Optical remote sensing image feature landslide information extraction method

Abstract

The invention relates to the technical field of earthquakes, in particular to an optical remote sensing image feature landslide information extraction method. A double-time-phase NDVI time sequence analysis strategy is adopted, and landslide area recognition is achieved by constructing a vegetation index difference chart before and after an earthquake. A landslide mass area before an earthquake presents a high NDVI value due to complete vegetation coverage, and a vegetation index of the area in an image after a disaster is significantly attenuated due to surface disturbance, so that an obvious change response is formed. If the normalized vegetation index does not change, a non-landslide area can be determined, and the greater the change of the normalized vegetation index is, the greater the possibility of landslide occurrence is, and a landslide preselection area is determined; then Otsu threshold segmentation is carried out by combining cloud layer features and water body features, cloud layer and water body change parts are effectively eliminated, object-oriented geometric shape rule fine recognition is carried out on a preselected area, a DEM is adopted to calculate a slope value to constrain low-lying or flat earth surface changes, and efficient and accurate landslide information extraction is achieved.

Description

Technical Field

[0001] The present invention relates to the field of earthquake technology, and in particular to a method for extracting landslide information from optical remote sensing images. Background Art

[0002] Optical remote sensing data has the characteristics of multi-scale, multi-spectral and multi-temporal. With the increasing abundance of high-resolution satellite image data, the information extraction technology of landslide disasters has also been continuously advanced. However, the current post-disaster landslide emergency investigation mainly relies on visual interpretation and field surveys, which is not only time-consuming and labor-intensive, but also difficult to meet the needs of emergency rescue. Existing single-temporal landslide remote sensing information extraction methods (whether pixel-oriented or object-oriented) have problems of over-identification and misidentification. This is because in complex terrain areas, the spectral response characteristics of man-made structures, exposed bedrock and transportation networks are highly similar to those of landslide bodies, resulting in classification models that rely solely on spectral information being easily confused and having low overall accuracy. Summary of the Invention

[0003] In order to improve the timeliness of remote sensing landslide emergency response and effectively reduce the rate of false positives and missed judgments, the present invention provides a method for extracting landslide information from optical remote sensing image features. It adopts a dual-phase NDVI time series analysis strategy and realizes landslide area identification by constructing a vegetation index difference map before and after the earthquake. Before the earthquake, the landslide area showed a high NDVI value due to complete vegetation coverage. However, in the post-disaster image, the vegetation index of this area was significantly attenuated due to surface disturbance, forming a significant change response. If the normalized vegetation index does not change, it can be clearly identified as a non-landslide area. The greater the change in the normalized vegetation index, the greater the possibility of a landslide, and the landslide pre-selected area is determined; then, Otsu threshold segmentation is performed in combination with cloud characteristics and water body characteristics to effectively exclude the changed parts of the cloud and water bodies, and the pre-selected area is finely identified based on object-oriented geometric shape rules. The slope value calculated by DEM is used to constrain the changes in low-lying or flat surfaces, thereby achieving efficient and accurate landslide information extraction.

[0004] The present invention provides a method for extracting landslide information based on optical remote sensing image features, which comprises:

[0005] Obtain optical images of earthquake areas before and after landslides;

[0006] Preprocessing of optical images before and after landslide;

[0007] Based on the pre-processed optical images before and after the landslide, the difference in the Normalized Difference Vegetation Index (NDVI) between the two temporal images before and after the landslide was calculated.

[0008] Based on the normalized vegetation index difference, dividing the earthquake area into a changed area and an unchanged area according to a preset threshold;

[0009] The cloud mask map was established by combining the shortwave infrared (SWIR) and NDVI thresholds to remove cloud interference areas in the change area, and the normalized difference water index (NDWI) was used to remove water interference areas in the change area.

[0010] For the change area after removing the cloud interference area and water interference area, the Otsu global threshold method is used to perform binary segmentation on the NDVI difference map to extract the landslide candidate area;

[0011] Based on the slope constraint generated by the pre-acquired digital elevation model (DEM), the landslide area is finally extracted from the landslide candidate area.

