A method for detecting rapidly changing areas in dam break accidents based on optical remote sensing images
By performing image difference calculation and connectivity analysis in the LAB color space, the real-time performance and noise resistance issues of remote sensing change detection in dam break accidents were resolved, enabling rapid and accurate identification of disaster change areas and supporting emergency decision-making.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA AGRI UNIV
- Filing Date
- 2026-05-28
- Publication Date
- 2026-06-23
Smart Images
Figure CN122265872A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of remote sensing image processing technology, specifically to a method for detecting rapidly changing areas of dam break accidents based on optical remote sensing images. Background Technology
[0002] Against the backdrop of global climate change and frequent extreme weather events, sudden disasters such as dam breaks, landslides, and floods pose serious threats to people's lives and property, the ecological environment, and socio-economic development. After a disaster occurs, rapidly obtaining dynamic information about the disaster situation and accurately identifying areas of change are prerequisites for conducting emergency rescue and disaster assessment. Remote sensing technology, with its advantages of wide-area, real-time, and non-contact monitoring, has become a core tool for disaster monitoring, and remote sensing image change detection is a crucial step in extracting disaster information from remote sensing data.
[0003] Currently, the mainstream remote sensing change detection methods can be mainly divided into the following two categories:
[0004] Pixel-based algebraic operation methods: These methods directly perform algebraic operations on corresponding pixels of two temporal remote sensing images. Common methods include the difference method (including grayscale difference, band difference, etc.) and the ratio method. The core principle is to determine whether a location has changed by calculating the degree of difference in pixel values. When the difference exceeds a preset threshold, it is determined to be a changed pixel. The significant advantages of this type of method are its simple principle, low computational complexity, and fast processing speed, enabling rapid processing of large-scale image data. However, it also has significant limitations: it requires extremely high image registration accuracy; even a small registration error can lead to a large amount of false change information. Furthermore, it is susceptible to interference from atmospheric scattering, changes in illumination, and terrain shadows, resulting in poor accuracy and reliability of the detection results.
[0005] Feature-based detection methods: To overcome the shortcomings of pixel-level methods, researchers have proposed feature-based change detection methods, such as Principal Component Analysis (PCA), Change Vector Analysis (CVA), Support Vector Machine (SVM), and deep learning models. These methods first extract features from the image (such as texture features, shape features, and high-dimensional semantic features), and then identify changed regions by analyzing the differences in these features. Compared to pixel-level methods, they offer higher detection accuracy and can better handle change detection needs in complex scenes. However, these methods suffer from high computational complexity and long processing times. Deep learning models, in particular, require a large amount of labeled data for training and have high hardware computing resource requirements, making it difficult to meet the real-time demands of emergency monitoring in sudden disasters such as dam failures.
[0006] Therefore, the shortcomings of existing technologies are mainly concentrated in the following aspects:
[0007] (1) Severe lack of real-time performance: After a dam failure, the golden rescue time is short, and the speed of obtaining disaster information is extremely important. However, existing feature-based complex change detection algorithms have cumbersome operation processes and huge computational loads. Even on high-performance hardware, it is difficult to complete the processing and analysis of large-scale images in a short time, which cannot meet the real-time requirements of emergency monitoring. While some pixel-level methods are faster, their accuracy is difficult to guarantee, and they cannot provide reliable support for emergency decision-making.
[0008] (2) Poor noise resistance: The quality of optical remote sensing images is easily affected by various factors, such as lighting conditions during shooting, atmospheric scattering, shadows caused by terrain undulations, and minor errors generated during image registration. However, after sudden disasters such as dam failures, data is often collected by means of UAV aerial photography and emergency satellite rapid imaging in order to quickly obtain disaster images. Such images are easily affected by shooting environment (such as severe weather, complex terrain) and equipment performance limitations, making it difficult to distinguish between real changes and noise. Existing pixel-level methods mostly adopt simple threshold segmentation strategies, which lack adaptability to complex scenes and have weak resistance to the above-mentioned interference factors. They are prone to identifying non-disaster changes such as cloud cover areas, terrain shadows, seasonal changes in vegetation, and differences in image brightness as disaster changes, resulting in a high false alarm rate in the detection results. Although feature-level methods have improved noise resistance, they still cannot completely eliminate interference and have problems such as increased computational costs and insufficient real-time performance.
