A Fire Recognition and Early Warning Method Based on Weather Radar

Abstract

The present invention discloses a method for fire recognition and early warning based on weather radar, which relates to the technical field of fire recognition and early warning, and includes the following steps: Step S1, input the data in the volume scan mode of the weather radar into the system; Step S2, data quality control: use the method of fuzzy logic to filter out ground clutter echoes; Step S3, echo image processing: extract the smoke and dust echoes from all echoes; perform dilation and erosion processing on the smoke and dust echoes, and then perform connected component extraction; Step S4, preliminary image recognition; Step S5, in-depth recognition based on physical quantities; Step S6, final recognition. The present invention is reasonably designed and can realize a fire detection method with large-range detection, all-weather operation, and being not easily interfered by external factors such as temperature and light.

Description

Technical Field

[0001] The present invention relates to the technical field of fire recognition and early warning, and more specifically to the technical field of a fire recognition and early warning method based on weather radar. Background Art

[0002] Fire is a disaster with strong suddenness and great destructiveness, posing a serious threat to people's lives and property safety. However, the existing fire detection technologies and devices, including laser smoke detectors, temperature sensors, light sensors, etc., still have many drawbacks and limitations in practical applications.

[0003] The detection range of a laser smoke detector is small. For large-scale fire scenarios such as forest fires and large urban fires, it is often difficult to detect in time, resulting in a lag in early warning and missing the best fire extinguishing opportunity. In forest fires, the smoke spreads widely and is greatly affected by terrain and wind direction. The laser smoke detector can only cover a limited area, making it easy for the fire to spread into a disaster before being discovered. In large urban fires, with high-rise buildings standing in great numbers and narrow streets, the installation position of the smoke detector is restricted, there are detection blind spots, and it is difficult to comprehensively monitor the smoke situation in the initial stage of the fire.

[0004] The temperature sensor is affected by temperature fluctuations, and the accuracy of its readings is greatly reduced. In a cold environment, low temperature may cause the sensor sensitivity to decline and it is not sensitive to the temperature change in the initial stage of the fire; while in a high-temperature environment, the sensor may generate errors due to overheating and misjudge normal temperature changes as fire signals, resulting in false alarms. In addition, the temperature sensor usually needs the fire to develop to a certain stage and the temperature to rise significantly to trigger an alarm, and it is difficult to respond in time at the budding stage of the fire.

[0005] The light sensor is extremely sensitive to light changes and is prone to misjudgment with slight light fluctuations. Under strong light irradiation, such as direct sunlight in summer or large-scale lighting, the light sensor may misjudge the light source change as a flame flicker, resulting in false alarms; while in a dim night or dark environment, its ability to recognize weak flame signals is limited and there may be missed alarms. This instability makes its reliability insufficient under different lighting conditions and it is difficult to accurately and stably detect fires.

[0006] In view of the above technical bottlenecks of these sensors or detectors, the existing equipment is stretched in practical applications. Therefore, a fire detection method that can achieve large-range detection, all-weather operation, and is not easily affected by external factors such as temperature and light is needed. Summary of the Invention

[0007] The purpose of the present invention is: to solve the above technical problems, the present invention provides a fire recognition and early warning method based on weather radar.

[0008] The present invention specifically adopts the following technical solutions to achieve the above purpose:

[0009] The present invention provides a fire identification and early warning method based on weather radar, comprising the following steps:

[0010] Step S1, inputting data in the weather radar volume scanning mode into the system;

[0011] Step S2, data quality control: The weather radar volume scan data input in step S1 contains ground object echoes, which will interfere with the subsequent processing of smoke echoes. Fuzzy logic is used to filter out the ground object echoes;

[0012] Step S3, echo image processing: The data type in the weather radar volume scan mode is in the form of multi-elevation angle, multi-radial and multi-range libraries; first, the combined reflectivity is calculated based on the data in the weather radar volume scan mode, and then echo image processing is performed based on the combined reflectivity to eliminate clutter with poor connectivity and extract the smoke echo from all echoes; the smoke echo is subjected to expansion and corrosion processing, and then connected blocks are extracted;

[0013] Step S4, preliminary image recognition: After extracting connected blocks, when smoke rises from the fire source, a "smoke plume" is formed. The smoke plume is an upward turbulent flow caused by the buoyancy generated by the combustion, usually including the lower combustion area. As the smoke rises, the smoke plume will widen. This characteristic of the smoke plume is reflected in the echo image as a narrow, long smoke echo. Two feature quantities, aspect ratio and compactness, are proposed for each connected block.

