the fire. The present invention can improve the efficiency of drone inspections, provide real-time feedback on forest fires, predict the trend of fire spread, and improve the pertinence and effectiveness of emergency response.

[0021] In particular, the present invention sets up a diffusion analysis module, which predicts the fire diffusion coverage area based on the tree distribution characteristics combined with wind direction data, and the distribution of vegetation and trees that can act as combustibles to affect the severity of the fire in forest fires determines the material basis for the spread of the fire. At the same time, combined with the wind direction, a factor that can directly affect the direction of the fire, a comprehensive analysis is performed to accurately predict the area that the fire may affect, and then, the degree of fire spread is determined in combination with the terrain and topography of the corresponding area. Therefore, the present invention determines the diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area, so as to characterize the influence of the internal terrain structure of the area affected by the fire on the severity of the fire, and provide data support for the subsequent determination of the diffusion alarm level. The present invention can improve the efficiency of drone inspections, provide real-time feedback on forest fires, and predict the trend of fire spread, thereby improving the pertinence and effectiveness of emergency response.

[0022] In particular, the present invention sets up a path matching module, which, on the premise that the fire spread coverage area is predicted, considers the steepness of the terrain of the path at a certain distance from the corresponding area, and determines the available firefighting path that can allow firefighting equipment to pass smoothly and provide sufficient space for firefighting implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a functional module diagram of a deep learning-based patrol drone fire emergency system according to an embodiment of the invention; Figure 2 A logical step diagram for predicting the spread and coverage area of ​​a fire according to an embodiment of the invention; Figure 3 A logical decision diagram for determining feasible paths for embodiments of the invention; Figure 4 Logic decision diagram for determining available firefighting paths for an embodiment of the invention. DETAILED DESCRIPTION

[0024] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0025] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0026] It should be noted that in the description of the present invention, terms such as "upper", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0027] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0028] See also Figure 1 As shown, it is a functional module diagram of a patrol drone fire emergency system based on deep learning according to an embodiment of the present invention. The patrol drone fire emergency system based on deep learning according to an embodiment of the present invention includes: A fire situation collection module, which includes an image collection unit for acquiring forest image data and an environment collection unit for acquiring forest environment data; A fire source analysis module, which is connected to the fire situation collection module, is used to determine the location of the fire source, identify the tree distribution characteristics in the area where the fire source is located, and analyze the fire spread trend characterization value; a diffusion analysis module connected to the fire source analysis module, for predicting the diffusion coverage area of ​​the fire according to the diffusion trend characterization value combined with wind direction data, and determining the diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area to determine the diffusion alarm level of the diffusion coverage area; A path matching module, which is connected to the fire situation collection module and the diffusion analysis module respectively, and is used to determine the terrain steepness corresponding to a number of feasible paths based on the predetermined distance of the diffusion coverage area, so as to determine the available fire fighting path; An alarm emergency module, connected to the diffusion analysis module, for issuing an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module; The tree distribution characteristics include the accumulation density of vegetation in the area and the average spacing between trees, and the terrain characteristics include the depth of valleys and the verticality of slopes.

[0029] Specifically, there is no limitation on the specific structure of the image acquisition unit, which only needs to have the function of acquiring forest image data. In some possible implementations, a laser radar is mounted on a drone, which emits a laser beam to the forest and measures the time from the laser being emitted to the laser being reflected back to the receiver, calculates the distance between the laser beam and the target object, obtains a large number of three-dimensional spatial coordinate points, and forms point cloud data. The height and distribution of vegetation are analyzed based on the point cloud data, and then the vegetation accumulation density is calculated. For example, the vegetation coverage area can be obtained by calculating the projection area of ​​the vegetation points on the horizontal plane. For regular grid areas, the number of grids occupied by the projection of vegetation points in the grid can be counted and then multiplied by the area of ​​a single grid. For irregular areas, the polygon area calculation method can be used to connect the boundaries of the vegetation point projections into polygons and calculate their areas. Then, the ratio of the vegetation coverage area to the total area of ​​the region is used as the vegetation accumulation density. By identifying the position and shape of trees, the spacing between trees is determined, and a high-precision digital terrain model is constructed to obtain the depth of the valley and the shape information of the slope, so as to calculate the verticality of the slope; and based on the digital terrain model, the elevation difference and horizontal distance between adjacent points are calculated to obtain the slope value of the road area, and the slope value is determined as the steepness of the terrain. Of course, other methods can also be used to obtain forest image data, which will not be repeated here; The forest image data includes tree distribution characteristics, terrain characteristics, and terrain steepness.

