A method for constructing a potential surface anomaly remote sensing detector for single-temporal images

Abstract

The present invention discloses a method for constructing a potential surface anomaly remote sensing detector for single-phase images. A single-phase remote sensing image is input, and the remote sensing image is subjected to scene segmentation. For each scene in the remote sensing image, its objects are segmented separately, and various spectral characteristics and spatial characteristics of each object are calculated. Then, various spectral characteristics and spatial characteristics of the scene are calculated based on the characteristics of each object in the scene. Then, a difference index between the attributes of each object and the attributes of the scene in which it is located is constructed, namely a potential surface anomaly index. Finally, a suitable threshold is selected to classify the potential surface anomaly index. If the potential surface anomaly index of an object exceeds the threshold, it is determined to be a potential surface anomaly object. Different from existing surface anomaly remote sensing detection methods that require two-phase or multi-phase remote sensing images and can only detect specific surface anomaly events, this method can effectively detect various potential surface anomaly areas in remote sensing images based only on single-phase images. The method is simple in concept and easy to implement. It can be widely used in remote sensing detection of natural disasters, environmental pollution, ecological damage, safety accidents, illegal development and other events, and plays an important early warning role in protecting people's lives and property safety, social security, economic security and political security.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image processing, and in particular to a method for extracting regions of interest from remote sensing images. Background Art

[0002] Surface anomalies are phenomena that threaten the natural or social environment. They are caused by either natural or human factors, either singly or in combination, causing disruptions to the Earth's surface. These disruptions disrupt the stable equilibrium that has been established over time. These anomalies can be caused by natural disasters, environmental pollution, safety incidents, illegal development, and other events. Potential surface anomalies are areas of the Earth's surface that exhibit a state of disharmony with surrounding areas. Potential surface anomalies are essential for these conditions. With climate change and intensified human activities, the frequency, scope, and impact of surface anomalies are increasing. These events not only threaten people's lives and property, but also seriously endanger social stability and national security. Therefore, effectively detecting potential surface anomaly areas and providing early warnings is not only imperative for safeguarding people's lives and property, social security, and national security, but also a crucial approach to improving human well-being and protecting the ecological environment.

[0003] Satellite remote sensing, with its wide detection range, strong dynamic monitoring capabilities, rich information acquisition, and low monitoring costs, is the most effective method for detecting large-scale surface anomalies. Current remote sensing methods for detecting surface anomalies primarily rely on change detection in two- or multi-temporal imagery. These methods can be broadly categorized into manual visual interpretation and computer-generated automatic recognition. Manual visual interpretation involves manually interpreting remote sensing imagery from the same location at different times to detect surface anomalies. This method typically requires significant labor and has a low degree of automation. Computer-generated automatic recognition methods can be categorized into three types: thresholding, modeling, and machine learning classification. Thresholding methods first calculate the intensity of change in characteristic variables in remote sensing imagery over different time periods and then use the threshold to identify surface anomalies. The selection of thresholds relies on experience and regional characteristics, such as the Change Vector Analysis (CVA) method. Modeling methods primarily detect surface anomalies by fitting multi-temporal data to detect data that deviates from a reasonable range. These methods are relatively complex, such as the Additive Seasonal and Trend Decomposition and Breakpoint Detection (BFAST) algorithm. Machine learning classification methods, such as direct classification and post-classification comparison methods, train models by selecting samples and then use the trained models to detect surface anomalies. However, these methods face challenges in sample selection and model generalization. Furthermore, existing remote sensing methods for surface anomaly detection generally target specific anomalous events, such as thermal anomalies, floods, wildfires, earthquakes, and forest disturbances. These detection frameworks, tailored to specific events, cannot be applied to other surface anomalies. In summary, existing remote sensing methods for surface anomaly detection require bi-temporal or multi-temporal remote sensing imagery and can only detect surface anomalies caused by specific events.

