An adaptive flood inundation area extraction method fusing multi-source remote sensing data

Abstract

The present application belongs to the technical field of flood disaster monitoring, and specifically relates to a self-adaptive flood inundation area extraction method fusing multi-source remote sensing data. The method comprises the following steps: acquiring optical and SAR remote sensing image data; preprocessing the multi-source remote sensing images; fusing the optical and SAR remote sensing images; extracting a high-confidence discrete water body area; extracting a continuous water body area, fusing a digital elevation model, and constructing a high-quality water body data set; training a water body recognition deep learning model based on the water body data set; and extracting a flood inundation area using the water body recognition model. The present application solves the problems of threshold difficulty to determine, complex shadow interference, and less training data for deep learning when extracting a flood inundation area using remote sensing images. In addition, the method has strong robustness, high recognition accuracy, and can quickly calculate the flood inundation area, thereby providing technical support for flood prevention and disaster relief.

Description

Technical Field

[0001] This invention belongs to the field of flood disaster monitoring technology, specifically relating to a method for extracting the flood inundation range. Background Technology

[0002] Floods, as a natural disaster, have a profound impact on human society and the natural environment. Accurately and rapidly extracting the inundation area is crucial for effectively responding to floods. Currently, remote sensing technology is widely used in flood monitoring and assessment due to its ability to provide large-scale, timely surface information. However, existing methods for extracting the inundation area still suffer from limitations due to the reliance on single types of image data, such as optical or SAR remote sensing images. Optical remote sensing images provide high resolution and rich surface feature information under favorable weather conditions and without cloud cover, but become almost ineffective under adverse weather conditions such as clouds and rain, while floods are often caused by heavy rainfall. In contrast, SAR remote sensing images can penetrate clouds for all-weather observation, but their ability to describe fine surface features is inferior to that of optical images due to limitations in radar wavelength and incident angle. Flood inundation area extraction methods relying on a single data source are prone to misclassification under complex surface conditions, such as urban areas and vegetated areas. For example, angular reflection phenomena in SAR remote sensing images can cause water bodies to have similar reflection characteristics to wet surfaces, increasing the difficulty of classification. By fusing optical remote sensing imagery and SAR remote sensing imagery, we can combine the advantages of both and overcome the limitations of a single data source.

[0003] Thresholding methods can quickly extract water areas from SAR remote sensing images, but pixels located at water body boundaries are often difficult to define and are also affected by complex factors such as mountain shadows. Deep learning-based water extraction from remote sensing images can effectively solve the classification problem of water body boundary pixels. However, due to the lack of training data, deep learning-based methods tend to learn noise from the training set rather than the underlying true patterns, resulting in poor generalization ability of the model.

[0004] In summary, the above practical needs and problems reveal that the current field still lacks a comprehensive adaptive method for extracting flood inundation ranges. Existing methods suffer from difficulties in determining thresholds and interference from complex shadows. Furthermore, when using deep learning methods to extract flood inundation ranges, insufficient training data leads to poor model generalization ability, limiting their ability to identify water bodies only in specific areas. Therefore, developing an adaptive flood inundation range extraction method based on deep learning, applicable to different environments, to improve the accuracy and efficiency of flood monitoring, accelerate the acquisition of disaster information, and provide strong support for flood prevention and disaster relief, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive flood inundation range extraction method that integrates multi-source remote sensing data, so as to improve the accuracy and efficiency of flood monitoring, accelerate the acquisition of disaster information, and provide technical support for flood control and disaster relief.

[0006] The adaptive flood inundation range extraction method provided by the present invention, which integrates multi-source remote sensing data, includes six steps: acquiring multi-source remote sensing image data, preprocessing multi-source remote sensing image data, extracting high-confidence discrete water body areas, constructing a high-quality water body dataset, training a deep learning model for water body recognition, and automatically extracting the flood inundation range. Among them, multi-source remote sensing images include optical remote sensing images and SAR remote sensing images, which aim to overcome the limitations of a single data source; the preprocessing of multi-source remote sensing image data includes radiometric correction, atmospheric correction and image registration, etc., to achieve optimized processing and high-precision registration of remote sensing image data; high-confidence discrete water body regions are used to generate prompts for the Segment Anything Model (SAM) [1] to obtain continuous water body regions; the obtained water body regions are fused with Digital Elevation Model (DEM) data [2] to generate a high-quality water body dataset, which is used to train the water body recognition model (denoted as U-net model) constructed in this invention; using the trained water body recognition model, the water body in the remote sensing images is automatically extracted and its flood inundation range is obtained.

