A SAR image town flood mapping method based on multi-scale information fusion
By employing multi-scale information fusion and machine learning methods, the accuracy and reliability issues of SAR image flood mapping in complex urban scenarios were addressed, resulting in an efficient and automated flood mapping method that generates high-precision flood spatial distribution maps.
Patent Information
- Application Number
- CN202310406034.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-11
- Publication Date
- 2025-12-16
- Estimated Expiration
- 2043-04-11
AI Technical Summary
Existing flood mapping methods based on SAR imagery lack accuracy and reliability in complex urban scenarios, fail to fully mine and utilize multi-scale information, and involve significant human intervention, resulting in low mapping accuracy and reliability.
By acquiring and preprocessing SAR data before and after the flood, spatial and pseudo-color features are extracted to generate a flood probability map and perform multi-scale superpixel segmentation. Unsupervised flood mapping is performed using machine learning, and adaptive fusion is carried out through the principle of minimum uncertainty to generate a final reliable flood spatial distribution map.
It enables high-precision, automated flood mapping in complex urban scenarios, improving the accuracy and reliability of mapping, reducing manual intervention, and increasing computational efficiency and operability.
Smart Images
Figure CN116563678B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of remote sensing image processing, and in particular relates to a method for mapping urban floods based on SAR images using multi-scale information fusion. Background Technology
[0002] In recent years, large-scale flood disasters have occurred frequently around the world. Flood mapping based on SAR imagery can obtain timely and effective information on the spatiotemporal distribution of floods, which is of great help in the formulation of disaster prevention, mitigation, and post-disaster relief measures. Therefore, researching and proposing effective SAR flood mapping methods is a very important and necessary task.
[0003] Due to the limited information available in SAR image bands, current SAR-based flood mapping often involves thresholding SAR images before and after a flood to extract water bodies separately, then comparing the spatial distribution differences of water bodies before and after the flood to identify the flood-inundated area. This approach often yields unsatisfactory results. Furthermore, current SAR-based flood mapping research does not adequately mine and utilize multi-scale flood mapping information, and significant human intervention further reduces the accuracy and reliability of SAR flood mapping. Although many researchers have attempted to improve flood mapping accuracy by extracting and utilizing the spatial features of SAR images, there is still considerable room for improvement in the precision and reliability of SAR-based flood mapping, especially in urban scenarios with complex terrain. Therefore, researching how to effectively mine the information contained in SAR images to improve the precision and reliability of flood mapping and constructing efficient and automated SAR flood mapping methods is crucial. Summary of the Invention
[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a SAR imagery-based method for urban flood mapping based on multi-scale information fusion. This method first acquires SAR data of the target area before and after a flood, and performs necessary data preprocessing such as radiometric calibration, filtering, and terrain correction. Then, considering the complexity of urban scenes and the limited number of SAR data bands, spatial and pseudo-color features are extracted from the SAR intensity images before and after the flood, and these features are integrated to obtain feature sets for the periods before and after the flood. Subsequently, a flood probability map is generated based on the feature sets, and multi-scale superpixel segmentation is performed on it. Unsupervised flood mapping based on machine learning is then performed at different segmentation scales to generate a series of flood maps at different scales. Finally, the mapping uncertainty of the flood maps generated at different scales is evaluated, and based on the principle of minimum uncertainty, the flood maps at different scales are adaptively fused to generate a final reliable spatial distribution map of the flood.
[0005] The technical solution adopted in this invention is: a method for urban flood mapping based on SAR imagery using multi-scale information fusion, comprising the following steps:
[0006] Step 1: Acquire SAR data of the target area before and after the flood, and perform a series of necessary data preprocessing on the acquired SAR data, such as radiometric calibration, filtering and terrain correction.
[0007] Step 2: Considering the complexity of urban scenes and the limited number of SAR data bands, spatial features and pseudo-color features are extracted from the SAR intensity images before and after the flood. Spatial features include spatial morphological features and texture features, while pseudo-color features refer to features such as color saturation obtained through color image synthesis and color space transformation. All extracted feature images are then combined with the intensity images using layer stacking to form the feature sets before and after the flood.
