A method for aerial forest fire detection

Abstract

This application provides an aerial forest fire detection method, comprising: acquiring aerial image data of a forest fire area; employing dark channel feature analysis technology for fog interference; combining fog density estimation and transmittance map calculation to extract atmospheric light values ​​and perform color correction on the image to obtain a defogging initial processed image; if the clarity assessment result of the fire edge meets preset conditions, then using texture density distribution characteristic analysis technology to extract features of regional boundary noise interference and local brightness fluctuation amplitude; combining brightness distribution uniformity and regional contrast difference values ​​to determine the preliminary distribution characteristics of the fire area.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an aerial method for detecting forest fires. Background Technology

[0002] In modern forestry management and disaster prevention, aerial photography technology is crucial due to its ability to rapidly cover large areas and acquire real-time image data. Especially in forest fire monitoring and response, aerial imagery provides irreplaceable support for timely fire detection and disaster assessment. However, despite its widespread application, its performance in real-world scenarios is often constrained by various complex factors, affecting image quality and the accuracy of subsequent analysis. Therefore, targeted technological breakthroughs are urgently needed to enhance its practical value in fire prevention and control.

[0003] Currently, although numerous image processing and target detection methods have been applied to aerial image analysis, these methods often struggle to adapt to the unique environmental challenges of forest fire scenarios. Many existing solutions neglect the image quality degradation caused by atmospheric conditions and complex backgrounds from an aerial perspective, particularly in situations with dense smoke or insufficient light, where image details are severely lost, making it difficult to meet the demands for accurate fire identification. This limitation often prevents existing technologies from providing reliable visual information support in real-world fire scenarios.

[0004] More importantly, the core technical challenges of aerial imagery in fire scenarios lie in the degradation of image quality and the difficulty in extracting detailed features. Aerial images are often blurred due to fog interference from high altitudes, obscuring the outlines and textures of the fire area. On top of this, interference factors in the complex forest background further exacerbate the difficulty of identifying key targets such as flames. The low contrast caused by fog makes it difficult to capture subtle features of flames, while the mixing of background elements such as trees and shadows blurs the boundaries of target areas. For example, in aerial monitoring of a forest fire, images may be obscured by smoke, making it difficult to clearly show the location of the fire source or even determine the direction of fire spread, directly affecting the timeliness and accuracy of rescue decisions.

[0005] Therefore, effectively eliminating fog interference and mitigating the impact of complex backgrounds in aerial images to accurately extract detailed features of fire areas has become a key issue in improving forest fire monitoring and response capabilities. Solving this problem not only affects the quality of the images themselves but also directly impacts the efficiency and success rate of fire early warning and rescue operations. Summary of the Invention

[0006] This invention provides an aerial method for detecting forest fires, mainly comprising:

[0007] By collecting aerial image data of forest fire areas, dark channel feature analysis was used to address fog interference. Combined with fog density estimation and transmittance map calculation, atmospheric light values ​​were extracted, and the images underwent color correction to obtain a defogging initial image. Based on the defogging initial image, pixel grayscale distribution range data was obtained. Histogram statistical analysis was used to optimize grayscale mapping accuracy, and local brightness fluctuation adjustment techniques were combined to process contrast differences within regions block by block, determining an intermediate image with higher brightness distribution uniformity. For the local detail preservation and grayscale transition smoothness in the intermediate image, detail texture clarity information was extracted. Combined with edge gradient intensity analysis, if the texture blur judgment value exceeded a preset threshold, the pixel gradient distribution range was optimized through boundary noise interference filtering and color channel balance adjustment to obtain a smoothed refined image. From the refined image, based on local edges... By enhancing the applicability of edge continuity and detail, the initial outline of the potential fire area is segmented. Boundary gradient detection technology is used to analyze the pixel gradient distribution range of the fire edge. Combined with local grayscale transition smoothness data, edge continuity is detected to determine the degree of closure of the fire outline, thus obtaining the clarity assessment result of the fire edge. If the clarity assessment result of the fire edge meets the preset conditions, texture density distribution characteristic analysis technology is used to extract features for the noise interference degree of the region boundary and the amplitude of local brightness fluctuation. Combined with the uniformity of brightness distribution and the contrast difference value within the region, the initial distribution characteristics of the fire area are determined. Based on the initial distribution characteristics of the fire area, spatial location mapping data from aerial images is fused. A joint processing of grayscale value mapping accuracy analysis and detail texture clarity is used to track local detail changes of adjacent points within the fire area. Combined with local edge continuity and boundary texture density characteristics, the potential direction of fire spread is determined, resulting in the final fire monitoring distribution map.