[0007] The specific details of each step are as follows:

[0008] Step 1: Acquire multi-source remote sensing image data

[0009] Multi-source remote sensing imagery includes optical remote sensing imagery and SAR remote sensing imagery. Specifically, optical remote sensing imagery data prior to the flood disaster event, as well as synthetic aperture radar (SAP) remote sensing imagery data prior to and during the event, are acquired as image data sources for automated extraction of flood inundation areas. The SAR remote sensing imagery data is Sentinel-1 SLC format data with VH polarization, and the optical remote sensing imagery data is Landsat 8OLI / TIRS C1Level-1 optical imagery. Digital elevation model (DEM) data is used as auxiliary data for screening water areas in the SAR remote sensing imagery; this invention employs the ASTER GDEM 30m resolution global digital elevation model dataset.

[0010] Step 2: Preprocessing of multi-source remote sensing image data;

[0011] This involves preprocessing the acquired optical and SAR remote sensing images. Specifically, multi-view processing is performed on the SLC-format SAR images to obtain intensity maps. The output of multi-view processing is an optimized image with lower speckle noise, improved spatial and radiometric resolution, and enhanced image usability and interpretability. Based on the obtained intensity maps, a Lee filter (7*7) is applied. The filter output is controlled using the local statistical characteristics of the image to remove the inherent speckle noise of the SAR image. The filtered image has reduced noise, is clearer, and water bodies are more prominent. The filtered result is then radiometrically calibrated, converting the intensity values ​​of the remote sensing image into backscattering coefficients. The acquired Landsat 8OLI / TIRS C1 Level-1 optical images have already undergone geometric and topographic corrections and are directly subjected to radiometric and atmospheric correction. Radiometric calibration eliminates errors generated by the sensor itself, while atmospheric correction eliminates the influence of atmospheric and illumination factors on ground reflection. The radiometric calibration type is radiance values ​​with a coefficient of 0.1, and the atmospheric correction uses the FLASSH model.

[0012] The optical and SAR remote sensing images, after radiometric and atmospheric correction, are registered. Registration is performed based on NDF and an improved SIFT: When constructing the scale spaces of the optical and SAR images, NDF is used instead of Gaussian filtering to construct the Harris scale space, which better preserves image edge information while denoising. Furthermore, for gradient calculation, multi-scale Sobel and multi-scale ROEWA operators are used to calculate the gradients of the optical and SAR images, respectively, resulting in consistent gradient information. GLOH templates are used to construct descriptors to increase their robustness. Finally, by constraining feature matching using the scale, position, and orientation of matching points, false matches are effectively eliminated. High-precision registration of the optical and SAR remote sensing images is achieved.

[0013] Step 3: Extract high-confidence discrete water body regions;

[0014] Calculate the water body index from optical remote sensing imagery and SAR remote sensing imagery. Wherein:

[0015] Optical remote sensing images use a normalized water index, the formula of which is:

[0016]

[0017] In the formula, Green and NIR are the brightness values ​​of the green band and near-infrared band, respectively.

[0018] The NDWI value range is [-1, 1]. To facilitate threshold selection and improve its accuracy, the NDWI index range is stretched from [-1, 1] to [0, 2000]. Using the histogram bimodal method, the gray level corresponding to the valley between the two peaks is selected as the image segmentation threshold. Areas above the threshold are considered high-confidence water bodies in the image.

[0019] For SAR remote sensing images, based on the relationship between VV and VH polarization images and water body information, the SDWI water body index is used to extract the extent of water bodies in SAR images. The formula is as follows:

[0020] SDWI=ln(10*VV*VH), (2)

[0021] In the formula, SDWI is the water body extraction index, VV and VH represent the values ​​obtained by the two polarization methods, respectively, and when the SDWI value is greater than 0, it is a high-confidence water body.

[0022] The overlapping area of ​​high-confidence water bodies obtained from optical remote sensing images and SAR remote sensing images is used as a candidate area for high-confidence water bodies.