[0012] Preferably, the preprocessing includes radiation correction, atmospheric correction, geometric correction and image registration.

[0013] Preferably, the normalized difference vegetation index NDVI difference is used to quantify the change in vegetation coverage in the same area in two temporal remote sensing images. The calculation formula for the normalized difference vegetation index NDVI between the two temporal images before and after the landslide is:

[0014] ΔNDVI=NDVI after -NDVI before

[0015] Among them, NDVI before and NDVI after are the normalized vegetation indices of the images before and after the landslide, respectively.

[0016] Preferably, the area where the absolute value of the normalized vegetation index difference exceeds a preset threshold is divided into a change area, and the area where the absolute value does not exceed the preset threshold is divided into a non-change area.

[0017] Preferably, the cloud mask is established by combining the shortwave infrared (SWIR) and NDVI thresholds to remove cloud interference areas in the change area, including:

[0018] Pixels with NDVI < 0.8 were set as potential cloud pixels, and pixels with B11 (1610 nm) > 0.2 were set as marker cloud pixels;

[0019] Mark the pixels that meet any of the above settings as clouds, assign all pixels marked as clouds a value of 1, and assign the remaining pixels a value of 0 to generate the final cloud mask map;

[0020] According to the cloud mask image, the cloud interference area in the change area is determined to be removed.

[0021] Preferably, using the Normalized Difference Water Index (NDWI) to eliminate water interference areas in the change zone includes:

[0022] Pixels with NDWI > -0.15 are marked as water pixels to generate a water mask map;

[0023] According to the water mask map, the water interference area in the change area is determined to be removed.

[0024] As a preference, the Otsu global threshold method is used to perform binary segmentation on the NDVI difference map to extract landslide candidate areas, including:

[0025] Pixels with ΔNDVI values ​​between -0.05 and 0.05 were marked as 0, indicating non-landslide areas where vegetation did not change significantly;

[0026] Pixels with ΔNDVI values ​​outside this range are marked as 1, and areas with significant changes in NDVI are extracted as landslide candidate areas.

[0027] Preferably, the step of finally extracting the landslide area from the landslide candidate area based on the slope constraint generated by the pre-acquired digital elevation model (DEM) comprises:

[0028] Calculating a slope distribution map of a landslide candidate area using the digital elevation model;

[0029] The areas with slopes less than a preset threshold are removed from the landslide candidate areas to finally extract the landslide area.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The method performs radiometric calibration, geometric correction, and registration on the collected remote sensing images. The difference in the Normalized Difference Vegetation Index (NDVI) between the images before and after the landslide is calculated to distinguish between changed and unchanged areas. A cloud mask is established by combining shortwave infrared and NDVI thresholds, and water interference is removed using the Normalized Difference Water Index (NDWI). After removing the influence of clouds and water, the NDVI difference map is subjected to binary segmentation using the Otsu global threshold method to preliminarily extract candidate landslide areas. Finally, the candidate areas are further screened and optimized based on the slope constraints generated by the digital elevation model (DEM), completing the precise extraction and verification of the landslide area. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The pre-processed images before and after the landslide in the embodiment;

[0033] Figure 2 This is an extraction flow chart of a method for extracting landslide information from optical remote sensing images in an embodiment;

[0034] Figure 3 It is a catalog of visual interpretation of landslide in the embodiment;

[0035] Figure 4 NDVI calculation result diagram before and after the earthquake in the embodiment;

[0036] Figure 5 This is the NDVI difference map before and after the landslide in the embodiment;

[0037] Figure 6 is the NDVI difference in the embodiment Figure 2 Valued threshold segmentation result map;

[0038] Figure 7 This is a water body threshold segmentation extraction image in the embodiment;

[0039] Figure 8 The DEM and slope distribution map of the experimental area in the embodiment;

[0040] Figure 9 1 is a map of landslide and non-landslide areas extracted from the change area in the embodiment. DETAILED DESCRIPTION