[0008] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0009] This invention discloses a forest fire monitoring method based on aerial image processing. Addressing the interconnected operational problems of fog interference in aerial images of forest fire areas, such as uneven pixel grayscale distribution, blurred details and textures, difficulty in clearly segmenting fire edges, and inaccurate tracking of fire spread direction, the method collects aerial image data and uses dark channel feature analysis combined with fog density estimation and transmittance map calculation for defogging to obtain an initial image. Subsequently, histogram statistics are optimized based on pixel grayscale distribution intervals, and local brightness fluctuations are adjusted to generate an intermediate processed image. Further, detail and texture clarity information is extracted, and pixel gradient distribution is optimized through boundary noise filtering and color channel balancing to obtain a refined image. Then, the outline of potential fire areas is segmented, and edge continuity is analyzed to assess clarity. If conditions are met, texture density distribution features are extracted and fused with spatial location mapping data to track the fire spread direction, ultimately obtaining a fire monitoring distribution map. This method effectively solves the image processing challenges under fog interference, improves the accuracy of fire area identification and fire tracking, and achieves efficient forest fire monitoring. Attached Figure Description

[0010] Figure 1 This is a flowchart of an aerial forest fire detection method according to the present invention.

[0011] Figure 2 This is a schematic diagram of an aerial forest fire detection method according to the present invention.

[0012] Figure 3 This is another schematic diagram of an aerial forest fire detection method according to the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0014] like Figure 1-3 This embodiment of a method for aerial forest fire detection may specifically include:

[0015] S101. By collecting aerial image data of the forest fire area, dark channel feature analysis technology is used to target the fog interference part. Combined with fog density estimation and transmittance map calculation, atmospheric light value is extracted and the image is color corrected to obtain the initial processed image after defogging.

[0016] Based on the aforementioned business content and extracted relevant attributes, the following business solution is generated. Focusing on the image processing and analysis objectives for forest fire areas, and combining attributes such as forest fire, aerial imagery, fog interference, dark channel features, fog density, transmittance map, atmospheric light value, color correction, defogging, preliminary image, image acquisition, and regional analysis, the following technical process steps are generated: Images of the forest fire area are acquired using aerial photography equipment to obtain raw image data containing fog interference, completing preliminary data storage and obtaining an initial image set. For the fog interference portion in the initial image set, dark channel feature analysis technology is used to process image pixels region by region to determine the salient areas covered by fog. Based on the dark channel feature results of the salient areas, combined with fog density estimation methods, the fog concentration distribution of each region is calculated to obtain a fog density distribution map. Using the fog density distribution map, the corresponding transmittance map is calculated, and atmospheric light values ​​are extracted from the image to obtain key parameters for subsequent correction. Using the extracted atmospheric light values ​​and transmittance map, color correction processing is performed on the initial image set to complete the defogging operation, generating a preliminary defogging image. If localized blurry areas still exist in the initial image, the pixel differences before and after dehazing are compared to determine if residual interference exists, and an optimized image with improved clarity is obtained. Based on the optimized image, a detailed regional analysis of the forest fire area is conducted to delineate the fire range and affected zones, determining the final fire area distribution map.

[0017] For example, by collecting aerial image data of the forest fire area, the system first uses a high-definition camera mounted on a drone to acquire RGB images with a resolution of 1920x1080. The image data is stored with a depth of 8 bits per pixel, covering an area of ​​approximately 5 square kilometers. For fog interference, the system employs dark channel feature analysis technology. Specifically, it calculates the minimum value of each pixel's RGB three channels to form a dark channel image. For example, if a pixel's RGB value is (120, 150, 180), its dark channel value is 120. Then, a minimum value filter is applied to the dark channel image using a 15x15 window to obtain a smoothed dark channel image, which is used for subsequent fog density estimation. Next, the system estimates the fog density using the dark channel values, assuming that a lower dark channel value indicates denser fog. The calculation formula is density value = 1 - dark channel value / 255. For example, when the dark channel value is 51, the density value is approximately 0.8, and a fog density distribution map is constructed based on this. Furthermore, the system calculates the transmittance map using the formula Transmittance = 1 - ω * Density value, where ω is an adjustment parameter set to 0.95. A point in the transmittance map with a value of 0.24 indicates low light transmittance in that area. Next, the system extracts atmospheric light values. By selecting the top 0.1% of the brightest pixels in the dark channel image, its RGB average value is calculated, for example, resulting in atmospheric light values ​​of (220, 230, 240), which serve as a global atmospheric light reference. Finally, the system performs color correction on the image based on the dehazing model I(x)=J(x)*t(x)+A*(1-t(x)), where I(x) is the original image, J(x) is the dehazed image, t(x) is the transmittance, and A is the atmospheric light value. By inversely calculating J(x)=(I(x)-A*(1-t(x))) / t(x), each pixel is corrected. For example, if a pixel's original RGB value is (150, 160, 170), its transmittance is 0.24, and its atmospheric light value is (220, 230, 240), the calculated dehazed RGB value is approximately (83, 87, 91). This generates the initial processed image after dehazing, providing clear visual data for subsequent fire area analysis. The above process is automated through algorithms, ensuring high efficiency and accuracy in image processing.

