Glacier identification method based on remote sensing image
Through multi-source remote sensing data fusion and BoTNet network architecture, the problem of insufficient recognition accuracy of remote sensing image glacier recognition technology in high latitudes and complex terrain is solved, and efficient and accurate glacier recognition is achieved, adapting to changes in glacier characteristics, reducing costs and labor Intervention.
Patent Information
- Application Number
- CN202411216869.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-26
AI Technical Summary
Existing glacier identification technology based on remote sensing images has high data acquisition cost and low recognition accuracy in high latitude and high altitude areas. Especially under complex terrain and diverse climatic conditions, it has limited ability to identify small glaciers, and has great subjectivity and dependence. Test knowledge.
Multi-source remote sensing data fusion optimization, including the fusion of high-resolution remote sensing images, digital elevation data and multi-spectral data, the glacier recognition cycle model is constructed using the BoTNet network architecture, and glacier features are extracted through Gram-Schmidt transformation and CNN model to reduce manual intervention And improve the level of automation.
It significantly improves the accuracy and efficiency of glacier identification, reduces time and economic costs, and can stably identify glacier changes under complex conditions, providing strong technical support for climate change research and water resource management.
Smart Images

Figure CN119027830A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of computer vision, and in particular relates to a glacier identification method based on remote sensing images. Background Art
[0002] As global climate change intensifies, glaciers, as important freshwater resources on Earth and sensitive indicators of climate change, are of great significance for understanding the function of the Earth system and predicting future climate change trends. Traditional glacier monitoring methods mainly rely on ground observations, which have shortcomings such as high cost, low efficiency, and limited temporal and spatial resolution. The development of remote sensing technology has provided a new way for glacier monitoring, but the existing glacier identification methods based on remote sensing images have many technical bottlenecks. For example, the cost of obtaining high-quality remote sensing image data is high, especially in high-latitude and high-altitude areas. Cloud cover and extreme weather conditions further limit data acquisition and quality.
[0003] Existing algorithms have low accuracy in identifying glaciers under complex terrain and diverse climatic conditions. In particular, the ability to identify small glaciers or glaciers partially covered by gravel is limited. For example, traditional statistical methods based on empirical thresholds for identifying glaciers, such as the ratio threshold method, the (enhanced) snow cover index method, and the coherence threshold method, are highly subjective and vary with the study area and time period. In addition, although traditional machine learning methods such as support vector machines and random forests reduce manual involvement and improve recognition accuracy, these methods also require more prior knowledge.
[0004] In summary, glacier identification technology based on remote sensing images still faces major challenges in terms of data source acquisition and identification accuracy. In order to improve the efficiency and accuracy of glacier monitoring, it is necessary to develop new technical methods, make full use of multi-source data, and improve the automation and intelligence level of the algorithm to adapt to the ever-changing glacier characteristics and monitoring needs. The present invention proposes an improved glacier identification method based on remote sensing images, in order to solve the deficiencies in the existing technology and provide more accurate and efficient technical support for glacier research and global change monitoring. Summary of the invention
[0005] In order to solve the above technical problems, the present invention proposes a glacier identification method based on remote sensing images to solve the problems existing in the above prior art.
[0006] To achieve the above object, the present invention provides a glacier identification method based on remote sensing images, comprising:
[0007] Acquiring multi-source remote sensing data with time series characteristics based on the target area, wherein the multi-source remote sensing data includes: high-resolution remote sensing images, digital elevation data and multi-spectral data;
[0008] Based on the digital elevation data, the high-resolution remote sensing image and the multispectral data are fused and optimized to obtain a fused image;
[0009] A glacier recognition cycle model is constructed based on the BoTNet network architecture; and the fused image is input into the glacier recognition cycle model to obtain a glacier recognition result.