[0030] Specifically, there is no limitation on the specific structure of the environment collection unit, which only needs to have the function of collecting forest environment data. In some possible implementations, a wind direction sensor is carried on the drone to obtain wind direction information at the location of the drone in real time; In some possible implementations, the drone itself is equipped with a high-precision anemometer, which measures the changes in the drone's flight speed in different directions, combines the drone's own flight attitude and speed data, and uses relevant algorithms to infer the wind direction, which will not be elaborated here.

[0031] Specifically, there is no limitation on the method of dividing the forest into areas, as long as it meets the requirement of efficient fire inspection by drones. For example, the forest can be divided into several square areas according to the area of ​​the forest, and the drone can be controlled to inspect the area where the fire source is located first, and then the adjacent areas of the area can be inspected in combination with the wind direction data. This will not be elaborated here.

[0032] Specifically, there is no limitation on the specific structures of the fire source analysis module, diffusion analysis module, path matching module and alarm emergency module, which themselves or each unit therein can be composed of logic components or a combination of logic components, and the logic components include a field programmable processor, a computer or a microprocessor in a computer.

[0033] Specifically, the fire situation analysis module is used to determine the location of the fire source, including Using the forest image data collected by the image acquisition unit to determine smoke distribution and smoke concentration changes, so as to locate the direction of the fire source; It is used to extract the heat distribution along the direction of the fire source, and determine the distribution position corresponding to the highest heat as the fire source position.

[0034] It is understandable that the distribution of smoke is that it diffuses from the fire source to the surrounding area under the action of wind, and generally speaking, the closer to the fire source, the higher the smoke concentration. As the distance increases, the smoke will gradually dilute and the concentration will decrease. Observe the changes in smoke concentration along the direction of smoke diffusion. At the same time, combined with the heat changes caused by the fire, the heat at the fire source is much higher than the heat of the surrounding environment, and the location of the fire source can be accurately determined. This will not be repeated.

[0035] In this embodiment, based on the deep learning algorithm, a large amount of forest fire smoke image data is learned to distinguish smoke from other interference data in the forest, such as shadows, fog, etc. The pre-trained deep learning model can automatically extract the characteristics of smoke to identify smoke; For example, in implementation, the Mask RCNN model can be used. Mask RCNN is a powerful target detection and segmentation model that can accurately identify smoke areas in images. Of course, other forms of models can also be used. Those skilled in the art can choose according to their needs, which will not be repeated here.

[0036] By using a thermal infrared camera mounted on a drone, thermal infrared images of the forest area are collected from different heights and angles, the heat distribution and changes in the thermal infrared images are analyzed, and the areas with the highest heat are quickly located. This will not be elaborated here.

[0037] Specifically, the fire situation analysis module is used to analyze the fire spread trend representation value, including: The ratio of the accumulation density of vegetation in the area to the accumulation density threshold is used as the first spreading trend feature; The ratio of the average spacing threshold to the average spacing between trees in the area is used as the second spreading trend feature; The first spreading trend feature and the second spreading trend feature are weightedly summed to obtain the spreading trend representation value.

[0038] In actual fire situations, the distance between trees has a greater impact on the spread of fire. In implementation, the impact of the distance between trees on the spread of fire is given priority. Therefore, a slightly higher weight is given to the second spread trend feature calculated based on the average distance. When performing weighted summation, the weight of the first spread trend feature is set to 0.45, and the weight of the second spread trend feature is set to 0.55. In this embodiment, the purpose of setting the vegetation accumulation density threshold and the average spacing threshold between trees is to characterize the situation that the vegetation and trees in each area are closely distributed, which is easy to aggravate the trend of fire spread. The vegetation accumulation density threshold is pre-set by pre-acquiring a number of historical data related to forest fire records, calling the vegetation accumulation density historical data, solving the vegetation accumulation density mean, and setting the vegetation accumulation density threshold as the product of the accumulation density mean and the offset coefficient. The offset coefficient is selected in the interval [1.05, 1.1]. It is understandable that when the distance between trees is short, the branches and leaves of the trees are close to each other, and the flames can easily spread from one tree to another, forming a crown fire. The crown fire burns quickly and intensely, is difficult to extinguish, and can cause serious damage to the forest. Therefore, in this embodiment, the average spacing threshold between trees is selected within the interval [3m, 4m].