[0023] Step 4: Construct a high-quality water body dataset;

[0024] Based on the high-confidence candidate water bodies, pixels are randomly selected as prompts for the Segment Anything Model to extract water bodies from continuous regions. Topographic slope data is calculated using a Digital Elevation Model (DEM) to remove mountain shadows from the continuous water body areas, resulting in high-quality water body extent data. Based on SAR remote sensing image data and the high-quality water body extents, labels are manually assigned to construct a high-quality water body dataset.

[0025] Step 5: Train the deep learning model for water body recognition

[0026] The deep learning model for water body identification adopts the U-Net model, whose structure includes three parts: encoder, decoder, and skip connections; among which:

[0027] The encoder employs a ResNet101 (Residual Network-101) network (a deep convolutional neural network architecture used for computer vision tasks such as image classification and object detection). It extracts features through residual blocks, mitigating the vanishing gradient problem and effectively capturing deep image features. The encoder consists of multiple convolutional layers, batch normalization layers, and ReLU activation functions, with each residual block containing several convolutional layers. Each stage of the encoder reduces the spatial size of the feature map while increasing the number of channels, thereby extracting features at different levels. The encoder first gradually reduces the spatial dimension of the input image through a series of convolutional operations, ensuring the extraction of high-level features. The introduction of batch normalization layers helps accelerate the training process and improve model stability, while the ReLU activation function enhances the model's non-linear expressive power. To better recover detailed information during decoding, the U-Net model introduces skip connections between the encoder and decoder.

[0028] The decoder progressively restores the spatial resolution of the feature maps by expanding their size through upsampling operations (such as transposed convolution). Each decoder stage includes convolutional layers, batch normalization layers, and ReLU activation functions, and is fused with feature maps passed through skip connections. The decoder restores the spatial dimensions of the feature maps layer by layer, ultimately outputting a water body mask of the same size as the input image. Specifically, the decoder restores the spatial resolution of the image through layer-by-layer upsampling operations and further enhances the accuracy of the segmentation results by fusing with feature maps passed through skip connections. Each upsampling process is accompanied by convolutional operations and activation functions, ensuring that the output image has high-quality spatial and semantic information.

[0029] The skip connections directly pass feature maps from each layer of the encoder to the corresponding layers of the decoder, ensuring that high-resolution features are fully utilized during the decoding process. This design allows the model to combine low- and high-level features during decoding, thereby improving segmentation accuracy. Specifically, skip connections operate at each layer of the U-Net model. By combining the high-resolution features retained in the encoder stage with the upsampling results in the decoder stage, the model can better recover the boundary and detail information of the target region, thus improving the accuracy of the segmentation results.

[0030] The specific training process of the model is as follows: First, a high-quality water body dataset is input into the U-Net model as training data. Second, the encoder uses ResNet101 to extract deep features from the image, and obtains feature maps at different scales through multiple convolutional operations. Skip connections pass the feature maps from each layer of the encoder to the decoder, ensuring that high-resolution features are fully utilized during the decoding process. The decoder restores the spatial resolution of the feature maps layer by layer, expands the feature map size through upsampling operations, and fuses it with the feature maps from the skip connections. Finally, the model outputs a water body range mask of the same size as the input image, and extracts water body edge pixels through post-processing steps to provide a high-precision water body range.

[0031] Through the above process, the U-Net model can accurately extract the water body extent in complex remote sensing images and solve the problem of difficult-to-define water body edge pixels. Combining a high-quality water body dataset with the ResNet101 feature extractor, the model performs exceptionally well in extracting water body extents.

[0032] Step 6: Automatically extract the flood inundation area

[0033] Using the aforementioned deep learning model for water body identification, the extent of water bodies in SAR remote sensing images from the pre-flood and mid-flood periods is extracted. The extracted water body extents are then converted to a vector surface format to calculate the difference between the water body extents in the mid-flood and pre-flood periods; the result is the inundation area of ​​this flood.

[0034] This invention overcomes the limitations of single data sources and traditional threshold segmentation methods, improves the accuracy and robustness of flood inundation range extraction, and provides stronger technical support for flood control and disaster relief. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method of the present invention.

[0036] Figure 2 This is a Gaofen-1 remote sensing image of Jiashan County.

[0037] Figure 3 This is a remote sensing image of Sentinel-1 in Jiashan County.