[0010] Optionally, the process of fusion optimization based on the multi-source remote sensing data also includes multi-source remote sensing data registration; wherein the process of multi-source remote sensing data registration includes:
[0011] Preprocessing the multi-source remote sensing data to obtain preprocessed data;
[0012] Performing coordinate transformation on the preprocessed data to obtain high-resolution remote sensing images, digital elevation data and multispectral data in the same coordinate system;
[0013] Extract feature points of high-resolution remote sensing images, digital elevation data, and multispectral data in the same coordinate system based on SIFT algorithm;
[0014] Based on the feature points of the high-resolution remote sensing image, digital elevation data and multispectral data in the same coordinate system, geographic registration is performed to obtain the registered multi-source remote sensing data.
[0015] Optionally, the process of geo-referencing feature points of the high-resolution remote sensing image, digital elevation data, and multispectral data in the same coordinate system includes:
[0016] Based on the fast nearest neighbor search library algorithm, the corresponding relationship between the extracted feature points is established between the high-resolution remote sensing image and the digital elevation data, and between the digital elevation data and the multispectral data;
[0017] Based on the corresponding relationship, a two-dimensional affine transformation matrix is used to perform geographic transformation to obtain the registered multi-source remote sensing data.
[0018] Optionally, the formula for the geographic transformation is:
[0019]
[0020] Where (X1, y1) is the original coordinate, (X2, y2) is the transformed coordinate, and a, b, c, d, e, f are transformation parameters.
[0021] Optionally, the process of performing data fusion optimization based on the registered multi-source remote sensing data includes:
[0022] The Gram-Schmidt transformation method is used to preliminarily fuse the registered high-resolution remote sensing image and multispectral data to obtain the fused multispectral image.
[0023] Extract terrain features based on the fused multispectral images and the registered digital elevation data;
[0024] The terrain features and the fused multispectral image are input into the CNN model for fusion optimization to obtain a fused image.
[0025] Optionally, the process of performing preliminary fusion using the Gram-Schmidt transform method includes:
[0026] Perform principal component analysis on several bands of multispectral data to obtain the first principal component;
[0027] Based on the first principal component, Gram-Schmidt orthogonalization is performed on each remaining band in the multispectral data to obtain a plurality of orthogonal components;
[0028] Calculate the correlation between high-resolution remote sensing images and each orthogonal component to obtain several correlation indices;
[0029] Based on the several correlation indices, weighted calculation is performed on the high-resolution remote sensing image and the first principal component to obtain a fusion component;
[0030] The fused components are inversely transformed to reconstruct a fused multispectral image.
[0031] Optionally, a glacier identification cyclic model constructed based on the BoTNet network architecture includes: an input layer, a feature extraction layer, a number of BoTNet modules, a cyclic network structure and an output layer;
[0032] The process of performing glacier identification based on the glacier identification cycle model includes:
[0033] Input the fused image with time series characteristics into the input layer of the glacier recognition cycle model, perform image coding to obtain image coding information with time series characteristics;
[0034] The feature extraction layer receives the image encoding information with time series features and performs low-level feature extraction to obtain low-level time series features of the glacier;
[0035] The low-level time series features of the glacier are input into the BoTNet module to extract the high-level time series features of the glacier, and the target glacier and the glacier status are obtained based on the high-level time series features of the glacier.
[0036] Optionally, the glacier identification cycle model further includes optimizing a BoTNet module based on a glacier state, wherein the process of optimizing the BoTNet module based on a glacier state includes:
[0037] Inputting the glacier state into a cyclic network structure to obtain a previous state target glacier and a next state target glacier;
[0038] Calculating target errors between the previous state target glacier and the previous state fused image of the input layer, and between the next state target glacier and the next state fused image of the input layer respectively;
[0039] Optimizing the BoTNet module based on the target error;
[0040] An error threshold is set, and when the errors of the two targets are within the error threshold, the output layer outputs the glacier recognition result.
[0041] The present invention also provides a computer terminal device, comprising:
[0042] one or more processors;
[0043] A memory, coupled to the processor, for storing one or more programs;
[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement a glacier identification method based on remote sensing images.
[0045] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a glacier identification method based on remote sensing images is implemented.