[0018] S102. Based on the initial processed image after dehazing, obtain pixel grayscale distribution range data, use histogram statistical analysis method to optimize grayscale value mapping accuracy, combine local brightness fluctuation amplitude adjustment technology, process the contrast difference values ​​in the region block by block, and determine the intermediate processed image with higher brightness distribution uniformity.

[0019] Data extraction is performed on the preliminary image after dehazing to obtain the pixel grayscale distribution range information, completing the collection and organization of basic data and obtaining preliminary grayscale distribution results. Based on the preliminary grayscale distribution results, histogram analysis is used to statistically analyze the grayscale values, classify the data features within the distribution range, and determine the accuracy range of grayscale mapping. For the accuracy range of grayscale mapping, local brightness fluctuation data is obtained, and the brightness changes within the region are analyzed to obtain information on local brightness adjustment requirements. Based on the local brightness adjustment requirements and combined with fluctuation adjustment technology, the brightness values ​​within the image region are corrected block by block. If the local brightness fluctuation exceeds a preset threshold, the region is smoothed to determine the corrected brightness distribution data. Based on the corrected brightness distribution data, a secondary analysis is performed on the regional contrast differences. If the contrast difference value exceeds a preset range, the grayscale mapping of the region is fine-tuned to obtain a balanced brightness distribution image. With the balanced brightness distribution image obtained, block-by-block processing technology is used to optimize the overall brightness distribution of the image. If the brightness distribution in a certain region is still uneven, the pixel grayscale in that region is locally enhanced to obtain the final intermediate processed image. By combining the results of histogram analysis with the final intermediate processed image, the consistency of the grayscale distribution of the overall image is checked to determine that the uniformity of the image brightness distribution meets the predetermined standard.

[0020] For example, for the initial processed image after dehazing, the pixel grayscale distribution range data is first obtained through grayscale value statistics. Assuming the image is an 8-bit grayscale image with a grayscale value range of 0 to 255, the histogram statistical method is used to calculate the number of pixels for each grayscale value. It is found that the grayscale values ​​are concentrated in the 50 to 150 range, accounting for about 80%, while the 0 to 50 and 150 to 255 ranges have fewer pixels, accounting for only 20%, indicating uneven grayscale distribution. Next, the grayscale value mapping accuracy is optimized. The original grayscale values ​​are remapped using a histogram equalization algorithm. The cumulative distribution function (CDF) is calculated, mapping the grayscale value 50 to a new value of 30, and the grayscale value 150 to a new value of 220, expanding the distribution range to 0 to 255, improving the overall contrast. After equalization, the uniformity of grayscale value distribution is improved by about 35%, and the standard deviation is reduced from the original value of 40.5 to 26.3 through standard deviation analysis. Subsequently, combining local brightness fluctuation adjustment techniques, a 5x5 pixel window is used to calculate the local mean and variance. If the mean of a certain area is 100 and the variance is 15, then a linear transformation is used to adjust the variance to the target value of 10, reducing the brightness fluctuation amplitude by approximately 33% and ensuring a smooth transition in local brightness. Further, the contrast difference values ​​within each area are processed block by block. The image is divided into 16x16 pixel blocks, and the contrast of each block is calculated. If the contrast of a block is 0.3, lower than the target value of 0.5, then Laplacian sharpening is used to enhance the contrast to 0.48. After block-by-block processing, the overall contrast standard deviation decreases from 0.2 to 0.15. Finally, combining the above processing, an intermediate processed image with higher brightness distribution uniformity is generated. Through global brightness mean analysis, the value is adjusted from the original value of 110 to the target value of 128, improving the uniformity index by approximately 25%, laying the foundation for subsequent image processing. The above method is automatically implemented through an algorithm, ensuring rigorous logic and data correlation between steps to form a complete processing chain.

[0021] S103. Based on the degree of local detail preservation and grayscale transition smoothness in the intermediate processed image, extract the detail texture clarity information, combine it with the edge gradient intensity value analysis, and if the texture blur judgment value is higher than the preset threshold, then through the boundary noise interference filtering method and color channel balance adjustment, optimize the pixel gradient distribution range to obtain a smoothed fine-processed image.

[0022] For the original image, local detail information is acquired by segmenting the image region into blocks and extracting texture clarity data for each region to obtain preliminary texture feature distribution results. Based on the texture feature distribution results, gray-level transition regions are analyzed, and the smoothness of gray-level changes is calculated using a layered scanning method to determine the continuity index of gray-level transitions. Using the continuity index of gray-level transitions and edge gradient data, intensity analysis values ​​are calculated. If the intensity analysis value is lower than a preset threshold, edge regions are enhanced to obtain edge-enhanced image data. For the edge-enhanced image data, blur detection is performed. If the blur detection value is higher than a preset threshold, boundary noise filtering is initiated to reduce noise interference and obtain denoised image content. Based on the denoised image content, the distribution of color channels is analyzed, and equalization adjustment technology is used to optimize the balance of color channels and determine the adjusted color data. Using the adjusted color data, the distribution range of pixel gradients is optimized, and smoothing processing technology is applied to generate the final refined image result.