[0038] Figure 4 This is DEM data for Jiashan County.

[0039] Figure 5 This is a partial image of Sentinel-1 after preprocessing.

[0040] Figure 6 This is a partial image of Gaofen-1 after preprocessing.

[0041] Figure 7The results show the water extraction results (partial). Among them, a is the water extraction result from Gaofen-1 optical image, b is the water extraction result from Sentinel-1 image, and c is the result of overlaying the two.

[0042] Figure 8 This is a partial result of high-quality water extraction.

[0043] Figure 9 This refers to the extent (local) of the flood inundation. Detailed Implementation

[0044] The present invention is illustrated using Jiashan County, Zhejiang Province as an example, and is further described in conjunction with the accompanying drawings.

[0045] The provided adaptive flood inundation range extraction method, which integrates multi-source remote sensing data, has the following flowchart: Figure 1 As shown.

[0046] Step 1: Acquire multi-source remote sensing image data

[0047] Taking Jiashan County, Zhejiang Province as the study area, multi-source remote sensing imagery was used as the data source to automate the extraction of flood inundation areas. Optical remote sensing imagery was selected from Gaofen-1 acquired on August 12, 2019; Synthetic Aperture Radar (SAR) remote sensing imagery was selected from Sentinel-1 SLC format imagery acquired on August 4, 2019, with VH polarization (see Appendix). Figure 2 , 3 In addition, to assist in the selection of water bodies in SAR remote sensing imagery, the ASTER GDEM global digital elevation model (DEM) dataset with a resolution of 30 meters was used. See also Figure 4 .

[0048] Step 2: Preprocessing of multi-source remote sensing image data

[0049] The acquired optical and SAR remote sensing image data require preprocessing. First, multi-view processing is performed on the SLC-format SAR remote sensing image to generate an intensity map. Multi-view processing optimizes the image, reduces speckle noise, and improves spatial and radiometric resolution, thereby enhancing the image's usability and interpretability. Based on the generated intensity map, a 7×7 window Lee filter is applied. By utilizing the local statistical characteristics of the image to control the filter output, the inherent speckle noise of the SAR image is removed. The filtered image exhibits significantly reduced noise, resulting in a clearer image and more prominent water bodies. Next, the filtering results are radiometrically calibrated, converting the intensity values ​​of the remote sensing image into backscattering coefficients. Figure 5 , Figure 6 This is the image after preprocessing.

[0050] Radiometric calibration and atmospheric correction were performed on the acquired Gaofen-1 optical images. Radiometric calibration was used to eliminate errors introduced by the sensor, while atmospheric correction was used to eliminate the influence of atmospheric and illumination factors on ground reflection. The radiometric calibration type was radiance values ​​with a coefficient of 0.1, and the atmospheric correction used the FLAASH model.

[0051] Optical and SAR remote sensing images, after radiometric and atmospheric correction, were registered. The registration process was based on NDF and a modified SIFT method. When constructing the scale spaces of the optical and SAR images, NDF was used instead of Gaussian filtering to better preserve image edge information while denoising. For gradient calculation, multi-scale Sobel and multi-scale ROEWA operators were used to calculate the gradients of the optical and SAR images, respectively, to obtain gradient information with good consistency. GLOH templates were used to construct descriptors to enhance their robustness. Finally, by constraining feature matching using the scale, location, and orientation of matching points, mismatches were effectively eliminated, achieving high-precision registration between the optical and SAR images.

[0052] Step 3: Extract high-confidence discrete water body regions

[0053] Calculate the water index in optical and SAR remote sensing images.

[0054] For optical remote sensing imagery, the Normalized Difference Water Index (NDWI) is used, and its calculation formula is as follows:

[0055]

[0056] In the formula, Green and NIR are the brightness values ​​of the green band and near-infrared band, respectively, and the calculated NDWI value range is [-1, 1]. To facilitate threshold selection and improve accuracy, the NDWI value range is stretched to [0, 2000]. Using the histogram bimodal method, the gray level corresponding to the valley between the two peaks is selected as the image segmentation threshold, and the area above the threshold is considered a high-confidence water body area.