[0041] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more. In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0042] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0043] Optical remote sensing technology obtains surface reflection information through visible light, near-infrared and other electromagnetic spectrum segments. It has the advantages of spatial continuous coverage, multi-spectral resolution, and high temporal frequency. It has become the core data source for dynamic monitoring of the earth's environment. In the field of optical data, the Sentinel-2 image acquisition channel is open, and the data can be downloaded for free directly from the European Space Agency (ESA) data service website (https: / / scihub.copernicus.eu). Compared with the Landsat series of images, Sentinel-2 has a higher spatial resolution; and compared with the Gaofen series of satellites, its spectral advantage is more obvious. Therefore, this embodiment mainly uses the Sentinel-2 high-resolution multispectral imaging satellite (MSI). Sentinel-2 is an earth observation mission under the ESA Copernicus program. It consists of two satellites, A and B. Its orbit adopts a sun-synchronous orbit mode with an orbital phase difference of 180°, an altitude of 786km, and an inclination of 98.5°. The specific parameters of the satellite are listed in Table 1:

[0044] Table 1 Sentinel-2 Multispectral Imager (MSI) Mode Product Information

[0045]

[0046] Both Sentinel-2 satellites carry multispectral imagers covering 13 spectral bands. The sensor band configuration and related parameters are shown in Table 2. Spatial resolutions range from 10 m to 20 m, and from visible to near-infrared to shortwave infrared, respectively, with a swath width of 290 km and a maximum spatial resolution of 10 m. The two satellites complement each other with a revisit period of five days, enabling short-term, large-scale, and high-resolution Earth observation. This medium-period temporal resolution can reflect weekly spatial and temporal changes, while its spatial resolution also meets the needs of geological disaster emergency response. Notably, Sentinel-2 is the only high-resolution multispectral satellite with three red-edge bands, which is of great significance for green vegetation monitoring.

[0047] Table 2 Sentinel-2 sensor band parameter information

[0048]

[0049] Both Sentinel-2 satellites carry multispectral imagers covering 13 spectral bands. The sensor band configuration and related parameters are shown in Table 2. Spatial resolutions range from 10 m to 20 m, and from visible to near-infrared to shortwave infrared, respectively, with a swath width of 290 km and a maximum spatial resolution of 10 m. The two satellites complement each other with a revisit period of five days, enabling short-term, large-scale, and high-resolution Earth observation. This medium-period temporal resolution can reflect weekly spatial and temporal changes, while its spatial resolution also meets the needs of geological disaster emergency response. Notably, Sentinel-2 is the only high-resolution multispectral satellite with three red-edge bands, which is of great significance for green vegetation monitoring.

[0050] This example selects Sentinel-2 image data before the earthquake (July 8, 2022) and after the earthquake (October 1, 2022) (as shown in Table 2) and preprocesses them. The preprocessing steps include image mosaicking, cropping, radiometric calibration, atmospheric correction, orthorectification, geometric correction, pixel resampling, and band fusion. At the same time, the post-earthquake remote sensing image is used as the reference image to align the pre-earthquake image to minimize errors, and finally crop the pre-set study area (such as Figure 1 (As shown). Data from 12 bands, including near-infrared and shortwave infrared, were resampled to a spatial resolution of 10 meters. These data, with their high spatial resolution, short revisit intervals, rich spectral information, and open-source access, can be combined with NDVI temporal variation information to identify landslides, making them promising for application in disaster monitoring.

[0051] Table 3 Data information of landslide events in the experimental area

[0052]

[0053] NDVI (Normalized Difference Vegetation Index), also known as the Normalized Difference Vegetation Index, can separate vegetation from water and soil, and can objectively reflect changes in vegetation coverage. It is widely considered to be the core characterization parameter reflecting vegetation growth and its coverage density. NDVI data is closely related to many parameters of vegetation, which is conducive to quantifying the amount of vegetation and is an important indicator for monitoring changes in ground vegetation. The Normalized Difference Vegetation Index (NDVI) reflects the state of vegetation by quantifying the difference in the reflection of vegetation to red light (665nm, B4 band) and near-infrared (842nm, B8 band). Near-infrared indicates strong reflection of vegetation, and red light indicates absorption of vegetation. The calculation formula is:

[0054] NDVI=(B8-B4) / (B8+B4) (1)

