A method for estimating the early fire area in forest fire emergency rescue scenarios

Abstract

The present invention provides a method for estimating the early fire area in a forest fire emergency rescue scenario, which relates to the field of fire technology. The method includes: in the early stage of forest fire emergency rescue, using a panoramic camera mounted on an unmanned aerial vehicle to collect fire images, and performing denoising and thin smoke removal on the panoramic images; obtaining pre-disaster satellite remote sensing images according to the location of the forest fire, and performing orthorectification and cropping processing; using the processed pre-disaster satellite remote sensing images as reference images, registering the processed post-disaster unmanned aerial vehicle panoramic images; comparing the pre-disaster remote sensing images with the registered post-disaster unmanned aerial vehicle panoramic images, extracting the fire area, and counting the fire area. The present invention solves the problems of field of view obstruction, estimation efficiency and estimation accuracy not being compatible in the existing fire area estimation methods, and meets the actual needs of forest fire emergency rescue.

Description

Technical Field

[0001] The present invention belongs to the field of fire technology, and in particular relates to a method for estimating the early fire area in a forest fire emergency rescue scenario. Background Art

[0002] In the emergency rescue of forest fires, "fight early, fight small, and fight hard" is the basic principle, which means that the forest fire must be extinguished in the initial stage. Rapid estimation of the fire area and scope in the early stage of the fire is a crucial step. It helps emergency rescue command decision makers to formulate scientific and effective firefighting plans and reduce losses caused by the fire.

[0003] At present, there are usually two methods to estimate the area of ​​forest fires. One is manual on-site visual observation, and the other is to monitor and identify orthophoto data obtained by drone aerial survey and satellite remote sensing. The former is simple, convenient and time-effective, but it relies on manual experience and is often limited by the visual field in mountainous areas, making it impossible to see the whole picture. The assessment accuracy is low and cannot support precise rescue. The latter has high assessment accuracy, but the acquisition time of drone aerial survey data is long, and the transit time of remote sensing satellites may not meet the needs of disaster relief. Forest fires develop rapidly, and assessments based on drone aerial survey and satellite remote sensing are not conducive to early extinguishing of forest fires.

[0004] In forest fire emergency rescue, local forest fire fighting teams are usually the first to go out and arrive at the fire scene with equipment such as drones. In order to fully understand the fire situation and provide on-site data to the rear command team, forest fire rescue personnel can immediately launch drones after arriving at the scene, take panoramic images at a suitable location, and transmit them to the rear. Compared with drone aerial surveys, taking panoramic photos with drones usually only takes a few minutes and requires lower skills from technicians. However, compared with orthophotos obtained by drone aerial surveys, panoramic images only have the coordinates of the shooting center position, lack the coordinate information of each pixel, and cannot be directly used for the measurement and estimation of the fire area. Therefore, the study of early fire area estimation based on panoramic images can not only avoid the problems of visual field obstruction and low evaluation accuracy in manual on-site visual inspections, but also solve the problem of poor timeliness in drone aerial surveys or satellite remote sensing monitoring. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for estimating the early fire area in a forest fire emergency rescue scenario, which solves the problems of field of view obstruction and the lack of estimation efficiency and estimation accuracy in the existing fire area estimation methods.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is: a method for estimating the early fire area in a forest fire emergency rescue scenario, and its implementation method is as follows:

[0007] S1. In the early stage of forest fire emergency rescue, a drone equipped with a panoramic camera is used to collect fire scene images, and the panoramic images are denoised and smoke removed;

[0008] S2. Obtain pre-disaster satellite remote sensing images based on the location of the forest fire, and perform orthorectification and cropping;

[0009] S3, using the pre-disaster satellite remote sensing image processed by S2 as the reference image, register the post-disaster UAV panoramic image processed by S1;

[0010] S4. Compare the pre-disaster remote sensing images with the registered post-disaster drone panoramic images, extract the fire area, and count the fire area to complete the estimation of the early fire area.