[0057] For SAR remote sensing imagery, the Water Body Extraction Index (SDWI) is calculated using the relationship between VV and VH polarization images and water body information. The calculation formula is as follows:

[0058] SDWI = ln(10 * VV * VH)

[0059] In the formula, SDWI is the water body extraction index, VV and VH represent the values ​​obtained by the two polarization methods, respectively, and when the SDWI value is greater than 0, it is a high-confidence water body.

[0060] The overlapping area of ​​high-confidence water bodies obtained from optical remote sensing imagery and SAR remote sensing imagery is used as a candidate area for high-confidence water bodies. See also Figure 7 As shown.

[0061] Step 4: Construct a high-quality water body dataset

[0062] Based on the high-confidence water body candidate regions, pixels are randomly selected as prompts for the SegmentAnything Model to extract water bodies from continuous areas. Topographic slope data is calculated using a Digital Elevation Model (DEM) to remove mountain shadow interference from continuous water body regions, resulting in high-quality water body extent data. Based on SAR remote sensing image data and the high-quality water body extents, labels are manually assigned to construct a high-quality water body dataset. See also... Figure 8 .

[0063] Step 5: Train the deep learning model for water body recognition

[0064] The constructed high-quality water body dataset is imported into the U-Net model for training. A ResNet101 network is used as the feature extractor, and a residual learning mechanism is introduced to address the vanishing gradient problem, thereby effectively extracting deep features from the image. Integrating ResNet101 into the encoder part of the U-Net model significantly enhances the model's feature extraction capability when processing complex remote sensing images. When extracting water body extents, this method effectively solves the problem of difficult-to-define water body edge pixels, providing a high-precision water body extent mask.

[0065] Step 6: Automatically extract the flood inundation area

[0066] Using a trained deep learning model for water body identification, the extent of water bodies in SAR remote sensing images during the pre-flood and mid-flood periods was extracted. The extracted water body extents were converted to vector surface format to calculate the change in water body extent during the mid-flood period relative to the pre-flood period. The result represents the inundation extent of this flood. (See [link to relevant documentation]). Figure 9 .

[0067] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0068] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. References:

[0069] [1] The Segment Anything Model (SAM) originated from Facebook's Meta AI Lab. This model can identify and segment any object based on text commands or image recognition. SAM reference paper: Kirillov A, Mintun E, Ravi N, et al. Segment anything[C] / / Proceedings of the IEEE / CVF International Conference on Computer Vision.2023:4015-4026.

[0070] [2] Digital Elevation Model (DEM) is a digital simulation of ground topography (i.e., a digital representation of the surface morphology of the terrain) achieved through limited terrain elevation data. It is a physical ground model that represents ground elevation in the form of an ordered numerical array.