[0009] (3) Low accuracy of result screening: Existing technologies lack an effective screening mechanism and cannot accurately remove isolated noise and irrelevant minor change areas. The detection results often contain a large number of isolated noise points, minor irrelevant change patches and other non-core disaster information, which causes key change areas to be covered by redundant information. This not only increases the pressure on data storage and transmission, but also interferes with the judgment of emergency decision-makers on the core disaster situation, reduces the efficiency of emergency response, and may even hinder rescue work due to information delays. Summary of the Invention
[0010] In view of this, the present invention provides a method for detecting rapidly changing areas of dam break accidents based on optical remote sensing images, which can improve the real-time performance, accuracy and robustness of dam break accidents.
[0011] To solve the above-mentioned technical problems, the present invention is implemented as follows.
[0012] A method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery includes:
[0013] Step 1: Acquire pre-accident optical remote sensing images of the dam failure. and post-accident optical remote sensing images It is then converted to the LAB color space to obtain the first LAB image. Second LAB image ;
[0014] Step 2: Calculate the absolute difference of the three color channels in the two LAB images respectively to obtain the absolute difference map. For absolute difference plots The three color channel data of each pixel are weighted and summed to obtain a comprehensive color difference intensity value, which forms the intensity variation map. In the weighted summation, the weights of color elements A and B are greater than the weight of the brightness element L.
[0015] Step 3: Based on the change intensity map Identify areas of change following a dam failure.
[0016] Preferably, in step 2, when performing a weighted summation on the three color channel data, the weights of color elements A and B are selected as 0.4, and the weight of brightness element L is selected as 0.2.
[0017] Preferably, in step 3, based on the change intensity map Identifying areas of change following a dam break incident includes:
[0018] Step 31: Use Otsu's automatic thresholding method to analyze the intensity of change map. Perform binarization segmentation to generate an initial binary transformation map. ;
[0019] Step 32: Perform initial binary transformation on the graph. Morphological processing is performed to remove isolated noise and smooth boundaries;
[0020] Step 33: Perform connected component analysis on the morphologically processed image to identify all connected components that meet the minimum area requirement. ;
[0021] Step 34: For each connected region that meets the area requirement Calculate its intensity of change in the graph average intensity ;
[0022] Step 35: Based on average strength For all connected regions Sort in descending order and extract average intensity The highest set ratio of the connected area is used to obtain the area of change in the dam failure accident.
[0023] Preferably, in step 35, based on the average strength For all connected regions Sort in descending order and extract average intensity After identifying the highest proportion of connected regions, morphological filtering and isolated region removal are further applied to the extracted connected regions, and the region boundaries are optimized to obtain the change region of the dam failure accident.
[0024] Preferably, step 1 involves acquiring pre-dam failure optical remote sensing images. and post-accident optical remote sensing images Next, a maximum dimension scaling preprocessing is performed: if any dimension exceeds the preset maximum value, it is scaled down to the allowable range, and the scaling ratio is recorded. ;
[0025] In step 3, after identifying the area of change due to the dam failure, the scaling ratio is used... Restored to the original optical remote sensing image size.
[0026] Preferably, the method further includes:
[0027] In step 1, optical remote sensing images prior to the accident are extracted. and post-accident optical remote sensing images Geographic transformation information and projection information;
[0028] After identifying the dam failure accident change area in step 3, the outer contour of the dam failure accident change area is extracted. The extracted outer contour is converted into geographic coordinates using the geographic transformation information of the original optical remote sensing image. Then, the geographic coordinates are converted to the standard WGS84 coordinate system using the projection information. Finally, the outer contours of all dam failure accident change areas in the WGS84 coordinate system are output in GeoJSON format.