[0011] The beneficial effect of the present invention is: in order to meet the actual needs of forest fire emergency rescue and solve the problems of blocked field of view and lack of estimation efficiency and estimation accuracy in existing fire area estimation methods, the present invention is based on the actual needs of forest fire emergency rescue. In the early stage of forest fire fighting, based on drone panoramic images, intelligent alignment is performed with pre-disaster high-resolution satellite remote sensing images, and coordinates are marked for each pixel of the panoramic image. By comparing pre-disaster and post-disaster images, the fire area is quickly extracted, and the fire area is statistically calculated to provide data support for early forest fire rescue.

[0012] Furthermore, the S1 comprises the following steps:

[0013] S101. In the early stage of emergency rescue of forest fires, a panoramic camera is used to take panoramic images above the fire scene, where the panoramic images cover the entire fire scene;

[0014] S102, denoising the panoramic image using an adaptive median filtering algorithm;

[0015] S103, using a forest fire drone panoramic image thin smoke removal algorithm to perform thin smoke removal processing on the denoised panoramic image.

[0016] The beneficial effect of the above further scheme is: by obtaining panoramic images at the early stage of forest fire rescue and performing denoising and thin smoke processing, it can provide clear and intuitive on-site data to emergency decision-makers at the first moment of rescue, while laying a good foundation for subsequent image registration and fire area assessment.

[0017] Furthermore, the S103 includes the following steps:

[0018] A1. Construct an atmospheric scattering model;

[0019] A2. According to the characteristics of drone panoramic image shooting and the principle of atmospheric scattering model, the mean value of the pixel value of the sky in the image is selected as the estimated value of atmospheric light A;

[0020] A3. Construct a local window and estimate the smoke density index using the pixel value changes within the local window;

[0021] A4. Dynamically adjust the adjustment factor based on the estimated value of the smoke density index in the local window;

[0022] A5, estimating the transmittance of each pixel using a local minimum filter according to the estimated value of the atmospheric light A and the estimated value of the adjustment factor;

[0023] A6. According to the atmospheric scattering model, the smoke-free image is restored using the estimated value of transmittance and atmospheric light A, completing the smoke removal process for the panoramic image.

[0024] The beneficial effects of the above further scheme are: constructing a thin smoke removal algorithm for panoramic images of forest fire drones based on the atmospheric scattering model, taking into full account the characteristics of panoramic images of forest fires and the differences in smoke concentrations in the fire scene, estimating atmospheric light and transmittance, and then restoring image quality. The algorithm has high computational efficiency, does not require a complex optimization process, can process panoramic images in real time, and can flexibly process areas with different smoke concentrations, thereby improving the accuracy and effect of smoke removal processing and meeting the actual needs of forest fire emergency rescue.

[0025] Furthermore, the expression of the atmospheric scattering model in A1 is as follows: ;in, Indicates that there is smoke in the image The pixel value of the position, Indicates no smoke image The pixel value of the position, express The transmittance of the position, represents atmospheric light, Indicates the pixel position.

[0026] Furthermore, the expression of the smoke-free image in A4 is as follows: ; ; ; ; ;in, Indicates that there is smoke in the image The pixel value of the position, represents a constant, Indicates no smoke image The pixel value of the position, express Position transmittance, represents atmospheric light, N represents the number of pixels in the sky area, Represents the pixel value of the sky area in the panoramic image, express The adjustment factor for position, Indicates is the local window centered on Indicates the smoke density index. represents the pixel value in the local window of the smoke image, ω0 represents the basic adjustment factor, k represents the gain factor, represents the pixel position, Represents the pixel position within the local window.

[0027] Furthermore, the S2 comprises the following steps:

[0028] S201. Obtain pre-disaster satellite remote sensing images based on the location of the forest fire;

[0029] S202, performing orthorectification on the pre-disaster satellite remote sensing image to obtain an orthoimage;

[0030] S203. Crop the orthophoto according to the distribution of the fire scene in the panoramic image, wherein the range of the cropped pre-disaster satellite remote sensing image is larger than the range covered by the panoramic image.