[0008] Step 3: Use the log-ratio operator to calculate the difference between the feature sets before and after the flood, generating a series of difference features Δf. i (i = 1, 2, ..., N; and N represents the number of features contained in the feature set before and after the flood), let f i 1 with f i 2 If each feature corresponds to the i-th feature in the feature sets before and after the flood, then the i-th difference feature generated by the log-ratio operator can be represented as follows:
[0009]
[0010] In the above formula, log(·) represents the logarithmic function, and |·| represents the absolute value function.
[0011] Subsequently, by analyzing all the difference features Δf i The summation of (i = 1, 2, ..., N) generates a flood probability map (FPM), which reflects the probability that each pixel in the image belongs to a flood-inundated area.
[0012] Step 4: Set a series of scale parameters k1, k2, ..., kn from small to large (these scale parameters represent the expected number of superpixels to be obtained by segmentation), and then use the SLIC algorithm to perform superpixel segmentation on the flood probability map FPM at different scales according to these scale parameters.
[0013] Step 5: Perform unsupervised flood mapping based on machine learning at different segmentation scales to generate a series of flood maps at different scales.
[0014] Step 6: Evaluate the mapping uncertainty of flood maps generated at different scales, and perform adaptive fusion of flood maps at different scales according to the principle of minimum uncertainty (i.e., at each pixel, automatically take the classification label in the flood map corresponding to the minimum mapping uncertainty as the final mapping result at that pixel), thereby generating a final reliable flood spatial distribution map.
[0015] Furthermore, step 5 includes the following steps:
[0016] Step 5.1: Perform uncertainty analysis on the flood probability map (FPM) based on the Gaussian mixture model at each different scale, and decompose the flood probability map (FPM) into three parts: determined flood area, determined non-flood area, and uncertain area. Extract the determined flood area and determined non-flood area as simulation samples for subsequent flood mapping.
[0017] Step 5.2: At each different scale, the machine learning model is trained and predicted using simulated samples (defined flood areas and defined non-flood areas) to generate a series of binarized flood maps at different scales.
[0018] Furthermore, step 6 employs an information entropy model to assess the mapping uncertainty of flood mapping results at different scales. Let the mapping uncertainty of pixel x at different segmentation scales be U... k1 U k2 ,……,U kn The classification labels under different segmentation scales are C k1 C k2 ,……,C kn Where k1, k2, ..., kn correspond to a series of scale parameters set from small to large in step 4. Based on the principle of minimum uncertainty, after adaptive fusion of flood maps at different scales, the classification label C(x) of pixel x in the final flood spatial distribution map is determined by the following formula:
[0019] C(x)=C kl (2)
[0020]
[0021] In the above formula, l is an intermediate variable and l = 1, 2, ..., n, which represents the mapping uncertainty U at different scales at pixel x. k1 U k2 ,……,U kn The index corresponding to the minimum value. C kl This represents the classification label at pixel x in the flood map corresponding to the minimum mapping uncertainty.
[0022] Compared with the prior art, the present invention has the following advantages:
[0023] (1) Although the method proposed in this invention uses a machine learning model as the core classifier, it does not require manual labeling of samples. It can realize unsupervised and automated extraction of flood-inundated areas in urban scenarios and has strong practical application value.
[0024] (2) The method proposed in this invention can adaptively integrate multi-scale mapping information in SAR images during flood mapping and can ensure that the final flood map has the lowest mapping uncertainty, thus effectively improving the accuracy and reliability of flood mapping.
[0025] (3) The flood mapping method proposed in this invention has fast calculation efficiency and strong operability. It is not only highly accurate and automated, but also easy to implement and highly scalable. Therefore, this invention has high practical value. Attached Figure Description
[0026] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation
[0027] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0028] This invention provides a method for urban flood mapping based on SAR imagery using multi-scale information fusion. The technical solution of this invention will be further described below with reference to the accompanying drawings and embodiments.