[0029] Beneficial effects:
[0030] (1) Improved robustness and accuracy of detection: Existing technologies mostly perform simple interpolation in RGB space, which is easily affected by brightness. This scheme calculates the difference in LAB space, especially assigning higher weights to the A and B components, making the detection results more sensitive to the actual changes in ground object color and effectively distinguishing between real changes and changes in illumination / shade / seasonality.
[0031] (2) Ensuring Real-Time Performance: This invention abandons feature-level methods such as deep learning, PCA, and SVM, which are highly complex, hardware-dependent, and require a large amount of training data. These methods are computationally cumbersome and time-consuming, making them unsuitable for real-time emergency response. Instead, this invention uses LAB spatial pixel-level algebraic operations, which only include channel absolute differences and weighted summation calculations. This method has a small computational load and is fast, requiring only simple addition, subtraction, and multiplication pixel by pixel without complex matrix operations or model inference, thus significantly shortening the processing time. While improving detection accuracy, it also ensures detection speed, meeting the real-time requirements of "racing against time" in emergency monitoring of sudden disasters such as dam failures.
[0032] (3) Achieving focus on key information: Traditional methods only perform post-processing based on area or perimeter. This solution innovatively uses the average change intensity of connected regions for sorting and proportional filtering, ensuring that the final output is the most indicative of the disaster situation (i.e., the most drastic change), which greatly improves the efficiency of emergency decision-making.
[0033] (4) Improved noise resistance: This invention converts the image to the LAB color space to separate luminance and chromaticity information, thereby reducing the influence of luminance noise such as light, shadow, and atmosphere. Then, it assigns higher weights to color channels A and B and lower weights to luminance channel L for weighted difference calculation, which enhances the real color changes and suppresses luminance interference. At the same time, it combines morphological operations to remove isolated noise and smooth boundaries. Finally, it filters out small and weak noise areas by screening the area of connected regions and sorting the intensity of changes, thereby effectively solving the problem of poor noise resistance.
[0034] (5) High degree of engineering and can be directly applied: This invention integrates the reading of geographic information from TIFF images and the output of the final GeoJSON format. Furthermore, it solves the problem of geographic coordinate drift caused by preprocessing scaling through scaling ratio correction, so that the results can be directly imported into the GIS emergency system without secondary processing.
[0035] This invention is particularly suitable for rapid and accurate detection of changes in disaster-affected areas after a dam break, providing crucial technical support for emergency command, disaster assessment, and rescue deployment. Its core applications cover natural resource monitoring and emergency management, and it can be integrated into various emergency response platforms to achieve automated and efficient identification and feedback of disaster-affected areas. Attached Figure Description
[0036] Figure 1 This is a flowchart of a method for detecting rapidly changing areas of dam break accidents based on optical remote sensing images, according to a preferred embodiment of the present invention.
[0037] Figure 2 This is a block diagram of the device for detecting rapidly changing areas of dam break accidents based on optical remote sensing images in a preferred embodiment of the present invention.
[0038] Figure 3 This is a block diagram of the module for extracting the changing area of a dam break accident in a preferred embodiment of the present invention. Detailed Implementation
[0039] In remote sensing image detection of dam failure accidents, changes in illumination (such as shooting time and cloud cover) and differences in sensor radiation are the main interference factors. These interferences usually manifest as global or local brightness fluctuations rather than fundamental color changes. If the RGB color space is used, brightness and color information are coupled together, and changes in illumination will directly cause synchronous fluctuations in the values of the three RGB channels, which can easily be misjudged as "change areas". However, the LAB (luminance / green / red / blue / yellow) space can separate brightness and color, filter out invalid differences caused by illumination interference, and retain only the true color and brightness changes caused by the dam failure.
[0040] Based on the above analysis, this invention provides a method for detecting rapidly changing areas in dam-break accidents based on optical remote sensing imagery. The basic idea is to calculate the absolute difference between the images before and after the dam-break accident in the LAB color space, and obtain a change intensity map by weighted summation of three channels with specified weights. Then, the changed areas of the dam-break accident are extracted based on this change intensity map. This scheme innovatively combines the anti-interference characteristics of the LAB color space to identify changed areas in dam-break accidents without complex calculations, significantly improving detection accuracy while maintaining detection speed.