[0031] Furthermore, the S3 comprises the following steps:

[0032] S301, performing edge enhancement preprocessing on the pre-disaster satellite remote sensing images processed by S2 and the post-disaster drone panoramic images processed by S1;

[0033] S302, detecting feature points on the pre-disaster satellite remote sensing images and post-disaster drone panoramic images that have been pre-processed;

[0034] S303, performing rough matching on the feature points to obtain an initial matching point set of the image to be registered and the reference image;

[0035] S304, remove the mismatched points from the initial matching point set, estimate the best transformation model, and apply the best transformation model to the post-disaster UAV panoramic image, so that the post-disaster UAV panoramic image is matched with the pre-disaster satellite remote sensing image, and the geographic reference coordinate information of the pre-disaster satellite remote sensing image is assigned to the post-disaster UAV panoramic image to complete the registration.

[0036] The beneficial effect of the above further scheme is: according to the actual needs of forest fire emergency rescue, the present invention uses drones to obtain panoramic images of the fire scene as soon as possible, and through intelligent alignment with high-resolution satellite remote sensing images before the disaster, the coordinates of each pixel of the panoramic image are marked, and the panoramic image is given a geographic coordinate reference. By comparing the images before and after the disaster, the fire area is quickly extracted, and the fire area is statistically calculated, providing data support for early rescue of forest fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0038] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0039] Example

[0040] like Figure 1 As shown, the present invention provides a method for estimating the early fire area in a forest fire emergency rescue scenario, and the implementation method is as follows:

[0041] S1. In the early stage of forest fire emergency rescue, a panoramic camera equipped with a drone is used to collect fire scene images, and the panoramic images are denoised and smoke removed. The implementation method is as follows:

[0042] S101. In the early stage of emergency rescue of forest fires, a panoramic camera is used to take panoramic images above the fire scene, where the panoramic images cover the entire fire scene;

[0043] S102, denoising the panoramic image using an adaptive median filtering algorithm;

[0044] S103, using the forest fire drone panoramic image thin smoke removal algorithm, the denoised panoramic image is subjected to thin smoke removal processing, and the implementation method is as follows:

[0045] A1. Construct an atmospheric scattering model;

[0046] A2. According to the characteristics of drone panoramic image shooting and the principle of atmospheric scattering model, the mean value of the pixel value of the sky in the image is selected as the estimated value of atmospheric light A;

[0047] A3. Construct a local window and estimate the smoke density index using the pixel value changes within the local window;

[0048] A4. Dynamically adjust the adjustment factor based on the estimated value of the smoke density index in the local window;

[0049] A5, estimating the transmittance of each pixel using a local minimum filter according to the estimated value of the atmospheric light A and the estimated value of the adjustment factor;

[0050] A6. According to the atmospheric scattering model, the smoke-free image is restored using the estimated value of transmittance and atmospheric light A, completing the smoke removal process for the panoramic image.

[0051] In this embodiment, after a forest fire occurs, in the early stage of emergency rescue, a drone equipped with a high-definition panoramic camera is used to shoot panoramic images above the fire scene, and the panoramic image should cover the entire fire scene. During the specific shooting process, the height of the drone can be adjusted to achieve coverage of the entire fire scene, and the longitude and latitude coordinates of the shooting center position are recorded at the same time. .

[0052] In this embodiment, an adaptive median filtering algorithm is used to denoise the panoramic image, thereby reducing noise information generated by the sensor itself and the surrounding environment during image acquisition and transmission.