[0023] For example, regarding the analysis of local detail preservation and grayscale transition smoothness in intermediate processed images, the following steps are taken: First, the detail texture clarity information of local image regions is calculated, and the difference in Gaussian (DoG) algorithm is used to extract detail features. The standard deviation of the Gaussian kernel is set to 1.5 and 2.5. The difference images at two scales are calculated to obtain detail response values. If the response value is lower than 0.02, it is determined to be a blurred region. Next, combined with edge gradient intensity analysis, the Sobel operator is used to calculate the horizontal and vertical gradients to obtain gradient magnitudes. If the average gradient magnitude is less than 10, the degree of texture blurring is further confirmed, and a blur judgment threshold of 0.015 is set. If the response value is higher than this threshold, the optimization process begins. Subsequently, a boundary noise interference filtering method is used, employing a bilateral filtering algorithm with a spatial domain standard deviation of 3.0 and a value domain standard deviation of 30. Boundary noise is filtered out while preserving edge details. The noise variance reduction before and after filtering is calculated to be 15.6, verifying the filtering effect. Next, color channel balance adjustment is performed. The histogram distribution of the RGB three channels is analyzed. If the mean deviation of a certain channel exceeds 20, the channel value is adjusted to within ±10 of the mean through linear stretching to ensure color balance. Finally, the pixel gradient distribution range is optimized. Based on gradient histogram equalization, the gradient values ​​are remapped to the range of 0 to 255. The standard deviation of grayscale transition in the smoothed image is calculated to decrease from the initial 18.5 to 9.2, resulting in a refined image. Through the above process, a complete logical chain is formed from detail extraction to smoothing optimization. If the detail response value is inconsistent with the gradient magnitude analysis result, local contrast enhancement is introduced as an auxiliary judgment to ensure the accuracy of the judgment and enhance the robustness of image processing.

[0024] S104. From the finely processed image, based on the local edge continuity and the applicability of detail enhancement, the preliminary outline of the potential fire area is segmented. The boundary gradient detection technology is used to analyze the pixel gradient distribution range of the fire edge. Combined with the local grayscale transition smoothness data, edge continuity detection is performed to determine the degree of closure of the fire outline and obtain the clarity evaluation result of the fire edge.

[0025] Preliminary information about the fire area is obtained from the original image data. Image segmentation techniques are used to extract the contour range of the potential fire area, resulting in a first contour image. Based on the first contour image, the continuity of local edges is detected, and a preset gradient analysis method is used to calculate the boundary gradient change to determine the intensity of local edge continuity. From the local edge continuity intensity data, detailed features of the boundary pixel distribution are obtained, and combined with the smoothness of grayscale transitions, a comprehensive analysis is performed to determine the integrity of the contour closure. Based on the integrity result of the contour closure, a detail enhancement processing method is used to optimize the edge region, resulting in a second contour image. Relevant data on edge clarity are extracted from the second contour image. If the edge clarity intensity is lower than a preset threshold, the saliency of the edge is enhanced by local grayscale adjustment to determine the final edge contour. Based on the final edge contour, a clarity evaluation index is used for quantitative processing to obtain the clarity evaluation result of the fire area contour. Based on the clarity evaluation result, the contour closure and edge clarity data of the fire area are comprehensively compared. If the evaluation result meets the preset standard, the final fire area contour image is output.

[0026] For example, when processing a refined image to segment the initial outline of a potential fire area, the image is first processed using the Canny edge detection algorithm through local edge continuity analysis. A low threshold of 50 and a high threshold of 150 are set to extract edge pixels with significant grayscale changes, forming a preliminary fire area outline. The analysis results show that edge pixels account for approximately 3.2% of the total image pixels, indicating that the boundary of the potential fire area is initially revealed. Next, boundary gradient detection technology is used to calculate the pixel gradient distribution range of the fire edge. The Sobel operator is used to calculate the gradient in the horizontal and vertical directions of the image, resulting in gradient magnitudes ranging from 20 to 180. Pixels with gradient values ​​higher than 100 account for 1.5%, indicating strong grayscale changes in the fire edge area. Subsequently, edge continuity detection is performed by combining local grayscale transition smoothness data. By calculating the standard deviation of the grayscale difference between adjacent pixels and setting a threshold of 10, the detection results show that 85% of the edge pixels have a grayscale difference less than this threshold, indicating good edge continuity. To further determine the degree of closure of the fire outline, a outline tracking algorithm was used to calculate the distance between the start and end points of the outline, resulting in a closure score of 0.92 (out of 1.0), indicating that the outline is close to being closed. Finally, based on the above data, the sharpness of the fire edge was assessed. Combining gradient amplitude, continuity score, and closure, a weighted calculation formula (sharpness = 0.4 * mean gradient + 0.3 * continuity score + 0.3 * closure) was used, yielding a sharpness score of 0.85, indicating that the fire edge is relatively clear. These steps are automatically processed by the algorithm, with data logically progressing from edge extraction to sharpness assessment, progressively verifying the credibility of the fire area and providing a reliable basis for subsequent fire early warning.