[0041] This invention also proposes a strength filtering mechanism based on connected regions, further improving the real-time performance and accuracy of detection. Furthermore, this invention provides a scheme for outputting geographic information in a GeoJSON format, significantly enhancing the practicality of the detection results and facilitating direct integration into emergency response systems.
[0042] The following is combined with Figure 1 The preferred embodiment of the present invention is described in detail below. This preferred embodiment achieves rapid and accurate detection of areas affected by dam failures through a complete workflow of "geographic information image loading - preprocessing - LAB spatial difference detection - morphological postprocessing - change intensity screening - GeoJSON output".
[0043] See Figure 1 This implementation process involves processing pre-accident images. and post-accident footage This will generate the final GeoJSON file containing the changed regions.
[0044] Step 1: Geographic image loading and preprocessing.
[0045] Loading pre-dam failure images and post-accident footage Extract their geographic transformation information and projection information.
[0046] Pre-accident images and post-accident optical remote sensing images Perform maximum dimension scaling: If any dimension exceeds the preset maximum value, scale it down proportionally and record the scaling ratio. This is to ensure that the pixel coordinates are accurately converted back to the original geographic coordinates later. The scaling down can be achieved using cv2.INTER_AREA interpolation.
[0047] Step 2: LAB color space difference detection.
[0048] This step includes the following sub-steps:
[0049] Step 21: The pre-processed... and Convert to LAB color space to obtain the first LAB image. Second LAB image .
[0050] Step 22: Calculate the absolute difference map between the two LAB images. This formula represents the absolute difference data of the three color channels obtained by calculating the absolute difference of the three elements of each pixel in two LAB images. .
[0051] Step 23: Calculate the composite color difference intensity map Based on its application in dam failure accidents, the weights of the three elements are set as follows: L: 0.2, A: 0.4, B: 0.4, to calculate the overall color difference intensity.
[0052] The determination of this weight ratio is based on the dual results of the difference in the contribution of each LAB channel under the dam break scenario and experimental optimization: On the one hand, the core change of the dam break is the inundation of land features by water. The A / B color channels are sensitive to the differences between water and vegetation, soil, etc. and are less affected by light interference. Therefore, they are given high weights (0.4 each) as the main detection channel. The L brightness channel is easily affected by radiation and light noise. It is only given a low weight (0.2) to supplement the brightness difference information, forming an anti-interference strategy of "color-dominated and brightness-assisted". On the other hand, this ratio has been verified by the accuracy of multiple sets of dam break remote sensing image samples. It is an adaptive parameter that achieves the optimal balance between false detection rate and false negative rate.
[0053]
[0054] The above formula indicates that for an absolute difference plot Each pixel in the image contains three color channel data. Perform a weighted summation.
[0055] Step 24: Combine the color difference intensity map Normalized to the range of 0-255, the intensity of change is obtained. .
[0056] Step 3: Region segmentation based on LAB variation intensity.
[0057] Using Otsu's automatic thresholding method, the intensity of change map was analyzed. Perform binarization segmentation to generate an initial binary transformation map. .
[0058] Step 4: Morphological post-processing and connected region analysis.
[0059] Initial binary transformation plot Morphological processing is performed to remove isolated noise and smooth boundaries. For example, closing operations (filling small holes) and opening operations (removing small spots) can be performed sequentially, using a preset kernel of 7×7.
[0060] Connectivity analysis was performed on the morphologically processed image to identify all connected regions that met the minimum area requirement. subscript This indicates the sequence number of the connected region that meets the minimum area requirement.
[0061] Step 5: Screening key regions based on the intensity of change.
[0062] For each connected region that meets the area requirement Calculate its intensity of change in the graph average intensity That is, in the intensity change diagram Find connected components For the corresponding regional range, calculate the average intensity of the change within that region to obtain the average intensity. .