[0055] The data includes reflectance values ​​from the NIR Sentinel-2 near-infrared band (B8, 842nm, 10m resolution) and the Red band (B4, 665nm, 10m resolution). Bottom-of-Atmosphere (L2A) reflectance data, which has been corrected for atmospheric interference, was used to eliminate atmospheric interference. Statistics show that the Land Surface Vegetation Index (NDVI) ranges from –1 to 1. NDVI values ​​between –1 and 0 typically correspond to highly reflective areas such as clouds or water bodies. When NDVI is close to 0, it indicates a lack of vegetation cover, often consisting of bare soil or rock. When NDVI is greater than 0, it reflects varying degrees of vegetation cover, with higher values ​​indicating more lush greenery.

[0056] This embodiment uses a change detection method based on multi-temporal remote sensing images to identify landslides, and achieves target extraction by analyzing the differences in surface features before and after the landslide. The specific process is as follows:

[0057] Firstly, the high-resolution optical images before and after the Luding earthquake landslide were preprocessed, and the difference in the Normalized Difference Vegetation Index (NDVI) between the two periods was calculated.

[0058] Because the NDVI value of the vegetation-covered area was higher before the landslide occurred, and the NDVI decreased significantly after the landslide destroyed the vegetation.

[0059] By setting a threshold, areas with a sharp drop in NDVI are extracted as candidate landslide areas, effectively excluding areas with unchanged vegetation coverage (stable high NDVI) and interfering objects such as buildings and bare land (stable low NDVI).

[0060] Then, cloud interference was removed from areas where NDVI changes occurred. Clouds have high reflectivity in the blue and shortwave infrared (SWIR) bands, but low reflectivity in the red and near-infrared (NIR) bands. High cloud reflectivity in the visible light band can lead to artificially high NDVI values ​​(typically greater than 0.8), generating false positive change signals in change detection. To address cloud interference, a cloud mask was constructed by combining SWIR and NDVI thresholds (NDVI > 0.8) to remove cloud-covered areas. Sentinel-2 has multiple SWIR bands where clouds have high reflectivity, such as B11 (1610 nm), which can be used to identify clouds. Based on the spectral characteristics of the study area, a SWIR threshold of B11 > 0.2 was set. The cloud mask was then constructed by combining the NDVI and SWIR thresholds. Since cloud NDVI often appears negative or close to zero, pixels with an NDVI < 0.8 were identified as potential cloud pixels. High reflectance at B11 (1610nm) distinguishes clouds from vegetation; setting B11 > 0.2 marks cloud pixels. For each pixel that meets either of the two conditions above, it is marked as a cloud. All pixels marked as cloud are assigned a value of 1, and all other pixels are assigned a value of 0 to generate the final cloud mask.

[0061] The presence of water bodies often interferes with landslide detection. The strong absorption characteristics of water bodies in the NIR band make their NDVI values ​​usually less than 0 or close to 0, which can be easily confused with vegetation decline, leading to misjudgment or missed detection. By analyzing the spectral characteristics, water body identification and extraction are performed based on the Normalized Difference Water Index (NDWI). The threshold is set to 0.2 (i.e., NDWI>0.2) to extract the water body area and mask it to eliminate the impact of water bodies on landslide detection. The NDVI and NDWI indices are widely used in many vegetation and water body extraction studies. They can well enhance the information of vegetation and water bodies, suppress non-landslide interference signals, and achieve accurate removal of water body misjudgment pixels in vegetation surface disturbance areas.

[0062] Visual interpretation was combined with the selection of pixel combinations that matched the actual landslide distribution to delineate the landslide preselection area. To address residual errors in misidentification of roads, farmland, and rivers within the preselected area, the slopes were calculated using a DEM elevation model to constrain the extracted landslide body, further improving the accuracy of landslide boundary extraction and ultimately obtaining highly reliable landslide distribution results.