[0046] Compared with the prior art, the present invention has the following advantages and technical effects:
[0047] This paper proposes an innovative glacier identification method based on remote sensing images, aiming to solve the problems of limited data sources and insufficient identification accuracy in the existing technology. This method significantly improves the accuracy and efficiency of glacier identification by integrating high-resolution remote sensing images, digital elevation data and multispectral data and using deep learning technology.
[0048] The data fusion optimization link of the present invention adopts the Gram-Schmidt transformation method for preliminary fusion, and strengthens the spectral characteristics of the image through principal component analysis and orthogonalization processing. Further, the terrain characteristics and the CNN model are combined for deep fusion to generate a fused image rich in elevation and spectral information, which provides richer contextual information for the feature extraction of glaciers.
[0049] The present invention also introduces a glacier recognition cycle model constructed by the BoTNet network architecture. The model can effectively encode and extract features from fused images. In particular, the application of the BoTNet module extracts high-level features of glacier time series through an optimized network structure, significantly improving the performance of glacier recognition.
[0050] This method not only improves the accuracy of glacier identification, but also enhances the ability of automatic processing, reduces manual intervention, and reduces time and economic costs. By integrating multi-source data and deep learning technology, this method can effectively identify and monitor glacier changes, and has important application value in climate change research, water resources management, and disaster prevention and mitigation. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0052] Figure 1 This is a flow chart of a glacier identification method based on remote sensing images according to an embodiment of the present invention. DETAILED DESCRIPTION
[0053] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0054] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0055] Embodiment 1
[0056] like Figure 1 As shown, this embodiment provides a glacier identification method based on remote sensing images, including the following steps: acquiring multi-source remote sensing data with time series characteristics based on the target area, wherein the multi-source remote sensing data includes: high-resolution remote sensing images, digital elevation data and multi-spectral data; fusing and optimizing the high-resolution remote sensing images and multi-spectral data based on the digital elevation data to obtain a fused image; constructing a glacier identification cycle model based on the BoTNet network architecture; and inputting the fused image into the glacier identification cycle model to obtain a glacier identification result.
[0057] Furthermore, as a preferred solution of this embodiment, the following specific implementation steps are included:
[0058] S1. Obtain multi-source remote sensing data with time series characteristics in the target area: The multi-source remote sensing data used in this study mainly include: (1) Remote sensing images, including Landsat 8 and Sentinel-1, Sentinel-2 satellite image data; through multi-period remote sensing image comparison and analysis, combined with Google Earth historical images, obtain Landsat 8 and Sentinel-2 images with the lowest snow cover and cloud cover below 10% in the study area; (2) DEM data. Based on digital elevation model (DEM) data, various terrain information of glaciers in high mountain Asia can be obtained, such as elevation, slope, slope aspect, landform shading and other parameters. These terrain information can be used as auxiliary data for remote sensing image classification and participate in classification, which helps to improve the accuracy of image classification. The DEM data used in this embodiment include two types, namely ASTER GDEMV2 and SRTMDEM data. ASTER GDEM V2 data is obtained through the geospatial data cloud website, and SRTM DEM data is obtained through the USGS website. (3) Multispectral data, mainly obtained through multispectral sensors or hyperspectral sensors.
[0059] S2, multi-source data registration: The first step of multi-source remote sensing data fusion optimization is data registration, which ensures that high-resolution remote sensing images, digital elevation data and multispectral data are aligned in the same coordinate system. The process starts with preprocessing the original data to obtain preprocessed data. Subsequently, coordinate transformation technology is applied to the preprocessed data to achieve the unification of data in the same coordinate system. Next, the scale-invariant feature transform (SIFT) algorithm is used to extract feature points from high-resolution remote sensing images, digital elevation data and multispectral data, and these feature points are consistent in geographic space. Through these feature points, geographic registration is performed to accurately match data from different sources to eliminate spatial deviations and ensure geometric consistency between data. Finally, the registered multi-source remote sensing data is obtained. As a preferred embodiment of the present embodiment, the present embodiment uses DEM data as a geographic reference to accurately align high-resolution remote sensing images and multispectral data with temporal characteristics to the same geographic coordinate system. The specific implementation process includes:
[0060] S2.1 Preprocessing of multi-source remote sensing data
[0061] Data inspection: First, check the quality of DEM data, high-resolution remote sensing images and spectral data, including integrity, noise level and resolution.