[0029] like Figure 1 As shown, the process of this embodiment of the invention includes the following steps:
[0030] Step 1: Acquire SAR data of the target area before and after the flood, and perform a series of necessary data preprocessing on the acquired SAR data, such as radiometric calibration, filtering and terrain correction.
[0031] Step 2: Considering the complexity of urban scenes and the limited number of SAR data bands, spatial and pseudo-color features are extracted from the SAR intensity images before and after the flood, such as spatial morphological features, texture features, and color saturation features obtained through pseudo-color image synthesis and color space transformation. All extracted feature maps are then overlaid with the intensity images to form feature sets before and after the flood.
[0032] Step 3: Use the log-ratio operator to calculate the difference between the feature sets before and after the flood, generating a series of difference features Δf.i (i = 1, 2, ..., N; and N represents the number of features contained in the feature set before and after the flood), let f i 1 with f i 2 If each feature corresponds to the i-th feature in the feature sets before and after the flood, then the i-th difference feature generated by the log-ratio operator can be represented as follows:
[0033]
[0034] In the above formula, log(·) represents the logarithmic function, and |·| represents the absolute value function.
[0035] Subsequently, by analyzing all the difference features Δf i The summation of (i = 1, 2, ..., N) generates a flood probability map (FPM), which reflects the probability that each pixel in the image belongs to a flood-inundated area.
[0036] Step 4: Set a series of scale parameters k1, k2, ..., kn from small to large (these scale parameters represent the expected number of superpixels to be obtained by segmentation), and then use the SLIC algorithm to perform superpixel segmentation on the flood probability map FPM at different scales according to these scale parameters.
[0037] Step 5: Perform unsupervised flood mapping based on machine learning at different segmentation scales to generate a series of flood maps at different scales.
[0038] Step 5.1: Perform uncertainty analysis on the flood probability map (FPM) based on the Gaussian mixture model at each different scale, and decompose the flood probability map (FPM) into three parts: determined flood area, determined non-flood area, and uncertain area. Extract the determined flood area and determined non-flood area as simulation samples for subsequent flood mapping.
[0039] Step 5.2: At each different scale, the machine learning model (extreme learning machine or support vector machine model) is trained and predicted using simulated samples (determined flood areas and determined non-flood areas) to generate a series of binarized flood maps at different scales.
[0040] Step 6: Use the information entropy model to evaluate the mapping uncertainty of flood maps generated at different scales, and perform adaptive fusion of flood maps at different scales according to the principle of minimum uncertainty (i.e., at each pixel, automatically take the classification label in the flood map corresponding to the minimum mapping uncertainty as the final mapping result at that pixel), thereby generating a final reliable flood spatial distribution map.
[0041] Let the mapping uncertainty of pixel x at different segmentation scales be U k1 U k2 ,……,U kn The classification labels under different segmentation scales are C k1 C k2 ,……,C kn Where k1, k2, ..., kn correspond to a series of scale parameters set from small to large in step 4. After adaptive fusion of flood maps at different scales, the classification label C(x) of pixel x in the final flood spatial distribution map is determined by the following formula according to the principle of minimum uncertainty:
[0042] C(x)=C kl (2)
[0043]
[0044] In the above formula, l is an intermediate variable and l = 1, 2, ..., n, which represents the mapping uncertainty U at different scales at pixel x. k1 U k2 ,……,U kn The index corresponding to the minimum value. C kl This represents the classification label at pixel x in the flood map corresponding to the minimum mapping uncertainty.
[0045] In practice, the above process can be automated using computer software technology.
[0046] It should be understood that any parts not described in detail in this specification belong to the prior art.
[0047] It should be understood that the above description of the preferred embodiments is quite detailed, but it should not be considered as a limitation on the scope of protection of this invention. Those skilled in the art, under the guidance of this invention, can make substitutions or modifications without departing from the scope of protection of the claims of this invention, and all such substitutions or modifications fall within the scope of protection of this invention. The scope of protection of this invention should be determined by the appended claims.