[0053] In this embodiment, a thin smoke removal algorithm for panoramic images of forest fire drones is constructed based on the atmospheric scattering model to remove smoke from panoramic images and improve image clarity. After a forest fire occurs, the smoke produced by the burning of the forest fire will make the overall image of the scene too bright, and some objects will be blocked. In order to improve the quality of the panoramic image and improve the image clarity, the panoramic image smoke removal process is carried out according to the following steps:

[0054] 1. Establish an atmospheric scattering model. The atmospheric scattering model describes the formation process of smoke images. Assume that The pixel value at position is , there is smoke in the image The pixel value at position is , atmospheric light is A, The transmittance at the position is , then the smoke image can be expressed as: ;in, represents the pixel position, Represents transmittance, which describes the degree to which light from the light source to the camera penetrates the medium, and its value range is between [0,1].

[0055] 2. Atmospheric light estimation. According to the characteristics of drone panoramic image shooting, the average value of the pixel value of the sky in the image is selected as the estimated value of atmospheric light A: ;in, N represents the number of pixels in the sky area, Represents the pixel value of the sky area in the panoramic image, It means the sky, Indicates the pixel position.

[0056] 3. Construct a local window and use the pixel value changes within the local window to estimate the smoke density index. Areas with higher smoke density usually have lower contrast and higher brightness. For each pixel position, define a local window (such as 3×3, 5×5, etc.), then the smoke density index It can be expressed as: ;in, Indicates is the local window centered on is the pixel value in the local window of the smoke image, Represents the pixel position within the local window.

[0057] 4. Dynamically adjust the adjustment factor based on the estimated value of the smoke density index in the local window and define the adjustment factor for: ; Among them, ω0 represents the basic adjustment factor, which is used to control the degree of smoke removal. Usually, the value range of ω0 is between [0.9, 0.95]; k Represents the gain factor, which is used to control the range of the adjustment factor. It is usually between [0,0.1]. When the smoke concentration is high ( lower), Close to ω0; when the smoke concentration is low ( Higher), Increase, thereby enhancing the smoke removal effect;

[0058] 5. Transmittance estimation. Based on the estimated value of atmospheric light A and the estimated value of the adjustment factor, the transmittance of each pixel is estimated using a local minimum filter. In the constructed local window, a minimum pixel value is found. This local minimum value approximately represents the brightness in the absence of smoke. Transmittance It can be calculated by the following formula: ;in, Indicates The local window at the center.

[0059] 6. Image restoration. Using the estimated transmittance and atmospheric light A , restore the smoke-free image based on the atmospheric scattering model : ;in, Indicates a constant (such as 0.001) to prevent the denominator from being zero.

[0060] S2. According to the location of the forest fire, obtain the pre-disaster satellite remote sensing image, and perform orthorectification and cropping. The implementation method is as follows:

[0061] S201. Obtain pre-disaster satellite remote sensing images based on the location of the forest fire;

[0062] S202, performing orthorectification on the pre-disaster satellite remote sensing image to obtain an orthoimage;

[0063] S203. Crop the orthophoto according to the distribution of the fire scene in the panoramic image, wherein the range of the cropped pre-disaster satellite remote sensing image is larger than the range covered by the panoramic image.

[0064] In this embodiment, after a forest fire occurs, a search is immediately conducted to obtain high-resolution satellite remote sensing images of the forest fire area. Under the premise of ensuring coverage of the entire fire area, the higher the spatial resolution of the image and the closer the shooting time is to the current time, the better. The acquired satellite remote sensing image is orthorectified to obtain an orthorectified image. This step is usually completed in remote sensing image processing software such as ENVI. During emergency rescue, the satellite remote sensing images stored before the disaster are usually orthorectified and do not need to be processed at this time. According to the distribution of the fire in the panoramic image, the orthorectified satellite remote sensing image is cropped so that the cropped satellite remote sensing image range is slightly larger than the range covered by the panoramic image. The specific method is to position the satellite remote sensing image to the center of the panoramic image shooting in geographic information software such as ENVI and ArcGIS. By visual interpretation, the cropping area can be quickly selected to complete the image cropping. Compared with the registration of the entire satellite remote sensing image, after cropping, the search and comparison range is reduced, which can greatly reduce the calculation amount of image registration.