[0027] S105. If the clarity assessment results of the fire edge meet the preset conditions, the texture density distribution characteristic analysis technology is used to extract features of regional boundary noise interference and local brightness fluctuation amplitude. Combined with the brightness distribution uniformity and the contrast difference value within the region, the preliminary distribution characteristics of the fire area are determined.

[0028] Step 1: Acquire raw image data of the fire scene using image acquisition equipment. Perform preliminary segmentation processing on fire edges and area boundaries. Preprocess the image using grayscale conversion technology to obtain a pre-segmented grayscale image. Step 2: From the pre-segmented grayscale image, remove irrelevant interference information from the image by using filtering techniques to address area boundaries and noise interference. Extract key data related to texture density to determine the clean image after processing. Step 3: Based on the clean image, analyze local brightness and fluctuation characteristics. Combined with the uniform distribution of brightness, calculate the brightness change trend of each area in the image to obtain a brightness fluctuation distribution map. Step 4: Using the brightness fluctuation distribution map, combined with area comparison and difference values, use the support vector machine algorithm to classify the image areas, determine the distribution of fire areas and non-fire areas, and obtain a classification result map. Step 5: Analyze the correlation between distribution characteristics and texture density in the classification result map. If the texture density of a certain area in the classification result map is higher than a preset threshold, mark it as a potential fire core area and determine the core area distribution map. Step Six: Based on the core area distribution map, combined with the fire edge and clear standards, refine the boundaries of the potential fire core area to obtain the final fire area distribution map and complete the precise location of the fire area.

[0029] For example, in the assessment of fire scene edge sharpness, the input fire scene image is first processed using image processing techniques for edge detection. The Canny algorithm is used with a low threshold of 50 and a high threshold of 150 to extract edge pixels and calculate the mean gradient intensity of these pixels. Assuming the result is 120, which meets the preset condition (mean gradient greater than 100), the analysis proceeds to the next step. Next, using texture density distribution analysis techniques, Gaussian filtering (standard deviation of 2.0) is applied to smooth the boundary region for noise interference. The noise interference is calculated to be 0.15 (below the preset threshold of 0.2). Simultaneously, features of local brightness fluctuations are extracted. By calculating the standard deviation of the brightness difference between adjacent pixels, the fluctuation amplitude is found to be 8.5 (less than the preset value of 10), indicating that the boundary region is relatively stable. Subsequently, considering the uniformity of brightness distribution, the entropy value of the brightness value within the region was calculated to be 6.8 (close to the ideal value of 7.0), indicating a relatively uniform distribution. Further analysis of the contrast difference within the region, using the contrast formula (maximum brightness minus minimum brightness divided by average brightness), yielded a difference value of 0.3 (less than the preset value of 0.5), indicating moderate contrast within the region. Finally, based on the above characteristics, a threshold-based segmentation algorithm was used to mark areas meeting the following criteria as preliminary fire distribution areas: noise interference level below 0.2, fluctuation amplitude less than 10, brightness entropy value greater than 6.5, and contrast difference value less than 0.5. This generated a distribution feature map, providing data support for subsequent fire range confirmation. The above process, through automatic calculation and analysis by the algorithm, ensured the objectivity and accuracy of feature extraction. Simultaneously, the logical relationships between the parameters were close; noise and brightness fluctuations affected boundary judgment, while uniformity and contrast difference further verified the effectiveness of the regions, forming a complete technical chain.

[0030] S106. Based on the preliminary distribution characteristics of the fire area, the spatial location mapping data in the aerial images are integrated, and the grayscale value mapping accuracy analysis and detail texture clarity are combined to track the local detail changes of adjacent points in the fire area. Combining the local edge continuity and boundary texture density characteristics, the potential direction of fire spread is determined, and the final fire monitoring distribution map is obtained.