[0039] Specifically, the present invention sets a fire source analysis module, which determines the location of the fire source and combines it with the tree characteristics of the corresponding area to analyze the spread trend of the fire. In actual situations, dense vegetation distribution means a large number of combustibles per unit area. When a fire occurs, a large amount of vegetation can continuously provide fuel for combustion, making it easier for the fire to spread and burn more violently. At the same time, since the distance between densely distributed vegetation is relatively close, heat is more easily transferred between vegetation through conduction, radiation and convection, so that the fire can spread to the surrounding area faster, expanding the impact range of the fire. For example, in an area with dense bushes, the heat generated by the flame can be quickly transferred to adjacent bushes, causing the bushes to burn one after another and the fire to spread rapidly. Similarly, if the distance between trees in the area where the fire source is located is too close, it will make it easier for the flame to spread directly from one tree to another, expanding the scope of the fire in a short period of time. For example, if the branches and leaves of adjacent trees are in contact with each other or are very close to each other, the flame can easily jump from the crown of one tree to another, forming a large area of ​​crown fire, quickly sweeping the entire area. In addition, if the distance between trees is too close to each other, it will hinder the normal circulation of air and form a relatively closed space environment. The heat and smoke generated by the burning of trees accumulate in a limited space. Once the air circulation conditions are improved due to the burning and collapse of trees, a large amount of fresh air will flow in, which will cause more intense combustion, and the fire will instantly intensify and spread faster. At the same time, the effect of thermal radiation generated in areas where trees are too close is more significant. Each burning tree will radiate a large amount of heat to the surrounding trees, causing the temperature of the adjacent trees to rise rapidly, greatly increasing the possibility of reaching the ignition point. This chain reaction of thermal radiation will cause more trees to catch fire one after another, further promoting the spread of the fire. Therefore, the present invention determines the fire spread trend characterization value through the vegetation accumulation density and the average spacing between trees in the area where the fire source is located, so as to characterize the degree of aggravation of the impact of the distribution of trees and vegetation in the area where the fire source is located on the fire spread, and provides data support for the subsequent prediction of the fire spread coverage area. The present invention can improve the efficiency of drone inspections, provide real-time feedback on forest fires, predict the fire spread trend, and improve the pertinence and effectiveness of emergency response.

[0040] Specifically, see Figure 2 As shown, it is a logical step diagram of predicting the diffusion coverage area of ​​a fire according to an embodiment of the present invention. The diffusion analysis module is used to predict the diffusion coverage area of ​​the fire according to the spreading trend characterization value combined with wind direction data, including: To obtain the wind direction in the area where the fire source is located; To determine the spreading trend characterization values ​​corresponding to several areas along the wind direction starting from the area where the fire source is located; If there is any area whose corresponding spreading trend characterization value is greater than or equal to the spreading trend characterization threshold, the area is predicted as the spreading coverage area of ​​the fire.

[0041] The threshold for characterizing the spreading trend is selected in the interval [1.23, 1.46].

[0042] Specifically, the diffusion analysis module is used to determine the spreading diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area, including: The ratio of the valley depth to the valley depth threshold is used as the first spreading and diffusion feature; The ratio of the slope verticality to the slope verticality threshold is used as the second spreading feature; The sum of the first spreading diffusion characteristic and the second spreading diffusion characteristic is used as the spreading diffusion characterization parameter.

[0043] It is understandable that valleys with a depth of more than ten meters are prone to form local microclimates, where air circulation is poor, heat and smoke are not easy to diffuse, and heat is easily accumulated, making it difficult for the high temperature generated by the fire to dissipate, thereby contributing to the spread of the fire. Therefore, in implementation, in order to characterize the situation where local microclimates are easily formed, the valley depth threshold is set to 20m. Of course, those skilled in the art can adjust the valley depth threshold, which will not be repeated here.