[0014] Thresholds are set for aspect ratio and compactness respectively. When the aspect ratio and compactness of a connected block meet both thresholds, it is marked as a type of target.

[0015] If only the aspect ratio and compactness features of the connected blocks are used, many misjudgments will occur. Therefore, Hough line detection is used based on the image, and the line detection results are marked as second-class targets.

[0016] For each first-class target, determine its degree of intersection with the second-class target and set an intersection threshold. When the degree of intersection between the two exceeds the set intersection threshold, the target is preliminarily identified as a preliminary fire target.

[0017] Step S5: Depth recognition based on physical quantities:

[0018] The fire points have been initially located in the preliminary fire targets. Next, these fire points need to be further screened. The specific method is to calculate the change gradient of the echo area and echo top height of the preliminary fire targets, set the echo area and echo top height to correspond to the preset conditions, and eliminate the fire points that do not meet the preset thresholds.

[0019] Step S6, Final Identification: If the echo of a preliminary fire target simultaneously meets the preset conditions of the echo area and two characteristic quantities, then this preliminary fire target is considered a fire target.

[0020] In one embodiment, in step S2, the method of fuzzy logic is used to filter the ground object echo. The characteristic quantities of the method of fuzzy logic include the vertical change of the echo intensity, the local change of the echo intensity, the radial change of the echo intensity, and the velocity value after median filtering.

[0021] Specifically, fuzzy logic refers to imitating the uncertain concept judgment and reasoning thinking mode of the human brain. For a description system that is unknown or cannot be determined for the model, as well as a control object with strong nonlinearity and large hysteresis, fuzzy sets and fuzzy rules are applied for reasoning to express transitional boundaries or qualitative knowledge and experience, simulate the human brain mode, implement fuzzy comprehensive judgment, and reason to solve the rule-based fuzzy information problems that are difficult to handle by conventional methods. Fuzzy logic is good at expressing qualitative knowledge and experience with unclear boundaries. It uses the concept of membership function to distinguish fuzzy sets, process fuzzy relationships, simulate the human brain to implement rule-based reasoning, and solve various uncertain problems caused by the logical break of the "law of excluded middle".

[0022] Fuzzy Logic is a mathematical method for processing uncertain and fuzzy information. It extends the scope of classical binary logic (true / false, 1 / 0), allowing variables to take any value within the interval [0,1], thus being closer to the fuzzy concepts in human thinking (such as "a bit hot", "relatively fast").

[0023] In one embodiment, in step S3, the specific method of echo image processing is as follows:

[0024] Step S31, After obtaining the combined reflectivity image, first perform dilation and erosion processing on the combined reflectivity image. A convolutional neural network with a convolution kernel of 3×3 is used during the dilation and erosion processing. The reason for such processing is that the soot echo is formed when the radar electromagnetic wave scatters when encountering soot particles in the air, and the radar receives the backscattering signals of these scattered waves. Therefore, the soot echo often shows discontinuous characteristics. After the erosion and dilation processing, the morphology of the soot echo will be better highlighted.

[0025] Step S32, Extract connected components from the soot echo that has undergone the dilation and erosion processing in step S31, and eliminate the clutter with poor connectivity, thereby realizing echo partitioning.

[0026] Specifically, connected component extraction is a basic technique in image processing used to identify and label pixel regions (connected components) that are connected to each other in an image. It is most common in binary images and is widely used in fields such as object segmentation, morphological analysis, and OCR preprocessing. After steps S31 and S32, a large amount of clutter will be removed due to poor connectivity, which will facilitate the identification of soot echoes.