[0063] Further, combined with Figure 2This embodiment uses a systematic remote sensing image processing process to conduct landslide extraction research. The specific implementation process includes: first, pre-processing the collected remote sensing images through radiometric calibration, geometric correction, and registration; calculating the difference in the Normalized Difference Vegetation Index (NDVI) between the images before and after the landslide to distinguish between changed and unchanged areas; establishing a cloud mask by combining shortwave infrared and NDVI thresholds, and using the Normalized Difference Water Index (NDWI) to eliminate water interference; after removing the influence of clouds and water bodies, using the Otsu global threshold method to perform binary segmentation on the NDVI difference map to preliminarily extract landslide candidate areas; finally, combining the slope constraints generated by the digital elevation model (DEM) to further screen and optimize the candidate areas, completing the accurate extraction and verification of the landslide area.

[0064] Because the Luding earthquake landslide extraction study area is located in a high mountain valley, the cloud coverage of multi-view Sentinel-2 imagery often exceeds 50%, resulting in insufficient effective data rate and reduced image quality, which in turn affects subsequent surface change analysis. To reduce cloud interference on the results, this example ultimately selected two Sentinel-2 multispectral images from July 8, 2022 (before the earthquake) and October 1, 2022 (after the earthquake). These two images have relatively low cloud coverage, which can effectively improve the reliability and accuracy of change detection.

[0065] This example analyzes Sentinel-2 post-earthquake images and visually interprets them to construct a highly reliable landslide verification sample set. Human-computer interactive interpretation is used, combined with pre-landslide historical images and terrain features for cross-validation, and the following is finally marked: Figure 3 The 4029 true value samples with clear landslide boundaries are shown. The validation samples contain 310,439 valid pixels, corresponding to a total landslide area of ​​27.07 km. 2 The largest single landslide area reached 0.36 km 2 .

[0066] The two acquired Sentinel-2 multispectral images were preprocessed. First, they were stitched and cropped under the same projection. Radiometric calibration, atmospheric correction based on an atmospheric model, and image orthorectification and geometric correction were then performed sequentially to eliminate radiometric and geometric errors caused by the sensor and terrain. Subsequently, each image band was resampled and fused according to research requirements to ensure consistency in spatial resolution and spectral information. Finally, using the post-earthquake image as a reference, the pre-earthquake image was precisely aligned through image registration to minimize positioning errors caused by phase differences and cropped to the pre-defined landslide study area.

[0067] Multi-temporal NDVI data can be used to analyze vegetation change trends. Due to the characteristics of near-infrared (strong reflection from vegetation) and red light (absorption from vegetation), high values ​​of the vegetation index indicate dense tree canopies, while low or negative values ​​indicate bare land, buildings, and water bodies. The NDVI before and after the earthquake is calculated using formula (1), where NIR is the B8 band and R is the B4 band. The calculation results are as follows: Figure 4 As shown, the NDVI reflects vegetation cover by quantifying the difference in spectral response to red light (665 nm, B4 band) and near-infrared light (842 nm, B8 band). Healthy vegetation exhibits low reflectance in the red band due to chlorophyll absorption, while it exhibits high reflectance in the near-infrared band due to multiple reflections from cellular structures. The NDVI scale ranges from -1 to 1. Colors closer to red indicate NDVI values ​​closer to -1, while colors closer to green indicate NDVI values ​​closer to 1. Other colors fall in between. Negative values ​​typically correspond to water, bare soil, or clouds (strong near-infrared absorption), values ​​close to 0 indicate bare soil or rock, and positive values ​​(0.3 to 0.8) indicate healthy vegetation (chlorophyll absorbs red light and cellular structures reflect near-infrared light).

[0068] The Normalized Difference Vegetation Index (ΔNDVI) is used to quantify the change in vegetation cover in two temporal remote sensing images of the same area. It is generated by calculating the difference in NDVI values ​​between two temporal remote sensing images. The mathematical expression is:

[0069] ΔNDVI=NDVI after -NDVI before (2)

[0070] NDVI before and NDVI after The NDVI difference map of the two periods (such as Figure 5 When the NDVI values ​​of the two periods are similar or the difference of ΔNDVI is close to zero, the vegetation condition in the experimental area has not changed significantly and can be excluded as a non-landslide area; while areas with extremely positive or negative ΔNDVI values ​​correspond to drastic increases or decreases in vegetation coverage, which are likely to be landslide areas, such as the red or blue areas in the figure.