[0031] Preliminary distribution data of the fire area was obtained through aerial imagery. Image segmentation techniques were used to separate the target area from the background area, resulting in an initial division of the fire area. Based on this initial division, spatial location information of the fire area was extracted. Combined with grayscale accuracy analysis, the brightness difference between adjacent points was calculated. If the brightness difference exceeded a preset threshold, it was identified as a region with significant local changes, thus identifying potential active fire points. For active fire points, the texture details of adjacent points were analyzed to obtain edge continuity information. If the edge continuity showed a linear expansion trend, it was identified as a priority path for fire spread, resulting in a preliminary distribution of spread paths. Based on the preliminary distribution of spread paths, boundary density changes were calculated. A convolutional neural network model was used to extract deep features from the texture details, obtaining the density distribution characteristics of the boundary area. Based on the density distribution characteristics, the potential direction of fire spread was tracked. If the density distribution characteristics showed an increasing trend in a certain direction, that direction was identified as the main trend of fire spread, thus identifying key monitoring areas for fire spread. Spatial location data of key monitoring areas was obtained. Combined with updated aerial imagery information, local changes and edge continuity were continuously tracked to determine the dynamic evolution trend of fire spread. By analyzing the dynamic evolution trend, a real-time distribution map of fire monitoring is generated, and spatial location and potential direction information are overlaid to obtain comprehensive monitoring results of the fire area.

[0032] For example, in the technical implementation of fire area monitoring, the preliminary distribution characteristics of the fire area are first obtained through aerial imagery. Assuming the image resolution is 1920x1080 pixels, the image is divided into 100x100 pixel grid cells using spatial location mapping data. The grayscale mean of each cell is calculated, with a grayscale value range of 0-255. Areas exceeding 180 are initially identified as high-risk fire areas. Next, grayscale value mapping accuracy analysis and detail texture clarity processing are combined. The Sobel operator is used to extract image edge features, and the edge intensity of each grid cell is calculated. A threshold of 50 is set. If the edge intensity of a cell is greater than 50, its texture details are considered significant and require further analysis. Subsequently, local detail changes at adjacent points within the fire area are tracked. Based on the grayscale difference between grid cells, the grayscale change rate of adjacent cells is calculated. If the change rate exceeds 20%, it is marked as a potentially active fire area. Simultaneously, combined with local edge continuity analysis, the Canny algorithm is used to detect edge continuity, with a minimum continuity length of 10 pixels. If the continuity is insufficient, it is considered an unstable boundary area. Subsequently, the boundary texture density characteristics were analyzed, and the number of texture feature points per square pixel was calculated. A density greater than 0.5 was identified as a potential fire spread boundary. The direction of fire spread was predicted using the Histogram of Oriented Gradients (HOG) algorithm, with a directional angle variation range of ±30 degrees considered the primary spread trend. Finally, based on the above analysis results, a fire monitoring distribution map was generated. Using heatmap visualization technology, high-risk fire areas were marked in red (RGB value 255,0,0), and potential spread areas were marked in orange (RGB value 255,165,0), forming an intuitive distribution map to provide a basis for subsequent decision-making. All the above steps were automatically processed by algorithms, with tight logical connections between data. Indicators such as grayscale values, edge intensity, and texture density cross-validated each other to ensure the accuracy of fire spread direction judgment. Simultaneously, the generated heatmap corresponded one-to-one with the previous analysis results, forming a complete technical loop.

[0033] The above are only some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and modifications can be made. Any improvements and modifications made based on the basic principles of the present invention should be considered to fall within the protection scope of the present invention.

Claims

1. A method for aerial forest fire detection, characterized in that, The method includes: By collecting aerial image data of forest fire areas, dark channel feature analysis was used to address fog interference. Combined with fog density estimation and transmittance map calculation, atmospheric light values ​​were extracted, and the images underwent color correction to obtain a defogging initial image. Based on the defogging initial image, pixel grayscale distribution range data was obtained. Histogram statistical analysis was used to optimize grayscale mapping accuracy, and local brightness fluctuation adjustment techniques were combined to process contrast differences within regions block by block, determining an intermediate image with higher brightness distribution uniformity. For the local detail preservation and grayscale transition smoothness in the intermediate image, detail texture clarity information was extracted. Combined with edge gradient intensity analysis, if the texture blur judgment value exceeded a preset threshold, the pixel gradient distribution range was optimized through boundary noise interference filtering and color channel balance adjustment to obtain a smoothed refined image. From the refined image, based on local edges... By enhancing the applicability of edge continuity and detail, the initial outline of the potential fire area is segmented. Boundary gradient detection technology is used to analyze the pixel gradient distribution range of the fire edge. Combined with local grayscale transition smoothness data, edge continuity is detected to determine the degree of closure of the fire outline, thus obtaining the clarity assessment result of the fire edge. If the clarity assessment result of the fire edge meets the preset conditions, texture density distribution characteristic analysis technology is used to extract features for the noise interference degree of the region boundary and the amplitude of local brightness fluctuation. Combined with the uniformity of brightness distribution and the contrast difference value within the region, the initial distribution characteristics of the fire area are determined. Based on the initial distribution characteristics of the fire area, spatial location mapping data from aerial images is fused. A joint processing of grayscale value mapping accuracy analysis and detail texture clarity is used to track local detail changes of adjacent points within the fire area. Combined with local edge continuity and boundary texture density characteristics, the potential direction of fire spread is determined, resulting in the final fire monitoring distribution map.