[0065] S3, using the pre-disaster satellite remote sensing image processed by S2 as the reference image, register the post-disaster drone panoramic image processed by S1. The implementation method is as follows:

[0066] S301, performing edge enhancement preprocessing on the pre-disaster satellite remote sensing images processed by S2 and the post-disaster drone panoramic images processed by S1;

[0067] S302, detecting feature points on the pre-disaster satellite remote sensing images and post-disaster drone panoramic images that have been pre-processed;

[0068] S303, performing rough matching on the feature points to obtain an initial matching point set of the image to be registered and the reference image;

[0069] S304, remove the mismatched points from the initial matching point set, estimate the best transformation model, and apply the best transformation model to the post-disaster UAV panoramic image, so that the post-disaster UAV panoramic image is matched with the pre-disaster satellite remote sensing image, and the geographic reference coordinate information of the pre-disaster satellite remote sensing image is assigned to the post-disaster UAV panoramic image to complete the registration.

[0070] In this embodiment, the Canny algorithm is used to perform edge enhancement preprocessing on the satellite remote sensing image and the drone panoramic image to reduce the influence of the panoramic image's own geometric deformation and viewpoint change on the registration. The accelerated robust feature algorithm SURF is applied to the preprocessed satellite remote sensing image and drone panoramic image to detect feature points and calculate descriptors. The accelerated robust feature algorithm SURF locates feature points by constructing an integral image and finding local extreme values, and generates a descriptor based on the Haar wavelet Haar response. The fast nearest neighbor search library algorithm FLANN is used to perform rough matching of feature points to obtain the initial matching point set of the image to be registered and the reference image. The random sample consistency algorithm RANSAC is used to eliminate mismatched points, estimate the optimal transformation model, and apply the transformation model to the drone panoramic image to match it with the remote sensing image, and assign the geographic reference coordinate information of the remote sensing image to the panoramic image.

[0071] S4. Compare the pre-disaster remote sensing images with the registered post-disaster drone panoramic images, extract the fire area, and count the fire area to complete the estimation of the early fire area.

[0072] In this embodiment, the pre-disaster remote sensing image and the registered post-disaster drone image are compared, and the fire area is quickly extracted and counted in ArcGIS geographic information system software through visual interpretation.

[0073] In summary, compared with the existing forest fire area estimation technology, the advantages of the present invention are as follows:

[0074] (1) Forest fire area assessment based on drone panoramic images is more timely and more in line with the actual needs of emergency rescue.

[0075] Compared with the commonly used forest fire area assessment based on drone orthophotos or remote sensing images, drone panoramic images are easier and faster to obtain in the early stages of forest fire emergency rescue. In the early stages of a fire, the fire area is generally small, and panoramic images can meet the fire coverage requirements. By aligning post-disaster drone panoramic images with pre-disaster satellite remote sensing images, giving the panoramic images geographic information coordinate references, and comparing pre-disaster remote sensing images, the forest fire area can be quickly identified and extracted, providing data support for emergency rescue decisions, which is more in line with the actual needs of forest fire emergency rescue.

[0076] (2) The assessment of forest fire area based on UAV panoramic images is more accurate and scientific than manual on-site visual inspection.

[0077] Manual on-site visual inspection methods are highly dependent on manual experience and are often limited by the field of vision in mountainous areas. They are unable to see the entire picture, have low assessment accuracy, and cannot support precise rescue. Forest fire area assessment based on drone panoramic images is less restricted by terrain, has a more scientific method, and has a higher assessment accuracy.

[0078] (3) An algorithm for removing thin smoke from panoramic images of forest fires was constructed, which improved the clarity of panoramic images and helped improve the accuracy of forest fire area estimation.

[0079] The smoke produced by forest fires reduces the clarity of panoramic images to a certain extent. The smoke produced by forest fires will make the overall image of the scene too bright, and some objects will be blocked. By constructing a thin smoke removal algorithm for forest fire drone panoramic images, the thin smoke can be removed, which will improve the clarity of drone panoramic images and thus improve the estimation accuracy of forest fire area.