[0004] This invention aims to use single-phase remote sensing imagery to detect various potential surface anomaly areas. The design principle is: if the attributes of an object in a remote sensing image are significantly inconsistent with those of its surrounding objects, the object is considered spatially incongruous with the surrounding objects, and is therefore highly likely to be the area where a surface anomaly event has occurred, thus being considered a potential surface anomaly area. Therefore, based on this characteristic of spatial incongruence between an object and its surrounding objects, we can first segment the image into several large-scale scenes, then further segment different objects within each scene, calculate multiple attributes of each object and each scene, and finally detect potential surface anomaly areas by comparing the differences between the attributes of each object and the attributes of its scene. Summary of the Invention

[0005] In response to the shortcomings of existing surface anomaly remote sensing detection methods, this paper proposes a method for constructing a potential surface anomaly remote sensing detector for single-temporal images. The method aims to detect all types of potential surface anomaly areas using only single-temporal remote sensing images. It not only has a high level of automation and generalization ability, but also is simple in concept and easy to implement.

[0006] To achieve this object, the present invention adopts the following technical solutions:

[0007] A method for constructing a potential surface anomaly remote sensing detector for single-temporal images includes the following steps:

[0008] A. Image input: Input a single-temporal remote sensing image.

[0009] B. Scene segmentation: perform scene segmentation on remote sensing images.

[0010] For remote sensing images of different spatial resolutions, it is necessary to select an appropriate downscaling factor to reduce the spatial resolution to an appropriate scale before performing scene segmentation. The appropriate scale refers to the minimum spatial resolution that can clearly distinguish the boundaries between different scenes. An inappropriate scale will make it difficult to obtain optimal scene segmentation results. The remote sensing image features used for scene segmentation must include spectral features. If they do not, optimal scene segmentation results cannot be obtained. Segmentation methods can include clustering, superpixel segmentation, and other methods.

[0011] C. Object segmentation: For each scene in the remote sensing image, its objects are segmented separately.

[0012] Segmentation methods include clustering, superpixel segmentation, etc. When performing object segmentation, the same object should be divided into the same object as much as possible.

[0013] D. Object attribute calculation: calculate various spectral and spatial features of each object and perform normalization.

[0014] When calculating object attributes, the attribute values ​​of all objects in the scene need to be normalized to unify the weights of each attribute, which facilitates the subsequent construction of potential surface anomaly indicators.

[0015] E. Scene attribute calculation: Calculate various spectral and spatial features of the scene based on the attributes of each object in the scene.

[0016] When calculating scene attributes, methods such as direct averaging, depolarization averaging, and maximum likelihood estimation of the attributes of all objects in the scene can be used.

[0017] F. Construction of potential surface anomaly indicators, constructing the difference indicators between the attributes of each object and the attributes of the scene in which it is located.

[0018] Statistical methods such as paired t-test and machine learning methods such as Euclidean distance, Mahalanobis distance, maximum likelihood, and local outlier factor can be used to construct potential surface anomaly indicators.

[0019] G. Detection of potential surface anomaly objects: Select appropriate thresholds to classify potential surface anomaly indicators. If the potential surface anomaly indicator of an object exceeds the threshold, it is identified as a potential surface anomaly object.

[0020] Threshold selection methods include cumulative frequency values, statistical test confidence levels, etc. When using thresholds to divide potential surface anomaly indicators, single or double thresholds should be used according to the detection purpose and the appropriate threshold size should be selected to ultimately extract potential surface anomaly areas.

[0021] The present invention has the following characteristics:

[0022] (1) Using only single-temporal remote sensing images

[0023] (2) Capable of detecting all types of potential surface anomalies.

[0024] (3) The idea is simple and easy to implement.

[0025] (4) It has wide applicability and can be widely used to detect various potential surface anomaly areas in remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a process for constructing remote sensing detectors of potential surface anomalies for single-temporal images.

[0027] Figure 2 It is a high spatial resolution satellite remote sensing image.

[0028] Figure 3 It is the result of downscaling of high spatial resolution satellite remote sensing images.

[0029] Figure 4 is the scene segmentation result.

[0030] Figure 5 is the object segmentation result.

[0031] Figure 6 It is the calculation result of potential surface anomaly index.

[0032] Figure 7 It is the extraction result of potential surface anomaly area. DETAILED DESCRIPTION

[0033] The technical implementation scheme of the present invention will be further described below in conjunction with the accompanying drawings.

[0034] We now use a single-phase, high-spatial-resolution satellite remote sensing image to detect potential surface anomalies. This image consists of three red, green, and blue bands, with a spatial resolution of 0.3 meters and 1024 columns x 1024 rows of pixels. The values ​​for each red, green, and blue band range from 0 to 255, and are byte data.

[0035] A. Image input: Input a single-temporal remote sensing image.