Claims

1. An adaptive flood inundation range extraction method that integrates multi-source remote sensing data, characterized in that, The specific steps are as follows: Step 1: Acquire multi-source remote sensing image data; Multi-source remote sensing imagery includes optical remote sensing imagery and SAR remote sensing imagery. Specifically, optical remote sensing imagery data prior to the flood disaster event, as well as synthetic aperture radar (SAR) remote sensing imagery data prior to and during the event, are acquired as image data sources for the automated extraction of flood inundation areas. The SAR remote sensing imagery data is Sentinel-1 SLC format data with VH polarization, and the optical remote sensing imagery data is Landsat 8 OLI / TIRS C1 Level-1 optical imagery. Digital elevation model (DEM) data is used as auxiliary data to remove mountain shadow interference. Step 2: Preprocessing of multi-source remote sensing image data; The acquired optical and SAR remote sensing images were preprocessed. Specifically, multi-view processing was performed on the SLC-format SAR images to obtain intensity maps. The output of the multi-view processing was an optimized image. Based on the obtained intensity maps, a 7x7 Lee filter was applied, utilizing the local statistical characteristics of the image to control the filter output and remove inherent speckle noise from the SAR images. The filtering results were then radiometrically calibrated, converting the intensity values ​​of the remote sensing images into backscattering coefficients. The acquired Landsat 8 OLI / TIRS C1 Level-1 optical images, having already undergone geometric and topographic correction, were directly subjected to radiometric and atmospheric correction. Radiometric calibration was used to eliminate errors generated by the sensor itself, while atmospheric correction was used to eliminate the influence of atmosphere and illumination on ground reflections. Register the radiometrically calibrated and atmospherically corrected optical remote sensing images with SAR remote sensing images; Step 3: Extract high-confidence discrete water body regions; Calculate the water index from optical remote sensing imagery and SAR remote sensing imagery; where: Optical remote sensing images use a normalized water index, the formula of which is: ， （1） In the formula, Green and NIR are the brightness values ​​of the green band and near-infrared band, respectively; The NDWI value range is [-1, 1]. The NDWI index value range is stretched from [-1, 1] to [0, 2000]. Using the histogram bimodal method, the gray level corresponding to the valley between the two peaks is selected as the image segmentation threshold. Areas above the threshold are high-confidence water bodies in the image. For SAR remote sensing images, based on the relationship between VV and VH polarization images and water body information, the SDWI water index is used to extract the extent of water bodies in the SAR image. The formula is as follows: ， （2） In the formula, SDWI is the water body extraction index, VV and VH represent the values ​​obtained by the two polarization methods, respectively. When the SDWI value is greater than 0, it is a high-confidence water body. The overlapping area of ​​high-confidence water bodies obtained from optical remote sensing images and SAR remote sensing images is used as a candidate area for high-confidence water bodies. Step 4: Construct a high-quality water body dataset; Based on the high-confidence water body candidate region, pixels are randomly selected as prompts for segmenting the entire model to extract water bodies in continuous regions. Topographic slope data is calculated using a digital elevation model to remove mountain shadow interference in continuous water body regions. After processing, high-quality water body range data is obtained. Based on SAR remote sensing image data and the high-quality water body range, labels are manually assigned to construct a high-quality water body dataset. Step 5: Train the deep learning model for water body recognition; The deep learning model for water body identification adopts the U-Net model, whose structure includes three parts: encoder, decoder, and skip connections; among which: The encoder uses a ResNet101 network; the encoder consists of multiple convolutional layers, batch normalization layers and ReLU activation functions, and each residual block contains several convolutional layers. The decoder is able to gradually restore the spatial resolution of the feature map and expand the feature map size through upsampling operations. Each decoder stage contains a convolutional layer, a batch normalization layer, and a ReLU activation function, and is fused with the feature map passed by skip connections. The decoder restores the spatial size of the feature map layer by layer and finally outputs a water body range mask with the same size as the input image. The skip connection directly passes the feature maps of each layer of the encoder to the corresponding layer of the decoder, ensuring that high-resolution features can be fully utilized during the decoding process. Specifically, the skip connection operates on each layer of the U-Net model by combining the high-resolution features retained in the encoder stage with the upsampling results in the decoder stage. The specific training process of the model is as follows: A high-quality water body dataset is input into the U-Net model as training data; the encoder uses ResNet101 to extract deep features of the image, and obtains feature maps of different scales through multiple convolution operations; skip connections pass the feature maps of each layer of the encoder to the decoder, ensuring that high-resolution features are fully utilized during the decoding process; the decoder restores the spatial resolution of the feature maps layer by layer, expands the size of the feature maps through upsampling operations, and fuses them with the feature maps of the skip connections; the model outputs a water body range mask of the same size as the input image. Step 6: Automatically extract the flood inundation area Using a trained deep learning model for water body identification, the extent of water bodies in SAR remote sensing images during the pre-flood and mid-flood periods is extracted. The extracted water body extents are then converted into vector surface format to calculate the difference between the water body extents during the mid-flood and pre-flood periods. The result is the flood inundation extent.

2. The adaptive flood inundation range extraction method according to claim 1, characterized in that, In step (2), the radiation calibration type is radiance value with a coefficient of 0.1, and the atmospheric correction uses the FLAASH model.

3. The adaptive flood inundation range extraction method according to claim 1, characterized in that, In step (2), the optical remote sensing image and SAR remote sensing image that have undergone radiometric calibration and atmospheric correction are registered. Specifically, the registration of optical remote sensing image and SAR remote sensing image is based on NDF and improved SIFT: when constructing the scale space of optical and SAR remote sensing images, NDF is used instead of Gaussian filtering to construct the Harris scale space; in addition, the gradients of optical and SAR images are calculated by multi-scale Sobel and multi-scale ROEWA operators respectively to obtain gradient information with good consistency; descriptors are constructed using GLOH templates; finally, by using the scale, position and orientation of matching points to constrain feature matching, mismatches are effectively eliminated.

Images