[0027] In one embodiment, in step S4, for each connected component, the aspect ratio calculation formula is as follows:

[0028] aspect_ratio = max(w, h) / min(w, h);

[0029] In the formula, h is the length of the bounding rectangle of the connected component, w is the width of the bounding rectangle of the connected component, max(w, h) is the larger value of the length and width, and min(w, h) is the smaller value of the length and width; the greater the aspect ratio is greater than 1, the more slender the shape of the connected component is, that is, it is closer to a linear or strip shape;

[0030] For each connected component, the compactness calculation formula is as follows:

[0031] compactness = (4 × π × area) / (perimeter^2);

[0032] In the formula, perimeter represents the contour perimeter of the connected component, area represents the area of the connected component, and π is the circumference ratio; the smaller the compactness, the more slender the shape of the connected component.

[0033] In one embodiment, in step S4, when the connected component is a type of target, the aspect ratio threshold and the compactness threshold are set to 3.47 and 0.26 respectively.

[0034] In one embodiment, in step S4, for each type of target, judge its intersection degree with the second type of target. The specific method is as follows: for each type of target (that is, each connected component), calculate the ratio of the length that each second type of target (that is, each straight line) passes through it to the diagonal length of the connected component. If the ratio exceeds 0.8, then the target is initially identified as a fire target.

[0035] In one embodiment, in step S5, the specific calculation method of the echo area is as follows:

[0036] For each connected block, the number of grid points with a combined reflectivity greater than 5 dBz in consecutive moments is counted, and the result is taken as the echo area in the connected block; if there is an echo area that maintains an increasing trend for five consecutive moments and the increment exceeds 5000, then the connected block is considered to meet the echo area gradient change condition.

[0037] In one embodiment, in step S5, the specific calculation method of the echo top height is as follows:

[0038] For each connected block, the echo top height at each moment is calculated, and then the difference between the two moments is taken to obtain the ET difference. During the development of the fire, the combustion particles are affected by the thermal uplift in the fire, and their motion trajectory presents an irregular up and down tumbling state. This phenomenon is reflected in the smoke echo as follows: the ET difference in meteorological echoes usually has a relatively stable vector displacement feature, because the movement of meteorological particles (such as raindrops, hail, etc.) is mainly affected by atmospheric circulation, and their motion trajectory is relatively regular; while the ET difference of the fire smoke echo shows a large variation and disorder in its vector displacement due to the complex movement of the combustion particles. This difference can be used as one of the important features to distinguish fire smoke echoes from meteorological echoes.

[0039] In one embodiment, the specific method for extracting the disorder feature of ET difference is as follows: the ET difference between two moments has been obtained before, that is, a window with a length of 2 is used to slide to extract the ET difference, and the two ET differences before and after the window sliding are used as A and B inputs, and the optical flow method is used to calculate the vector field of A and B; after obtaining the vector field, a 15×15 grid matrix is defined, and the vector field is gradually traversed from left to right and from top to bottom; for the vector field area selected by the grid matrix, the variance of the vector data in the vector field area is calculated separately ,like If it is greater than 50, the vector field region is judged to be an ET differential disorder region.

[0040] The beneficial effects of the present invention are as follows:

[0041] The present invention is rationally designed and can realize a fire detection method with wide range detection, all-weather detection, and is not easily disturbed by external factors such as temperature and light. The specific effects are as follows:

[0042] Large-scale detection: Leveraging the volume-scanning mode of weather radar, this solution can scan and monitor large areas, effectively avoiding the delays in fire detection often associated with a narrow detection range. Whether in vast forests or large urban areas, smoke echoes from fires can be captured promptly from a considerable distance, enabling early warning of large-scale fires. This effectively addresses the blind spots of traditional laser smoke detectors in large-scale fire monitoring, providing strong support for fire prevention and control over large areas.

[0043] In terms of all-weather monitoring: The fire detection method based on weather radar is not restricted by light conditions. Whether it is during the daytime with direct sunlight, at night when it is completely dark, or in a complex environment with dim light, it can operate stably, achieve continuous monitoring of fires, effectively make up for the deficiency of light-sensitive detectors in reliability under different light conditions, avoid false alarms or missed alarms caused by light changes, and ensure the stability and continuity of fire monitoring.