[0036] In this case, a single-phase high-spatial-resolution satellite remote sensing image is input, namely Figure 2 .

[0037] B. Scene segmentation: perform scene segmentation on remote sensing images.

[0038] For remote sensing images of different spatial resolutions, it is necessary to select an appropriate downscaling factor to reduce the spatial resolution to an appropriate scale before performing scene segmentation. The appropriate scale refers to the minimum spatial resolution that can clearly distinguish the boundaries between different scenes. An inappropriate scale will make it difficult to obtain optimal scene segmentation results. The remote sensing image features used for scene segmentation must include spectral features. If they do not, optimal scene segmentation results cannot be obtained. Segmentation methods can include clustering, superpixel segmentation, and other methods.

[0039] The remote sensing image used in this case has a spatial resolution of 0.3m, which is a high spatial resolution image. It is downscaled using a Gaussian pyramid with a downscaling factor of 128. The downscaling results are shown in Figure 3 ; Then use K-means clustering method to divide the scene of the downscaled image. The features used in scene division are the grayscale values ​​of the red, green and blue bands after downscaling, that is, spectral features. The scene segmentation results are shown in Figure 4 .

[0040] C. Object segmentation: For each scene in the remote sensing image, its objects are segmented separately.

[0041] In this case, the Felzenszwalb superpixel segmentation method is used to segment the image according to the scene to obtain each object. The object segmentation results are shown in Figure 5 .

[0042] D. Calculate the attributes of each object, calculate the various spectral characteristics and spatial characteristics of each object, and perform normalization.

[0043] In this case, eight attributes of each object in each scene were calculated: red light band, green light band, blue light band, distance coding diversity, angular second moment, contrast, homogeneity, and neighborhood information entropy. To unify the weights of each attribute, the attributes of all objects in each scene were normalized.

[0044] E. Scene attribute calculation: Calculate various spectral and spatial features of the scene based on the attributes of each object in the scene.

[0045] In this case, the mean of all object attributes in the scene is used as the scene attribute.

[0046] F. Construction of potential surface anomaly indicators, constructing the difference indicators between the attributes of each object and the attributes of the scene in which it is located.

[0047] In this case, a paired t-test is performed on the scene attributes and the attributes of each object in the scene. The p-value of the paired t-test is used as the potential surface anomaly indicator. The calculation results of the potential surface anomaly indicator are shown in Figure 6 .

[0048] G. Detection of potential surface anomaly objects: Select appropriate thresholds to classify potential surface anomaly indicators. If the potential surface anomaly indicator of an object exceeds the threshold, it is identified as a potential surface anomaly object.

[0049] In this case, p=0.05 is selected as the threshold. Objects with potential surface anomaly indicators less than the threshold are considered potential surface anomaly objects. The extraction results of potential surface anomaly areas are shown in Figure 7 .

[0050] The above is merely one specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by anyone familiar with the art within the technical scope disclosed by the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for constructing a potential surface anomaly remote sensing detector for single-temporal images, characterized in that: The following steps are involved: A. Image input: input a single-temporal remote sensing image; B. Scene segmentation: reducing the spatial resolution of the single-temporal remote sensing image input in step A to an appropriate scale and then performing scene segmentation, wherein the features used in scene segmentation are the grayscale values ​​of the red, green, and blue bands after downscaling; C. Object segmentation: for each scene in the remote sensing image, segment its objects separately; D. Object attribute calculation, calculate various spectral characteristics and spatial characteristics of each object, and perform normalization processing; E. Scene attribute calculation, which calculates various spectral and spatial features of each scene based on the characteristics of each object in the scene; F. Constructing a potential surface anomaly indicator: constructing a difference indicator between the attributes of each object and the attributes of the scene in which it is located. A paired t-test is performed on the attributes of the scene and the attributes of each object in the scene, and the p-value of the paired t-test is used as the potential surface anomaly indicator; G. Detection of potential surface anomaly objects: Select appropriate thresholds to classify potential surface anomaly indicators. If the potential surface anomaly indicator of an object exceeds the threshold, it is identified as a potential surface anomaly object.

2. The method for constructing a potential surface anomaly remote sensing detector for single-temporal imagery according to claim 1, characterized in that: In step B, the scene segmentation is characterized by: A. For remote sensing images with different spatial resolutions, it is necessary to select an appropriate downscaling factor; B. The remote sensing image features used for scene segmentation must include spectral features.

Images