[0062] Radiation correction: Radiation correction is performed on high-resolution remote sensing images and multispectral data to eliminate the effects of factors such as the atmosphere and solar altitude on the images, ensuring that the radiation values of the image data can accurately reflect the true surface conditions. This process involves the application of an atmospheric correction model, which can simulate and correct the absorption and scattering effects of the atmosphere on light.
[0063] Preliminary geometric correction: Perform preliminary geometric correction on the image to eliminate geometric distortion caused by factors such as sensor attitude and earth curvature. Preliminary geometric correction usually includes affine transformation or more complex geometric transformation on the image to ensure that the image data can accurately reflect the spatial position of surface objects.
[0064] Through the above preprocessing steps, the quality of remote sensing data can be significantly improved, laying a solid foundation for the fusion and optimization of multi-source remote sensing data. Preprocessing not only improves the availability of data, but also provides reliable data support for further analysis and applications, such as glacier identification, terrain analysis, etc.
[0065] S2.2 Coordinate system unification
[0066] Determine the reference coordinate system: select a unified geographic coordinate system as the reference coordinate system for all data. In this embodiment, the coordinate system of the DEM data is selected as the reference coordinate system, that is, this embodiment uses the WGS84 coordinate system as the unified geographic coordinate system.
[0067] Coordinate conversion: Use coordinate conversion tools or software to convert DEM data, high-resolution remote sensing images, and multispectral data from their respective original coordinate systems to a unified reference coordinate system. The coordinate conversion tools or software used include but are not limited to ArcGIS and QGIS.
[0068] S2.3 Geometric Correction and Registration
[0069] DEM-based georeferencing: Using DEM data as a georeference can ensure the consistency of terrain features during the registration process. DEM data contains surface elevation information and can be used as a benchmark for terrain matching.
[0070] Feature point extraction: Based on the SIFT algorithm, significant feature points are extracted from DEM data, high-resolution remote sensing images and spectral data. Significant feature points include mountain tops, valleys, river intersections, etc. These feature points play a key role in the registration process.
[0071] Registration algorithm selection: Select a suitable registration algorithm according to data characteristics and requirements, wherein the optional registration algorithms of the present invention include but are not limited to: feature point-based registration, grayscale-based registration, and terrain illumination model-based registration.
[0072] Furthermore, the present embodiment further performs registration based on the extracted feature points. The present embodiment establishes a correspondence between the extracted feature points in the high-resolution remote sensing image and the digital elevation data, and between the digital elevation data and the multispectral data based on the fast nearest neighbor search library algorithm (FLANN), and at the same time uses the RANSAC algorithm to remove the wrong matching points or incomplete matching points in the matching results.
[0073] Furthermore, this embodiment establishes registration based on the correspondence relationship of the optimized feature points. In the registration process, matrix transformation is the key to geometric correction to achieve data alignment. The matrix transformation in the registration process is as follows:
[0074] The coordinates of the point in the original coordinate system are (x1, y1), and the coordinates of the point in the transformed coordinate system are (x2, y2). The transformation relationship is represented by a two-dimensional affine transformation matrix T:
[0075]
[0076] Among them, a, b, c, d, e, and f are transformation parameters. For two-dimensional coordinates, we actually only care about the elements in the first two rows and columns. The transformation formula can be expressed as:
[0077]
[0078] Right now
[0079]
[0080] S2.4 Registration accuracy detection: This embodiment evaluates the registration accuracy by superimposing the display, and when the registration accuracy reaches a preset value, the registered multi-source remote sensing data is obtained.