[0044] In terms of strong anti-interference ability: The method of using fuzzy logic to filter out ground clutter echoes, as well as image processing means such as dilation and erosion processing and connected component extraction, can effectively remove clutter interference, making fire detection not easily affected by external factors such as temperature and light. This overcomes the problem that temperature sensors have inaccurate readings due to temperature fluctuations, and the deficiency that light-sensitive detectors are easily interfered by light intensity fluctuations, improves the accuracy of fire detection, reduces the occurrence of false alarms and missed alarms, and provides a more reliable basis for fire early warning.

[0045] In terms of accurately identifying fires: By calculating and analyzing feature quantities such as the aspect ratio and compactness of connected components, combined with means such as Hough line detection, it is possible to preliminarily screen and determine suspected fire targets; further, in-depth identification is carried out based on physical quantities such as the change gradient of the echo area and the change of the echo top height, and finally the fire target is accurately determined. This effectively solves the problem that traditional fire detection equipment is difficult to respond in a timely manner at the budding stage of a fire, can accurately identify a fire in the initial stage of the fire, buys precious time for fire extinguishing operations, and reduces the losses caused by the fire. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as a limitation of the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0047] Figure 1 is a flowchart of a fire recognition and early warning method based on weather radar of the present invention.

[0048] Figure 2 is a schematic diagram of reflectivity before quality control.

[0049] Figure 3 is a schematic diagram of reflectivity after quality control.

[0050] Figure 4 is a schematic diagram of combined reflectivity.

[0051] Figure 5 is a schematic diagram of connected component extraction.

[0052] Figure 6 It is a schematic diagram of a type of target determination.

[0053] Figure 7 It is a schematic diagram of a second type of target determination.

[0054] Figure 8 It is a schematic diagram for initially identifying fire targets.

[0055] Figure 9 It is a schematic diagram of the echo area change curve.

[0056] Figure 10 It is a schematic diagram of ET difference.

[0057] Figure 11 It is a schematic diagram of the optical flow vector field of the ET difference of meteorological echoes.

[0058] Figure 12 It is a schematic diagram of the optical flow vector field of the ET difference of soot echoes.

[0059] Figure 13 It is a schematic diagram of the final identification of fire targets. Detailed implementation manners

[0060] To make the technical problems, technical solutions, and technical effects of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Usually, the components of the embodiments of the present invention described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0061] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0062] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In addition, the terms "first", "second", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0063] Embodiment 1

[0064] As Figure 1 shown, this embodiment provides a method for fire identification and early warning based on a weather radar, including the following steps:

[0065] Step S1: Input the data in the weather radar volume scan mode into the system;

[0066] Step S2: Data quality control: There are ground clutter echoes in the weather radar volume scan data input in Step S1, and the ground clutter echoes will interfere with the subsequent processing of smoke and dust echoes. Use the fuzzy logic method to filter out the ground clutter echoes; use the fuzzy logic method to filter the ground clutter echoes. The characteristic quantities of the fuzzy logic method include the vertical change of echo intensity, the local change of echo intensity, the radial change of echo intensity, and the velocity value after median filtering, as Figure 2 and Figure 3 shown.

[0067] Specifically, fuzzy logic refers to imitating the uncertain concept judgment and reasoning thinking mode of the human brain. For a description system with unknown or uncertain models, as well as a control object with strong nonlinearity and large lag, apply fuzzy sets and fuzzy rules for reasoning, express transitional boundaries or qualitative knowledge and experience, simulate the human brain mode, and implement fuzzy comprehensive judgment and reasoning to solve the rule-based fuzzy information problems that are difficult to handle by conventional methods. Fuzzy logic is good at expressing qualitative knowledge and experience with unclear boundaries. It uses the concept of membership function to distinguish fuzzy sets, process fuzzy relations, simulate the human brain to implement rule-based reasoning, and solve various uncertain problems caused by the logical break of the "law of excluded middle".

[0068] Fuzzy Logic is a mathematical method for processing uncertain and fuzzy information. It extends the scope of classical binary logic (true / false, 1 / 0), allowing variables to take any value within the interval [0,1], thus being closer to the fuzzy concepts in human thinking (such as "a bit hot", "relatively fast").