[0044] Moreover, on steeper slopes, for example, those with a slope of more than 25°, the fire spreads significantly faster, especially when there is wind. The flames are more likely to spread rapidly upward with the help of the slope, forming a "fire rush", which spreads much faster than in flat terrain areas. Therefore, in practice, the slope verticality threshold is set to 30°; The angle between the slope surface and the vertical horizontal plane is taken as the verticality of the slope.

[0045] Specifically, the present invention sets up a diffusion analysis module, which predicts the diffusion coverage area of ​​the fire according to the tree distribution characteristics combined with the wind direction data, and determines the material basis for the spread of the fire according to the distribution of vegetation and trees that can act as combustibles to affect the severity of the fire in forest fires. At the same time, combined with the wind direction, which can directly affect the direction of the fire, a comprehensive analysis is conducted to accurately predict the area that the fire may affect, and then, combined with the terrain conditions of the corresponding area, the degree of spread of the fire is determined, including: If there are similar terrain structures such as depressions in the area where the fire spreads, deeper structures of this type will form unique airflow channels. The changes in the internal airflow may affect the propagation path of the flying fire. The flying fire may be carried by the airflow to a farther place in the depression, triggering a new fire source, thereby expanding the spread of the fire. For slope structures such as slopes, the higher the verticality of the slope, that is, the steeper the slope, the faster the fire spreads upward. At the same time, based on the rising characteristics of hot air, it is also easier for the flame to spread upward. The vegetation covered on such structures may collapse more easily due to gravity, resulting in more complete combustion. As the fire spreads rapidly upward, the vegetation above may be subjected to strong heat radiation in a short period of time, accelerating combustion, which makes the fire in the steep slope area more intense and the fire spreads faster. Therefore, the present invention determines the spread characterization parameters based on the terrain characteristics of the diffusion coverage area, so as to characterize the impact of the internal terrain structure of the area affected by the fire on the severity of the fire, and provide data support for the subsequent determination of the diffusion alarm level. The present invention can improve the efficiency of drone inspections, provide real-time feedback on forest fires, predict the trend of fire spread, and improve the pertinence and effectiveness of emergency responses.

[0046] Specifically, the diffusion analysis module is used to determine the diffusion alarm level of the diffusion coverage area, including: Preset the corresponding relationship between the diffusion alarm level and the preset diffusion characterization parameter range; Determining a preset spread diffusion characterization parameter interval to which the spread diffusion characterization parameter belongs; Setting the diffusion alarm level corresponding to the preset spreading diffusion characterization parameter interval as the diffusion alarm level of the diffusion coverage area; Among them, the diffusion alarm level corresponds one-to-one to the preset spread characterization parameter range.

[0047] In this embodiment, the diffusion alarm level is determined in the following manner: Divide the spread characterization parameters into three preset intervals, corresponding to three spread alarm levels; If the spreading diffusion characterization parameter is within the first preset interval [2.24, 2.37), the spreading alarm level corresponding to the spreading coverage area is the first spreading alarm level; If the spreading diffusion characterization parameter is within the second preset interval [2.37, 2.53], the spreading alarm level corresponding to the spreading coverage area is the second spreading alarm level; If the spreading diffusion characterization parameter is within the third preset interval (2.53, +∞], the spreading alarm level corresponding to the spreading coverage area is the third spreading alarm level.

[0048] The purpose of setting the alarm level is to issue an early warning when the spreading and diffusion characterization parameter is high. The spreading and diffusion characterization parameter is calculated by the depth of the valley and the verticality of the slope. The standard normal value of the spreading and diffusion characterization parameter is 2. Therefore, the first preset interval, the second preset interval and the second preset interval are set based on the deviation degree of the spreading and diffusion characterization parameter relative to the standard normal value. Each interval is continuous and represents a gradual increase in the deviation degree, which represents the deviation situation; In implementation, the left interval endpoint of the first preset interval is between 1.1 times and 1.13 times the standard normal value, and the right interval endpoint is set to be between 1.15 times and 1.2 times the standard normal value. The left interval endpoint of the second preset interval is continuous with the right interval endpoint of the first preset interval, and the right interval endpoint of the second preset interval is between 1.25 times and 1.3 times the standard normal value; The left interval endpoint of the third preset interval is continuous with the right interval endpoint of the second preset interval. The right interval endpoint of the third preset interval is usually set to positive infinity to cover all situations where the propagation and diffusion characterization parameters have a large deviation.