[0071] The NDVI difference map is subjected to binary segmentation using the Otsu global threshold method, as shown in Figure 5 As shown in the figure, binarization processing is performed based on this threshold, and pixels with ΔNDVI values ​​between -0.05 and 0.05 are marked as 0 (black), indicating non-landslide areas where vegetation has not changed significantly; pixels with ΔNDVI values ​​outside this range are marked as 1 (white), corresponding to areas where NDVI has changed significantly, thereby distinguishing NDVI change areas from non-changing ground objects.

[0072] Normalized water index is usually used to extract the distribution of water bodies in landslide areas. NDWI is calculated by the reflectivity of the near-infrared band (Band8, central wavelength 842nm) and the green band (Band3, central wavelength 560nm). Its value range is [-1, 1]. Water pixels usually have higher NDWI values, while non-water pixels have lower NDWI values. In order to automatically segment water bodies, this embodiment uses the Otsu global threshold method to calculate the optimal dividing point and sets the threshold to -0.15: when NDWI>-0.15, it is determined to be a water body area. When the NDWI of the local feature is ≤-0.15, the feature type is non-water body. Finally, the water body mask is compared with the Figure 5 The final water distribution in the landslide area is obtained by combining the change area extracted by NDVI difference (such as Figure 7 shown).

[0073] Clouds have high reflectivity in the visible and shortwave infrared bands, which can produce false positive signals in change detection based on vegetation index. In response to the interference of clouds in change detection, this embodiment uses the Otsu global threshold method of NDVI and SWIR bands to construct a cloud mask. First, the NDVI of each pixel is calculated. When the NDVI value is lower than 0.8, it is marked as a potential cloud pixel; at the same time, the B11 band (1610nm) of Sentinel-2 is extracted, and pixels with a reflectivity greater than 0.2 are also regarded as clouds. Then, the Otsu algorithm is applied to the NDVI and B11 images respectively to automatically determine the optimal threshold, and pixels that meet any condition are assigned a value of 1 (cloud), and the remaining pixels are assigned a value of 0 (non-cloud), generating a binary cloud mask. Finally, the cloud mask is compared with the Figure 5 The ΔNDVI change areas shown in the figure are combined to eliminate the influence of cloud cover in the landslide area and ensure the accuracy of subsequent change detection.

[0074] Figure 8 (a) is from the ALOS-12.5m resolution digital elevation model (DEM), which represents the elevation information of the earth's surface in digital form. Figure 8 (b) is the slope distribution calculated directly using DEM data. Slope is a key parameter in terrain analysis and is used to measure the steepness of surface units. It is usually expressed as the angle between the slope surface and the horizontal plane. The slope tangent (such as Formula 3) is the ratio of the vertical height change (Δh) of a point on the terrain surface to its horizontal distance (Δd).

[0075]

[0076] Combining the three difference maps with the slope distribution map reveals that landslides are often located in the upper and middle portions of slopes, revealing a correlation between topographic parameters and the spatial distribution of landslides. Furthermore, slope stability is significantly enhanced when the slope is below 15°. By calculating slope values ​​based on DEM data to apply topographic constraints, areas with slopes less than 15° are eliminated, effectively filtering out false change signals in plain areas and optimizing the identification of landslide areas.

[0077] This example uses multi-temporal remote sensing images and digital elevation model (DEM) data to sequentially perform NDVI change detection, NDWI water masking, cloud removal, and DEM terrain constraints. It also uses multi-condition threshold segmentation and mask exclusion methods to comprehensively eliminate water fluctuations and cloud interference. Figure 9 (a) As shown. Finally, the landslide area is accurately extracted, as shown in Figure 9 As shown in (b), the white area represents the identified landslide information.