[0069] Step S3: Echo image processing: The data type in the weather radar volume scan mode is in the form of multiple elevation angles, multiple radials, and multiple range bins; first calculate the composite reflectivity based on the data in the weather radar volume scan mode, and then perform echo image processing based on the composite reflectivity, as Figure 4 shown, eliminate the clutter with poor connectivity, and extract the smoke and dust echoes from all echoes; perform dilation and erosion processing on the smoke and dust echoes, and then perform connected component extraction; the specific method of echo image processing is as follows:

[0070] Step S31: After obtaining the composite reflectivity image, first perform dilation and erosion processing on the composite reflectivity image. A convolutional neural network with a convolution kernel of 3×3 is used during the dilation and erosion processing; the reason for this processing is that the smoke and dust echoes are formed by the radar electromagnetic waves scattered when encountering the smoke and dust particles in the air, and the radar receives the backscattered signals of these scattered waves. Therefore, the smoke and dust echoes often show discontinuous characteristics; after the erosion and dilation processing, the morphology of the smoke and dust echoes will be better highlighted;

[0071] Step S32: Perform connected component extraction on the soot echo that has undergone dilation and erosion processing in step S31, and eliminate the clutter with poor connectivity, thereby achieving echo partitioning.

[0072] Specifically, connected component extraction is a basic technique in image processing used to identify and label the pixel regions (connected components) that are connected to each other in an image. It is most common in binary images and is widely used in fields such as object segmentation, morphological analysis, and OCR preprocessing. After steps S31 and S32, a large amount of clutter will be eliminated due to poor connectivity, which will facilitate the identification of the soot echo, as Figure 5 shown.

[0073] Step S4: Preliminary image recognition: After extracting the connected components, when the flue gas rises from the fire source, a "smoke plume" will be formed. The smoke plume is an upward turbulent flow formed by the buoyancy generated by combustion and usually includes the combustion area at the lower part; as the flue gas rises, the smoke plume will become wider; and this characteristic of the smoke plume is manifested as a narrow-band and long-strip soot echo in the echo image; for each connected component, two characteristic quantities, the aspect ratio and the compactness, are proposed;

[0074] For each connected component, the formula for calculating the aspect ratio is as follows:

[0075] aspect_ratio = max(w, h) / min(w, h);

[0076] In the formula, h is the length of the circumscribed rectangle of the connected component, w is the width of the circumscribed rectangle of the connected component, max(w, h) is the larger value of the length and the width, and min(w, h) is the smaller value of the length and the width; the larger the aspect ratio is greater than 1, the more slender the shape of the connected component is, that is, it is closer to a linear or strip shape;

[0077] For each connected component, the formula for calculating the compactness is as follows:

[0078] compactness = (4 × π × area) / (perimeter^2);

[0079] In the formula, perimeter represents the contour perimeter of the connected component, area represents the area of the connected component, and π is the pi; the smaller the compactness is, the more slender the shape of the connected component is.

[0080] Set thresholds for the aspect ratio and the compactness respectively. When the aspect ratio and the compactness of a certain connected component simultaneously meet these two thresholds, it is marked as a type of target; when the connected component is a type of target, the thresholds of the aspect ratio and the compactness are set to 3.47 and 0.26 respectively, as Figure 6 shown.

[0081] If only the aspect ratio and compactness features of the connected components are used, there will be many misjudgments. Therefore, based on the image, Hough line detection is used, and the line detection results are marked as secondary targets, as Figure 7 shown.

[0082] For each primary target, judge its intersection degree with the secondary target, set an intersection threshold. When the intersection degree between the two exceeds the set intersection threshold, then this target is preliminarily identified as a preliminary fire target. The specific method is as follows: for each primary target (i.e., each connected component), calculate the ratio of the length of each secondary target (i.e., each line) passing through it to the diagonal length of the connected component. If the ratio exceeds 0.8, then this target is preliminarily identified as a fire target, as Figure 8 shown.