2. The aerial forest fire detection method according to claim 1, characterized in that, The process involves collecting aerial image data of forest fire areas, employing dark channel feature analysis technology to address fog interference, combining fog density estimation and transmittance map calculation, extracting atmospheric light values, and performing color correction on the images to obtain a defogging initial processed image, including: Based on the aforementioned business content and extracted relevant attributes, the following business solution is generated. Focusing on the image processing and analysis objectives of forest fire areas, and combining attributes such as forest fires, aerial images, fog interference, dark channel features, fog density, transmittance maps, atmospheric light values, color correction, defogging, preliminary images, image acquisition, and regional analysis, the following technical process steps are generated: Images of the forest fire area are acquired using aerial photography equipment to obtain raw image data containing fog interference; preliminary data storage is completed to obtain an initial image set. For the fog interference in the initial image set, dark channel feature analysis technology is used to process the image pixels region by region to determine the significant areas covered by fog. Based on the dark channel characteristics of significant areas, and combined with fog density estimation methods, the fog concentration distribution of each area is calculated to obtain a fog density distribution map. By using the fog density distribution map, the corresponding transmittance map is calculated, and the atmospheric light value in the image is extracted to obtain key parameters for subsequent correction. The extracted atmospheric light values ​​and transmittance maps are used to perform color correction on the initial image set to complete the dehazing operation and generate a preliminary dehazed image. If there are still local blurry areas in the initial image, the pixel differences before and after dehazing are compared to determine whether there is residual interference, and an optimized image with improved clarity is obtained. Based on the optimized images, a detailed regional analysis of the forest fire area was conducted to delineate the fire range and affected zones, and to determine the final fire area distribution map.

3. The aerial forest fire detection method according to claim 1, characterized in that, The process involves obtaining pixel grayscale distribution range data from the initial dehazed image, optimizing the grayscale value mapping accuracy using histogram statistical analysis, and combining this with local brightness fluctuation adjustment techniques. It also involves processing contrast differences within a region block by block to determine an intermediate processed image with higher brightness distribution uniformity. This includes: By extracting data from the preliminary image after dehazing, the distribution range information of pixel gray levels is obtained, the basic data is collected and organized, and the preliminary gray level distribution results are obtained. Based on the preliminary grayscale distribution results, histogram analysis was used to statistically analyze the grayscale values, and the data characteristics within the distribution range were classified to determine the accuracy range of grayscale mapping. Based on the accuracy range of grayscale mapping, local brightness fluctuation data is obtained, the brightness change within the region is analyzed, and the adjustment requirements for local brightness are obtained. By combining the local brightness adjustment requirements with fluctuation adjustment technology, the brightness values ​​in the image area are corrected block by block. If the fluctuation of local brightness exceeds the preset threshold, the area is smoothed to determine the corrected brightness distribution data. Based on the corrected brightness distribution data, a secondary analysis is performed on the regional contrast difference. If the contrast difference value exceeds the preset range, the grayscale mapping of the region is fine-tuned to obtain a balanced brightness distribution image. A balanced brightness distribution image is obtained, and the overall brightness distribution of the image is optimized by a block-by-block processing technique. If the brightness distribution of a certain area is still uneven, the pixel grayscale of that area is locally enhanced to obtain the final intermediate processed image. By combining the results of histogram analysis with the final intermediate processed image, the consistency of the grayscale distribution of the overall image is checked to determine that the uniformity of the image brightness distribution meets the predetermined standard.

4. The aerial forest fire detection method according to claim 1, characterized in that, The process involves extracting detail texture clarity information based on the degree of local detail preservation and grayscale transition smoothness in the intermediate processed image. Combined with edge gradient intensity analysis, if the texture blur determination value exceeds a preset threshold, the pixel gradient distribution range is optimized through boundary noise interference filtering and color channel balance adjustment to obtain a smoothed, finely processed image. This includes: For the original image, local detail information is obtained. By dividing the image region into blocks, the texture sharpness data of each region is extracted to obtain preliminary texture feature distribution results. Based on the texture feature distribution results, the gray-level transition region is analyzed, and the smoothness of gray-level changes is calculated using a layered scanning method to determine the continuity index of gray-level transition. By combining the continuity index of grayscale transition with edge gradient data, the intensity analysis value is calculated. If the intensity analysis value is lower than the preset threshold, the edge region is enhanced to obtain the image data with enhanced edges. For the image data after edge enhancement, the blur determination is detected. If the blur determination value is higher than the preset threshold, the boundary noise filtering method is activated to reduce noise interference and obtain the denoised image content. Based on the content of the denoised image, the distribution of color channels is analyzed, and equalization adjustment technology is used to optimize the balance of color channels and determine the adjusted color data. By adjusting the color data, optimizing the distribution range of pixel gradients, and applying smoothing techniques, the final refined image result is generated.