[0078] This embodiment evaluates the change detection results through subjective effects and objective quantitative indicators. Table 4 uses confusion matrix and overall accuracy quantitative indicator system to evaluate the detection performance of the NDVI change difference image construction method. A true positive example is when the model correctly predicts the actual landslide area as a landslide, a false positive example is when the model mistakenly predicts the non-landslide area as a landslide, a false negative example is when the model mistakenly predicts the landslide area as a non-landslide, and a true negative example model correctly predicts the non-landslide area as a non-landslide. Overall Accuracy (OA) refers to the proportion of correctly classified pixels. It reflects the proportion of all evaluation samples in landslide and non-landslide areas that are correctly identified, and is used to comprehensively evaluate the overall classification performance of the model. The higher the overall accuracy, the better the overall classification performance of the NDVI difference method.

[0079] Table 4 Quantitative analysis indicators of landslide extraction results using different methods

[0080]

[0081] In summary, landslide identification based on multi-temporal NDVI change detection is achieved by comparing changes in vegetation cover before and after the landslide. Before the landslide occurred, the NDVI in the vegetated area was high (healthy vegetation); after the landslide destroyed the vegetation, the exposed surface caused the NDVI to drop significantly. By setting a threshold to extract areas with large NDVI differences before and after the landslide, the vegetation-damaged area can be quickly identified as a landslide candidate area, and non-landslide features with stable NDVI (such as unchanged vegetation, bare soil, buildings, etc.) can be excluded. This method is highly timely and applicable to most areas covered by Sentinel-2. The preprocessing process (atmospheric correction, cloud mask) ensures the reliability of the NDVI data, and combined with the high-resolution (10-meter) characteristics, it can finely monitor surface changes. This improves the accuracy of landslide extraction and provides efficient support for disaster emergency assessment.

[0082] The method mainly includes the following three parts: (1) Multi-temporal data preprocessing stage, in which radiometric correction, geometric registration and atmospheric correction are performed on the acquired remote sensing images to ensure the temporal and spatial comparability of the image data; (2) Dynamic feature extraction stage, by constructing a dual-temporal NDVI difference image, the Otsu optimal threshold segmentation algorithm is used to identify the surface cover change area, and at the same time, a composite mask is established by combining an improved cloud detection algorithm with the normalized difference water index (NDWI) to effectively eliminate the spectral interference between cloud cover and water bodies; (3) Change area refinement stage, the terrain slope constraint condition is constructed based on the digital elevation model (DEM) data, and the pseudo-change information caused by the terrain undulation is eliminated through spatial morphological constraints, and finally a high-confidence surface change landslide extraction map is established. This technical process significantly improves the geometric accuracy of the change detection results and the reliability of land type identification through the organic combination of multi-source data collaborative analysis and multi-level terrain slope spatial constraint mechanism.

[0083] Earthquake-induced landslides can destroy existing vegetation, resulting in images of landslide areas appearing as light-colored, bare soil without vegetation. Accordingly, the NDVI in these areas is high before the landslide, but decreases significantly afterward. Therefore, the greater the NDVI change, the greater the probability of a landslide. By setting a reasonable NDVI difference threshold, we can exclude vegetation areas without landslides while also filtering out buildings and bare land with no significant NDVI change, thereby accurately locating areas of new surface changes.

[0084] The exposed soil surface of collapsed landslides often exhibits grayscale and spectral characteristics in remote sensing images that are very similar to other exposed soils (such as river floodplains, cultivated slopes, or dirt roads). This makes it difficult to distinguish between coseismic landslides and bare soil using single-phase images. Our proposed multi-temporal NDVI difference change detection method fully utilizes pre-landslide imagery and slope data to remove false information from flat areas and cleverly identify these areas as non-landslide areas.

[0085] The experimental results show that the slopes on both sides of the main rivers such as Dadu River and Moxi River and their tributaries in the Luding earthquake landslide experimental area are the main areas where earthquake-triggered collapse and landslides occur. The landslide results finally extracted by the multi-phase NDVI difference change detection method in this embodiment are as follows: Figure 9 As shown in (b), considering the developmental characteristics of earthquake collapse and landslides, vegetation destruction is related to slope gradient. Landslides caused by vegetation destruction are primarily concentrated in the upper and middle portions of slopes with gradients ranging from 15° to 50°. High-lying landslides are densely distributed, and the convergence of multiple landslide zones forms a planar expansion pattern, dominated by small and medium-sized, shallow rock slides. The highest density of landslide points is found in Moxi Town, the Hailuogou area, and along the Dadu River in Detuo Town.