[0083] Step S5: Identification based on physical quantity depth

[0084] Among the preliminary fire targets, the fire points have been preliminarily located. Next, these fire points need to be further screened. The specific method is to calculate the change gradient of the echo area and the change of the echo top height of the preliminary fire targets, set the corresponding preset conditions for the echo area and the echo top height, and eliminate the fire points that do not meet the preset thresholds;

[0085] The specific calculation method of the echo area is as follows:

[0086] The combined reflectivity refers to the maximum value of the reflectivity factor at all elevation angles at a specific position within the detection range of the radar. For each connected component, count the number of grid points with a combined reflectivity greater than 5 dbz within consecutive time instants, and the result is used as the echo area within this connected component. If there is a trend that the echo area continuously increases for 5 consecutive time instants and the increment exceeds 5000, then it is considered that this connected component meets the echo area gradient change condition, as Figure 9 shown.

[0087] The specific calculation method of the echo top height is as follows:

[0088] For each connected component, calculate the echo top height at each time instant, and then calculate the difference between two consecutive time instants to obtain the ET difference. During the fire development process, the combustion particles are affected by thermal uplift in the fire, and their movement trajectories show an irregular up-and-down tumbling state. The characteristics reflected on the smoke echo are as follows: the ET difference in the meteorological echo usually has a relatively stable vector displacement characteristic because the movement of meteorological particles (such as raindrops, hailstones, etc.) is mainly affected by the atmospheric circulation and their movement trajectories are relatively regular; while the ET difference of the fire smoke echo has a large change and disorder in its vector displacement due to the complex movement of the combustion particles; this difference can be used as one of the important characteristics to distinguish the fire smoke echo from the meteorological echo, as Figure 10 shown.

[0089] The specific method for extracting the ET differential disorder feature is as follows: The ET differentials at two pairwise moments have been obtained previously, that is, a window of length 2 is slid to extract the ET differentials, and the ET differentials at two adjacent moments extracted by the window are used as the reference image A and the target image B respectively; The optical flow method is used to calculate the vector fields of the reference image A and the target image B; After obtaining the vector fields, a 15×15 grid matrix is defined, and the vector fields are traversed step by step from left to right and from top to bottom; For the vector field regions selected by the grid matrix, the variance of the vector data within the vector field regions is calculated separately , if is greater than 50, then it is determined that the vector field region is an ET differential disorder region, as Figure 11 and Figure 12 .

[0090] Step S6, final identification: If the echo of a preliminary fire target simultaneously meets the preset conditions of the echo area and two characteristic quantities, then it is considered that the preliminary fire target is a fire target, as Figure 13 shown.

Claims

1. A fire recognition and early warning method based on weather radar, characterized in that, It includes the following steps: Step S1: Input the data in the weather radar volume scan mode into the system; Step S2: Data quality control: Use the method of fuzzy logic to filter out ground clutter echoes; Step S3: Echo image processing: First calculate the composite reflectivity based on the data in the weather radar volume scan mode, then perform echo image processing based on the composite reflectivity, eliminate the clutter with poor connectivity, and extract the smoke and dust echoes from all echoes; Perform dilation and erosion processing on the smoke and dust echoes, and then perform connected component extraction; Step S4: Preliminary image recognition: After extracting the connected components, two feature quantities, namely the aspect ratio and compactness, are proposed for the connected components; Thresholds are set for the aspect ratio and compactness respectively. When the aspect ratio and compactness of a certain connected component simultaneously meet these two thresholds, it is marked as a type of target; If only relying on the aspect ratio and compactness features of the connected components, there will be many misjudgments; Therefore, based on the image, Hough line detection is used, and the line detection results are marked as type-two targets; For each type-one target, judge its intersection degree with the type-two targets, set an intersection threshold, and when the intersection degree between the two exceeds the set intersection threshold, it is preliminarily determined that this target is a preliminary fire target; Step S5: Deep recognition based on physical quantities: Among the preliminary fire targets, calculate the change gradient of the echo area and the change of the echo top height of the preliminary fire targets, set the corresponding preset conditions for the echo area and the echo top height, and eliminate the fire points that do not meet the preset thresholds; Step S6: Final recognition: If the echo of a certain preliminary fire target simultaneously meets the preset conditions of the echo area and the two feature quantities, it is considered that this preliminary fire target is a fire target.