[0049] Specifically, see Figure 3 As shown, it is a logical decision diagram for determining a feasible path in an embodiment of the present invention. The path matching module is used to determine that the feasible path includes: To determine a number of paths, and determine the path that meets the road condition benchmark conditions as a feasible path; The road condition reference condition includes that the road width is greater than a path width threshold and the curve radius is greater than a curve radius threshold.

[0050] Specifically, for path identification, the forest image data collected by the drone can be input into a pre-trained deep learning model. The deep learning model can then classify the image or point cloud data at the pixel level or point level to segment the road from the forest background, thereby achieving the purpose of path identification.

[0051] Specifically, the path width threshold can be determined according to the way the fire truck implements the rescue. For example, in an application environment where the fire truck is required to travel smoothly in both directions, when multiple fire trucks are required to operate at the same time or to pass other rescue vehicles to provide sufficient space, the path width of the main road is generally not less than 6m. At the same time, the path width of the branch road that satisfies the one-way travel of the fire truck and ensures that rescue personnel can perform related operations next to the vehicle and leave a certain space is generally not less than 4m. This will not be repeated; The threshold value of the turning radius can be determined according to the type of fire truck that implements the rescue. For example, the rescue is carried out by a city main combat fire truck. Since the turning radius of this type of fire truck is relatively small, when road conditions permit, a turning radius of 10-12m can generally meet the turning needs of this type of fire truck; the rescue is carried out by an off-road fire truck. This type of fire truck is mainly used for fire fighting and rescue in complex terrains such as mountains and forests. Its turning radius is usually larger, generally around 15m. This is because the conditions of forest roads are poor, there are many obstacles and irregular curves, and a larger turning radius is required to ensure safe turning. This will not be elaborated here.

[0052] Specifically, see Figure 4 As shown, it is a logical decision diagram for determining an available fire-fighting path according to an embodiment of the present invention. The path matching module is used to determine the available fire-fighting path, including: If there is any feasible path whose terrain steepness is less than the terrain steepness threshold, the feasible path is determined as the available fire fighting path.

[0053] It is understandable that fire trucks generally used in special environments such as mountainous areas have higher climbing capabilities and can cope with steeper slopes, but generally should not exceed 30°. If it exceeds 30°, the fire truck may lose stability and be unable to drive normally, and may even cause accidents such as rollover or sliding. The terrain steepness threshold is set to 28° and will not be elaborated.

[0054] The present invention sets a path matching module. On the premise that the prediction of the spread and coverage area of ​​the fire is completed, the steepness of the terrain of the path at a certain distance from the corresponding area is considered to determine the available fire-fighting path that can provide smooth and smooth passage for fire-fighting equipment and sufficient space for fire-fighting implementation.

[0055] Specifically, the predetermined distance is determined based on the range of the firefighting equipment, wherein the firefighting equipment includes a large or special fire truck.

[0056] It is understandable that for forest fires, firefighting should be carried out at a safe distance, and the fire should be prevented from causing damage to rescue equipment and rescue personnel. Therefore, in this embodiment, firefighting and rescue are carried out on a road at a predetermined distance from the diffusion coverage area based on the range of the firefighting equipment. For example, when the firefighting equipment used is equipped with long-range fire extinguishing equipment, the predetermined distance can be set at about 50m-90m, and the fire can be extinguished by utilizing the range of the equipment. This will not be repeated.

[0057] Specifically, the alarm emergency module is used to send an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module, including: Pre-set the corresponding relationship between the alarm signal and the diffusion alarm level; The alarm signal corresponds to the diffusion alarm level one by one, and the alarm signal includes the sounding of an alarm.

[0058] In this embodiment, the correspondence between the alarm signal and the diffusion alarm level is determined in the following manner: If the diffusion alarm level is the first diffusion alarm level, an alarm signal 1 is issued; If the diffusion alarm level is the second diffusion alarm level, an alarm signal 2 is issued; If the diffusion alarm level is the third diffusion alarm level, an alarm signal 3 is issued; The alarm signal 1 includes: the alarm sounds for 30 seconds every 10 minutes; The alarm signal 2 includes: the alarm sounds for 30 seconds every 5 minutes; The alarm signal 3 includes that the alarm sounds for 30 seconds every 2 minutes.