[0086] To improve the response speed and extraction accuracy of remote sensing landslide monitoring, this paper, based on a review of existing landslide identification methods, proposes an automated identification process based on change detection in multi-temporal remote sensing imagery and object-oriented geometric rules. This process first locates potential landslide areas by using significant differences in NDVI values ​​before and after a landslide. It then reduces false detections by eliminating irrelevant features with similar spectral characteristics to those of landslides. Subsequently, object-oriented rule-based judgment is applied to the initial screening results based on the geometric morphological characteristics of the landslide bodies to further optimize boundary and morphological extraction. This method ensures efficient detection of landslide areas while significantly improving the accuracy and reliability of the extraction results.

[0087] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for extracting landslide information based on optical remote sensing image features, characterized in that: include: Obtain optical images of earthquake areas before and after landslides; Preprocessing of optical images before and after landslide; Based on the pre-processed optical images before and after the landslide, the difference in the Normalized Difference Vegetation Index (NDVI) between the two temporal images before and after the landslide was calculated. Based on the normalized vegetation index difference, dividing the earthquake area into a changed area and an unchanged area according to a preset threshold; The cloud mask map was established by combining the shortwave infrared (SWIR) and NDVI thresholds to remove cloud interference areas in the change area, and the normalized difference water index (NDWI) was used to remove water interference areas in the change area. For the change area after removing the cloud interference area and water interference area, the Otsu global threshold method is used to perform binary segmentation on the NDVI difference map to extract the landslide candidate area; Based on the slope constraint generated by the pre-acquired digital elevation model (DEM), the landslide area is finally extracted from the landslide candidate area.

2. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The preprocessing includes radiation correction, atmospheric correction, geometric correction and image registration.

3. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The Normalized Difference Vegetation Index (NDVI) reflects vegetation status by quantifying the difference in vegetation reflection to red light (665nm, B4 band) and near-infrared light (842nm, B8 band). Near-infrared light indicates strong vegetation reflection, while red light indicates vegetation absorption. The calculation formula is NDVI = (B8-B4) / (B8+B4). The calculation formula of the difference between the normalized vegetation index NDVI of the two phase images before and after the landslide is: ΔNDVI=NDVI after -NDVI before Among them, NDVI before and NDVI after are the normalized vegetation indices of the images before and after the landslide, respectively.

4. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The area where the absolute value of the normalized vegetation index difference exceeds a preset threshold is divided into a change area, and the area where the absolute value does not exceed the preset threshold is divided into a non-change area.

5. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The cloud mask is established by combining shortwave infrared (SWIR) and NDVI thresholds to remove cloud interference in the changing area. The areas include: Pixels with NDVI < 0.8 were set as potential cloud pixels, and pixels with B11 (1610 nm) > 0.2 were set as marker cloud pixels; Mark the pixels that meet any of the above settings as clouds, assign all pixels marked as clouds a value of 1, and assign the remaining pixels a value of 0 to generate the final cloud mask map; According to the cloud mask image, the cloud interference area in the change area is determined to be removed.

6. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The normalized difference water index (NDWI) is used to eliminate water disturbance areas in the change zone, including: Pixels with NDWI > -0.15 are marked as water pixels to generate a water mask map; According to the water mask map, the water interference area in the change area is determined to be removed.

7. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The Otsu global threshold method is used to perform binary segmentation on the NDVI difference map to extract landslide candidate areas, including: Pixels with ΔNDVI values ​​between -0.05 and 0.05 were marked as 0, indicating non-landslide areas where vegetation did not change significantly; Pixels with ΔNDVI values ​​outside this range are marked as 1, and areas with significant changes in NDVI are extracted as landslide candidate areas.

8. The method for extracting landslide information based on optical remote sensing image features according to claim 1, wherein: The landslide area is finally extracted from the landslide candidate area based on the slope constraint generated by the pre-acquired digital elevation model (DEM): Calculating a slope distribution map of a landslide candidate area using the digital elevation model; The areas with slopes less than a preset threshold are removed from the landslide candidate areas to finally extract the landslide area.

Images