Claims
1. A method for urban flood mapping based on SAR imagery using multi-scale information fusion, characterized in that, Includes the following steps: Step 1: Obtain SAR data of the target area before and after the flood, and preprocess the obtained SAR data respectively; Step 2: Extract spatial features and pseudo-color features from the SAR data intensity images before and after the flood, and combine all the extracted feature images with the intensity images by layer stacking to form a feature set before and after the flood. Step 3: The log-ratio operator is used to calculate the difference between the feature sets before and after the flood to generate a series of difference features. The flood probability map (FPM) is generated by summing all the difference features. The flood probability map (FPM) reflects the probability that each pixel in the image belongs to the flood-inundated area. Difference characteristics The calculation method is as follows: set up and If each feature corresponds to the i-th feature in the feature sets before and after the flood, then the i-th difference feature generated by the log-ratio operator can be represented as follows: （1） In the above formula, log(·) represents the logarithmic function, |·| represents the absolute value function, i=1,2,……,N; and N represents the number of features contained in the feature set before and after the flood. Step 4: Set a series of scale parameters k1, k2, ..., kn from small to large, and then use the SLIC algorithm to perform superpixel segmentation on the flood probability map FPM at different scales based on these scale parameters. Step 5: Perform unsupervised flood mapping using machine learning models at different segmentation scales to generate a series of flood maps at different scales; Step 6: Evaluate the mapping uncertainty of flood maps generated at different scales, and adaptively fuse the flood maps at different scales according to the principle of minimum uncertainty to generate a final reliable flood spatial distribution map. In step 6, the information entropy model is used to assess the mapping uncertainty of flood mapping results at different scales; Let the mapping uncertainty of pixel x at different segmentation scales be U k1 U k2 ,……, U kn The classification labels under different segmentation scales are C k1 C k2 ,……, C kn Where k1, k2, ..., kn correspond to a series of scale parameters set from small to large in step 4. Based on the principle of minimum uncertainty, after adaptive fusion of flood maps at different scales, the classification label C(x) of pixel x in the final flood spatial distribution map is determined by the following formula: （2） （3） In the above formula, intermediate variable and =1,2,……,n, where U represents the mapping uncertainty U at different scales at pixel x. k1 U k2 ,……, U kn The index corresponding to the minimum value, C kl This represents the classification label at pixel x in the flood map corresponding to the minimum mapping uncertainty.
2. The method for urban flood mapping based on multi-scale information fusion of SAR images as described in claim 1, characterized in that: The preprocessing includes radiometric calibration, filtering, and terrain correction of the SAR data.
3. The method for urban flood mapping based on multi-scale information fusion of SAR images as described in claim 1, characterized in that: The spatial features include spatial morphological features and textural features; Pseudo-color features refer to the color saturation features obtained through color image synthesis and color space transformation.
4. The method for urban flood mapping based on multi-scale information fusion of SAR images as described in claim 1, characterized in that: The scale parameter in step 4 represents the expected number of superpixels to be obtained from the segmentation, which can be set by customization.
5. The method for urban flood mapping based on multi-scale information fusion of SAR images as described in claim 1, characterized in that: The specific implementation method of step 5 is as follows: Step 5.1: Perform uncertainty analysis on the flood probability map (FPM) based on the Gaussian mixture model at each different scale, and decompose the flood probability map (FPM) into three parts: the determined flood area, the determined non-flood area, and the uncertain area. Extract the determined flood area and the determined non-flood area as simulation samples for subsequent flood mapping. Step 5.2: At each different scale, the machine learning model is trained and predicted using simulated samples to generate a series of binarized flood maps at different scales. The simulated samples are defined flood areas and defined non-flood areas, and the generated binarized maps only contain flood areas and non-flood areas.
6. The method for urban flood mapping based on multi-scale information fusion of SAR images as described in claim 1, characterized in that: The machine learning model is either an Extreme Learning Machine or a Support Vector Machine model.