Claims

1. A method for estimating the early fire area in a forest fire emergency rescue scenario, characterized in that: The following steps are involved: S1. In the early stage of forest fire emergency rescue, a drone equipped with a panoramic camera is used to collect fire scene images, and the panoramic images are denoised and smoke removed; S2. Obtain pre-disaster satellite remote sensing images based on the location of the forest fire, and perform orthorectification and cropping; S3, using the pre-disaster satellite remote sensing image processed by S2 as the reference image, register the post-disaster UAV panoramic image processed by S1; S4. Compare the pre-disaster remote sensing images with the registered post-disaster drone panoramic images, extract the fire area, and count the fire area to complete the estimation of the early fire area; The S1 comprises the following steps: S101. In the early stage of emergency rescue of forest fires, a panoramic camera is used to take panoramic images above the fire scene, where the panoramic images cover the entire fire scene; S102, denoising the panoramic image using an adaptive median filtering algorithm; S103, using a forest fire drone panoramic image thin smoke removal algorithm to perform thin smoke removal processing on the denoised panoramic image; The S103 comprises the following steps: A1. Construct an atmospheric scattering model; The expression of the atmospheric scattering model in A1 is as follows: in, Indicates that there is smoke in the image The pixel value of the position, Indicates no smoke image The pixel value of the position, express The transmittance of the position, represents atmospheric light, Indicates the pixel position; A2. According to the characteristics of drone panoramic image shooting and the principle of atmospheric scattering model, the mean value of the pixel value of the sky in the image is selected as the estimated value of atmospheric light A; A3. Construct a local window and estimate the smoke density index using the pixel value changes within the local window; A4. Dynamically adjust the adjustment factor based on the estimated value of the smoke density index in the local window; A5, estimating the transmittance of each pixel using a local minimum filter according to the estimated value of the atmospheric light A and the estimated value of the adjustment factor; A6. According to the atmospheric scattering model, the smoke-free image is restored using the estimated value of transmittance and atmospheric light A, completing the smoke removal process for the panoramic image. The expression of the smoke-free image is as follows: in, Indicates that there is smoke in the image The pixel value of the position, represents a constant, Indicates no smoke image The pixel value of the position, express Position transmittance, represents atmospheric light, N represents the number of pixels in the sky area, Represents the pixel value of the sky area in the panoramic image, express The adjustment factor for position, Indicates is the local window centered on Indicates the smoke density index. represents the pixel value in the local window of the smoke image, ω0 represents the basic adjustment factor, k represents the gain factor, represents the pixel position, Represents the pixel position within the local window.

2. The method for estimating the early fire area in a forest fire emergency rescue scenario according to claim 1 is characterized in that: The S2 comprises the following steps: S201. Obtain pre-disaster satellite remote sensing images based on the location of the forest fire; S202, performing orthorectification on the pre-disaster satellite remote sensing image to obtain an orthoimage; S203. Crop the orthophoto according to the distribution of the fire scene in the panoramic image, wherein the range of the cropped pre-disaster satellite remote sensing image is larger than the range covered by the panoramic image.

3. The method for estimating the early fire area in a forest fire emergency rescue scenario according to claim 1 is characterized in that: The S3 comprises the following steps: S301, performing edge enhancement preprocessing on the pre-disaster satellite remote sensing images processed by S2 and the post-disaster drone panoramic images processed by S1; S302, detecting feature points on the pre-disaster satellite remote sensing images and post-disaster drone panoramic images that have been pre-processed; S303, performing rough matching on the feature points to obtain an initial matching point set of the image to be registered and the reference image; S304, remove the mismatched points from the initial matching point set, estimate the best transformation model, and apply the best transformation model to the post-disaster UAV panoramic image, so that the post-disaster UAV panoramic image is matched with the pre-disaster satellite remote sensing image, and the geographic reference coordinate information of the pre-disaster satellite remote sensing image is assigned to the post-disaster UAV panoramic image to complete the registration.

Images