5. The aerial forest fire detection method according to claim 1, characterized in that, From the refined image, based on local edge continuity and the applicability of detail enhancement, a preliminary outline of the potential fire area is segmented. Boundary gradient detection technology is used to analyze the pixel gradient distribution range of the fire edge. Combined with local grayscale transition smoothness data, edge continuity detection is performed to determine the degree of closure of the fire outline, resulting in a sharpness assessment of the fire edge, including: By obtaining preliminary information about the fire area from the original image data, image segmentation technology is used to extract the contour range of the potential fire area to obtain the first contour image; Based on the first contour image, the continuity of local edges is detected, and the boundary gradient change is calculated using a preset gradient analysis method to determine the continuity intensity of local edges. From the continuous intensity data of local edges, detailed features of boundary pixel distribution are obtained, and combined with the smoothness of gray-level transitions for comprehensive analysis to determine the integrity of contour closure. To ensure the integrity of the contour closure, a detail enhancement processing method is used to optimize the edge region, resulting in a second contour image. Extract relevant data with clear edges from the second contour image. If the intensity of the clear edges is lower than a preset threshold, enhance the saliency of the edges through local grayscale adjustment to determine the final edge contour. Based on the final edge contour, the clarity assessment index is used for quantitative processing to obtain the clarity assessment result of the fire area contour. Based on the clarity assessment results, a comprehensive comparison is made between the data on the closure of the fire area's outline and the clarity of its edges. If the assessment results meet the preset standards, the final fire area outline image is output.

6. The aerial forest fire detection method according to claim 1, characterized in that, If the clarity assessment result of the fire edge meets the preset conditions, then through texture density distribution characteristic analysis technology, feature extraction is performed on the noise interference degree of the area boundary and the local brightness fluctuation amplitude. Combined with the brightness distribution uniformity and the contrast difference value within the area, the preliminary distribution characteristics of the fire area are determined, including: Step 1: Acquire raw image data of the fire scene using image acquisition equipment, perform preliminary segmentation processing on the fire edges and area boundaries, and preprocess the image using grayscale technology to obtain a grayscale image after preliminary segmentation; Step 2: From the initially segmented grayscale image, filter techniques are used to remove irrelevant interference information from the image, targeting region boundaries and noise interference, extracting key data related to texture density, and determining the clean image after processing; Step 3: Based on the cleaned image after processing, analyze the local brightness and fluctuation characteristics, and combine the uniform distribution of brightness to calculate the brightness change trend of each region in the image and obtain the brightness fluctuation distribution map; Step 4: Using the brightness fluctuation distribution map, combined with regional comparison and difference values, the support vector machine algorithm is used to classify the image regions, determine the distribution of fire areas and non-fire areas, and obtain the classification result map; Step 5: For the classification result map, analyze the correlation between distribution characteristics and texture density. If the texture density of a certain area in the classification result map is higher than the preset threshold, mark it as a potential fire core area and determine the core area distribution map. Step Six: Based on the core area distribution map, combined with the fire edge and clear standards, refine the boundaries of the potential fire core area to obtain the final fire area distribution map and complete the precise location of the fire area.

7. The aerial forest fire detection method according to claim 1, characterized in that, Based on the preliminary distribution characteristics of the fire area, spatial location mapping data from aerial images is fused. A combined processing approach of grayscale mapping accuracy analysis and detail texture clarity is employed to track local detail changes at adjacent points within the fire area. By combining local edge continuity and boundary texture density characteristics, the potential direction of fire spread is determined, resulting in the final fire monitoring distribution map, including: Preliminary distribution data of the fire area was obtained by aerial imagery, and the target area and background area were separated by image segmentation technology to obtain the initial division result of the fire area. Based on the initial division results, the spatial location information of the fire area is extracted. Combined with grayscale accuracy analysis, the brightness difference between adjacent points is calculated. If the brightness difference exceeds the preset threshold, it is determined to be a local area with significant changes, and potential active fire points are identified. For active fire points, analyze the texture details of adjacent points to obtain edge continuity information. If the edge continuity shows a linear expansion trend, it is judged as the preferred path for fire spread, and the preliminary distribution of spread paths is obtained. By calculating the boundary density change based on the initial distribution of the spread path, a convolutional neural network model is used to extract deep features of the texture details, thereby obtaining the density distribution features of the boundary region. Based on the density distribution characteristics, the potential direction of fire spread is tracked. If the density distribution characteristics show an increasing trend in a certain direction, then that direction is determined to be the main trend of fire spread, and the key monitoring area for fire expansion is identified. Acquire spatial location data of key monitoring areas, combine with updated aerial imagery, continuously track local changes and edge continuity, and determine the dynamic evolution trend of fire spread; By analyzing the dynamic evolution trend, a real-time distribution map of fire monitoring is generated, and spatial location and potential direction information are overlaid to obtain comprehensive monitoring results for the fire area.

Images