[0059] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A patrol drone fire emergency system based on deep learning, characterized in that: include: A fire situation collection module, which includes an image collection unit for acquiring forest image data and an environment collection unit for acquiring forest environment data; A fire source analysis module, which is connected to the fire situation collection module, is used to determine the location of the fire source, identify the tree distribution characteristics in the area where the fire source is located, and analyze the fire spread trend characterization value; a diffusion analysis module connected to the fire source analysis module, for predicting the diffusion coverage area of ​​the fire according to the diffusion trend characterization value combined with wind direction data, and determining the diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area to determine the diffusion alarm level of the diffusion coverage area; A path matching module, which is connected to the fire situation collection module and the diffusion analysis module respectively, and is used to determine the terrain steepness corresponding to a number of feasible paths based on the predetermined distance of the diffusion coverage area, so as to determine the available fire fighting path; An alarm emergency module, connected to the diffusion analysis module, for issuing an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module; The tree distribution characteristics include the accumulation density of vegetation in the area and the average spacing between trees, and the terrain characteristics include the depth of valleys and the verticality of slopes.

2. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The fire situation analysis module is used to determine the location of the fire source, including Using the forest image data collected by the image acquisition unit to determine smoke distribution and smoke concentration changes, so as to locate the direction of the fire source; It is used to extract the heat distribution along the direction of the fire source, and determine the distribution position corresponding to the highest heat as the fire source position.

3. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The fire situation analysis module is used to analyze the fire spreading trend representation value, including: The ratio of the accumulation density of vegetation in the area to the accumulation density threshold is used as the first spreading trend feature; The ratio of the average spacing threshold to the average spacing between trees in the area is used as the second spreading trend feature; The first spreading trend feature and the second spreading trend feature are weightedly summed to obtain the spreading trend representation value.

4. The deep learning-based inspection drone fire emergency system according to claim 2 is characterized in that: The diffusion analysis module is used to predict the diffusion coverage area of ​​the fire according to the spreading trend characterization value combined with wind direction data, include, To obtain the wind direction in the area where the fire source is located; To determine the spreading trend characterization values ​​corresponding to several areas along the wind direction starting from the area where the fire source is located; If there is any area whose corresponding spreading trend characterization value is greater than or equal to the spreading trend characterization threshold, the area is predicted as the spreading coverage area of ​​the fire.

5. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The diffusion analysis module is used to determine the spreading and diffusion characterization parameters based on the terrain characteristics of the diffusion coverage area, including: The ratio of the valley depth to the valley depth threshold is used as the first spreading and diffusion feature; The ratio of the slope verticality to the slope verticality threshold is used as the second spreading and diffusion feature; The sum of the first spreading diffusion characteristic and the second spreading diffusion characteristic is used as the spreading diffusion characterization parameter.

6. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The diffusion analysis module is used to determine the diffusion alarm level of the diffusion coverage area, including: Preset the corresponding relationship between the diffusion alarm level and the preset diffusion characterization parameter range; Determining a preset spread diffusion characterization parameter interval to which the spread diffusion characterization parameter belongs; Setting the diffusion alarm level corresponding to the preset spreading diffusion characterization parameter interval as the diffusion alarm level of the diffusion coverage area; Among them, the diffusion alarm level corresponds one-to-one to the preset spread characterization parameter range.

7. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The path matching module is used to determine that the feasible path includes: To determine a number of paths, and determine the path that meets the road condition benchmark conditions as a feasible path; The road condition reference condition includes that the road width is greater than a path width threshold and the curve radius is greater than a curve radius threshold.

8. The deep learning-based inspection drone fire emergency system according to claim 7 is characterized in that: The path matching module is used to determine the available fire fighting paths, including: If there is any feasible path whose terrain steepness is less than the terrain steepness threshold, the feasible path is determined as the available fire fighting path.

9. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The predetermined distance is determined according to the range of the fire-fighting equipment.

10. The deep learning-based inspection drone fire emergency system according to claim 1 is characterized in that: The alarm emergency module is used to send an alarm signal of a corresponding level according to the diffusion alarm level determined by the diffusion analysis module, including: Pre-set the corresponding relationship between the alarm signal and the diffusion alarm level; The alarm signal corresponds to the diffusion alarm level one by one, and the alarm signal includes the sounding of an alarm.

Images