2. The method for fire recognition and early warning based on weather radar according to claim 1, wherein, In step S2, when using the method of fuzzy logic to filter the ground clutter echoes, the feature quantities of the method of fuzzy logic include the vertical change of the echo intensity, the local change of the echo intensity, the change of the echo intensity along the radial direction, and the velocity value after median filtering.

3. The method for fire recognition and early warning based on weather radar according to claim 1, characterized in that, In step S3, the specific method of echo image processing is as follows: Step S31: After obtaining the composite reflectivity image, first perform dilation and erosion processing on the composite reflectivity image. During the dilation and erosion processing, a convolutional neural network with a convolution kernel of 3×3 is used to highlight the morphology of the smoke and dust echoes; Step S32: Perform connected component extraction on the smoke and dust echoes after the dilation and erosion processing in step S31, and eliminate the clutter with poor connectivity, so as to realize echo partitioning.

4. The method for fire recognition and early warning based on weather radar according to claim 3, characterized in that In step S4, for each connected component, the aspect ratio calculation formula is as follows: aspect_ratio = max(w, h) / min(w, h); In the formula, h is the length of the circumscribed rectangle of the connected component, w is the width of the circumscribed rectangle of the connected component, max(w,h) is the larger value of the length and width, and min(w,h) is the smaller value of the length and width; The larger the aspect ratio, the more slender the shape of the connected component; For each connected component, the compactness calculation formula is as follows: compactness = (4 × π × area) / (perimeter^2); In the formula, perimeter represents the contour perimeter of the connected component, area represents the area of the connected component, and π is the ratio of the circumference of a circle to its diameter; the smaller the compactness, the more slender the shape of the connected component.

5. The method for fire recognition and early warning based on weather radar according to claim 4, characterized in that, In step S4, when the connected component is a type-1 target, the aspect ratio threshold and the compactness threshold are set to 3.47 and 0.26 respectively.

6. The method for fire recognition and early warning based on weather radar according to claim 5, characterized in that, In step S4, for each type-1 target, judge its intersection degree with the type-2 target. The specific method is as follows: for each type-1 target, calculate the ratio of the length that each type-2 target passes through it to the diagonal length of the connected component. If the ratio exceeds 0.8, then this target is preliminarily identified as a fire target.

7. The method for fire recognition and early warning based on weather radar according to claim 6, characterized in that In step S5, the specific calculation method of the echo area is as follows: For each connected component, count the number of grid points with a combined reflectivity greater than 5 dbz within consecutive time instants, and the result is used as the echo area within this connected component; if there is an echo area that continuously shows an increasing trend for 5 consecutive time instants and the increment exceeds 5000, then it is considered that this connected component meets the echo area gradient change condition.

8. The method for fire recognition and early warning based on weather radar according to claim 6, characterized in that In step S5, the specific calculation method of the echo top height is as follows: For each connected component, calculate the echo top height at each time instant, and then calculate the difference between two adjacent time instants to obtain the ET difference; during the fire development process, the combustion particles are affected by thermal uplift in the fire, and their movement trajectories show an irregular up-and-down tumbling state. The characteristics reflected on the smoke echo are as follows: the ET difference in the meteorological echo usually has a relatively stable vector displacement characteristic because the movement of meteorological particles is mainly affected by the atmospheric circulation and their movement trajectories are relatively regular; while the ET difference of the fire smoke echo shows a large change and disorder in its vector displacement due to the complex movement of the combustion particles; this difference can be used as one of the important characteristics to distinguish the fire smoke echo from the meteorological echo.

9. A fire recognition and early warning method based on weather radar according to claim 8, characterized in that, The specific method for extracting the disorder feature of the ET difference is as follows: the ET differences between two adjacent time instants have been obtained before, that is, a window with a length of 2 is slid to extract the ET difference, and the ET differences of two adjacent time instants extracted by the window are used as the reference image A and the target image B respectively; use the optical flow method to calculate the vector fields of the reference image A and the target image B; After obtaining the vector field, define a 15×15 grid matrix and traverse the vector field step by step from left to right and from top to bottom; For the vector field region selected by the grid matrix, calculate the variance of the vector data within the vector field region separately , if is greater than 50, then determine that the vector field region is an ET differential disorder region.

Images