[0063] Based on average strength For all connected regions Sort in descending order.
[0064] Only retain average strength The top K% of connected regions As the final set of changed regions K is the set ratio value.
[0065] Set of changing regions Denoising is performed by removing discrete noise through morphological filtering and isolated small region removal methods, optimizing the region morphology and boundaries, and generating the final filtered transformation map M. final .
[0066] Step 6: Scale and restore, and output GeoJSON.
[0067] First, from M finalExtract the outer contour. For each outer contour, use the pixel coordinates... First, by scaling Restore to original image coordinates .
[0068] Then, using the geographic transformation information of the original imagery, Convert to geographic coordinates.
[0069] Finally, the geographic coordinates are converted to the standard WGS84 coordinate system (EPSG:4326) using projection information and the pyproj library (a Python library for geographic coordinate transformation and map projection).
[0070] Finally, all changed regions are output in GeoJSON format for easy use by downstream systems.
[0071] This concludes the process.
[0072] Based on the above method, the present invention also provides a device for detecting rapidly changing areas of dam break accidents based on optical remote sensing images, such as... Figure 2 As shown, it includes: an image loading module, a LAB color space conversion module, a change intensity acquisition module, and a dam failure accident change area extraction module. Preferably, the device also includes a dimension scaling module and a format extraction and conversion module.
[0073] The image loading module is used to load optical remote sensing images of the dam failure prior to the accident. and post-accident optical remote sensing images The dimension scaling module is invoked to perform maximum dimension scaling preprocessing. If any dimension exceeds the preset maximum value, it is scaled down to the allowed range, and the scaling ratio is recorded. This module also calls the format extraction and conversion module to extract geographic transformation and projection information from optical remote sensing images.
[0074] The LAB color space conversion module is used to convert the pre- and post-accident remote sensing images processed by the image loading module to the LAB color space to obtain the first LAB image. Second LAB image .
[0075] The variation intensity acquisition module is used to calculate the absolute difference of the three color channels in the two LAB images respectively, and obtain the absolute difference map. For absolute difference plots The three color channel data of each pixel are weighted and summed to obtain a comprehensive color difference intensity value, which in turn yields a change intensity map. In the weighted summation, the weights of color elements A and B are greater than the weight of the brightness element L.
[0076] The dam failure accident change area extraction module is used to extract changes based on the change intensity map. Identify areas of change following a dam failure. For example... Figure 3 As shown, the module for extracting the changing regions in this dam failure accident specifically includes an adaptive binarization segmentation unit, a morphological processing unit, a connected region analysis unit, and a changing region identification unit.
[0077] Adaptive binarization segmentation units are used to perform intensity map segmentation using the Otsu automatic thresholding method. Perform binarization segmentation to generate an initial binary transformation map. .
[0078] Morphological processing unit, used for initial binary transformation map Perform morphological operations to remove isolated noise and smooth boundaries.
[0079] The connected component analysis unit is used to perform connected component analysis on morphologically processed images, identifying all connected components that meet the minimum area requirement. .
[0080] The change region identification unit is used to identify each connected region that meets the area requirement. Calculate its intensity of change in the graph average intensity According to average strength For all connected regions Sort in descending order, retaining average intensity The highest set ratio area yields the change area of dam failure accidents.
[0081] After identifying the dam failure accident change area, the dam failure accident change area extraction module calls the dimension scaling module again, based on the stored scaling ratio. The changed area of the dam failure accident is restored to the original image size. Then, the format extraction and conversion module is called again to perform coordinate transformation operations, including: converting the outer contour of the changed area of the dam failure accident into geographic coordinates using the geographic transformation information of the original optical remote sensing image; and converting the geographic coordinates into the standard WGS84 coordinate system using the projection information, and outputting it in GeoJSON format.
[0082] The preferred embodiment of this invention solves the problems of insufficient real-time performance, poor noise resistance, and imprecise result selection in the prior art. Specifically, it includes the following beneficial effects:
[0083] (1) The present invention combines the robustness of color space differences and intelligent screening based on the intensity of change, ensuring detection efficiency and accuracy in rapid emergency scenarios.
[0084] (2) The LAB color space is used for difference calculation. Compared with the traditional RGB space, it can better separate the brightness (L) and chromaticity (A,B) information, and improve the noise resistance to different lighting conditions and atmospheric effects.
[0085] (3) Introducing weighted gray difference calculation (L:0.2,A:0.4,B:0.4) can more effectively highlight the difference in color information (A,B), which is especially important for detecting changes in land cover type (such as land / vegetation changing into water / soil).
[0086] (4) A post-processing screening mechanism based on change intensity is proposed: by sorting the average change intensity of connected regions, only the top K% regions with the highest intensity are retained, which realizes the focus from all changes to key changes and significantly reduces false alarm rate and noise interference.
[0087] (5) The solution fully realizes the closed loop from geographic image reading, coordinate correction (overcoming the coordinate offset caused by scaling) to the final GeoJSON geographic data output, and has high engineering practicality.
[0088] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.
Claims
1. A method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery, characterized in that, include: Step 1: Acquire pre-accident optical remote sensing images of the dam failure. and post-accident optical remote sensing images It is then converted to the LAB color space to obtain the first LAB image. Second LAB image ; Step 2: Calculate the absolute difference of the three color channels in the two LAB images respectively to obtain the absolute difference map. For absolute difference plots The three color channel data of each pixel are weighted and summed to obtain a comprehensive color difference intensity value, which forms the intensity variation map. In the weighted summation, the weights of color elements A and B are greater than the weight of the brightness element L. Step 3: Based on the change intensity map Identify areas of change following a dam failure.
2. The method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery as described in claim 1, characterized in that, In step 2, when performing a weighted summation on the three color channel data, the weights of color elements A and B are set to 0.4, and the weight of brightness element L is set to 0.
2.
3. The method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery as described in claim 1, characterized in that, In step 3, based on the change intensity map Identifying areas of change following a dam break incident includes: Step 31: Use Otsu's automatic thresholding method to analyze the intensity of change map. Perform binarization segmentation to generate an initial binary transformation map. ; Step 32: Perform initial binary transformation on the graph. Morphological processing is performed to remove isolated noise and smooth boundaries; Step 33: Perform connected component analysis on the morphologically processed image to identify all connected components that meet the minimum area requirement. ; Step 34: For each connected region that meets the area requirement Calculate its intensity of change in the graph average intensity ; Step 35: Based on average strength For all connected regions Sort in descending order and extract average intensity The highest set ratio of the connected area is used to obtain the area of change in the dam failure accident.
4. The method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery as described in claim 3, characterized in that, In step 35, based on the average strength For all connected regions Sort in descending order and extract average intensity After identifying the highest proportion of connected regions, morphological filtering and isolated region removal are further applied to the extracted connected regions, and the region boundaries are optimized to obtain the change region of the dam failure accident.
5. The method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery as described in claim 1, characterized in that, Step 1 involves acquiring pre-dam failure optical remote sensing images. and post-accident optical remote sensing images Next, a maximum dimension scaling preprocessing is performed: if any dimension exceeds the preset maximum value, it is scaled down to the allowable range, and the scaling ratio is recorded. ; In step 3, after identifying the area of change due to the dam failure, the scaling ratio is used... Restored to the original optical remote sensing image size.
6. The method for detecting rapidly changing areas of dam break accidents based on optical remote sensing imagery as described in claim 1, characterized in that, The method further includes: In step 1, optical remote sensing images prior to the accident are extracted. and post-accident optical remote sensing images Geographic transformation information and projection information; After identifying the dam failure accident change area in step 3, the outer contour of the dam failure accident change area is extracted. The extracted outer contour is converted into geographic coordinates using the geographic transformation information of the original optical remote sensing image. Then, the geographic coordinates are converted to the standard WGS84 coordinate system using the projection information. Finally, the outer contours of all dam failure accident change areas in the WGS84 coordinate system are output in GeoJSON format.