[0081] S3. Based on the registered multi-source remote sensing data, the high-resolution remote sensing images and multispectral data are fused, and the fusion results are optimized using digital elevation data to obtain a fused image. In the data fusion optimization process, the Gram-Schmidt transformation technology is first used to perform a preliminary fusion of the registered remote sensing images and multispectral data to generate a fused multispectral image. This transformation method enhances the spectral characteristics of the data through orthogonalization processing, providing rich information for further analysis. Subsequently, terrain features are extracted from the fused multispectral image. These features reflect the morphological changes of the surface and are crucial to understanding the surface characteristics. Finally, the terrain features and the fused multispectral image are input into the convolutional neural network (CNN) model for deep fusion optimization. The CNN model intelligently integrates the information by learning the deep features of the data, and finally generates a fused image with high spatial resolution and rich spectral information. The specific implementation process includes:
[0082] S3.1 Extract reference bands: Perform principal component analysis on several bands of multispectral data to obtain the first principal component; use the first principal component as the initial reference band.
[0083] S3.2 Gram-Schmidt orthogonalization: Take the selected reference band as the first orthogonal component; for each remaining band in the multispectral data, use the Gram-Schmidt process to calculate its orthogonal component with the existing orthogonal components. In this process, high-resolution remote sensing images are used to "enhance" the spatial resolution of each orthogonal component, that is, by weighting the residual of high-resolution remote sensing images with each orthogonal component.
[0084] The Gram-Schmidt transform method was used for preliminary fusion in the data fusion optimization phase. The spectral characteristics of the image were enhanced through principal component analysis and orthogonalization. The terrain features and CNN model were further combined for deep fusion to generate a fused image rich in elevation and spectral information, which provided richer contextual information for glacier feature extraction.
[0085] S3.3 Replace high-resolution information: Calculate the correlation between the high-resolution remote sensing image and each orthogonal component, and weight the high-resolution remote sensing image and the original orthogonal components according to these correlations to generate new fused components.
[0086] S3.4 Reconstruction of fused image and post-processing: Reconstruct the fused multispectral image using the new fused components through inverse transformation, and adjust the color of the fused image to ensure that it is visually consistent with the original multispectral data.
[0087] S3.5 Use digital elevation data to optimize the fused multispectral image to obtain a fused image: This embodiment uses DEM data to improve the terrain expression ability of the fused image and enhance the spatial analysis ability. The specific process includes:
[0088] Extract terrain features related to multispectral images from DEM data, such as slope, aspect, elevation, etc. These features can be used as an additional source of information to optimize the fused image.
[0089] The extracted terrain features are matched and overlaid with the multispectral images in the form of raster data to ensure their spatial consistency.
[0090] The fused multispectral image and DEM terrain features are input into the CNN model. The input layer receives multiple bands of the multispectral image and the DEM terrain feature map; the convolution layer extracts high-level representations of the image and terrain features through multiple convolution layers; then, the high-level representations of the image and terrain features are upsampled using the transposed convolution layer to obtain transposed image features and transposed terrain features; the fusion layer effectively fuses the transposed image features and transposed terrain features based on the attention mechanism to obtain fused features. The output layer generates an optimized fused image based on the fused features. The CNN model automatically adjusts and optimizes the pixel values of the image based on the input multispectral image and DEM terrain features to generate a more accurate and rich fused image.
[0091] S4. Further, BoTNet is an architecture that combines convolutional neural network (CNN) and Transformer. This embodiment constructs a glacier recognition cycle model based on the BoTNet network architecture. The glacier recognition cycle model includes: an input layer, a feature extraction layer, a BoTNet module, a cycle network structure, and an output layer;
[0092] 1) Input layer: After the image is input, the input layer performs an embedding operation on the image, maps the input data to a multi-dimensional vector, and encodes the sequential information in the data.
[0093] 2) Feature extraction layer: ResNet is used as the front end to extract the features of the encoded image information to obtain low-level temporal features. The output of the feature extraction layer is used as the input of BoTNet to further extract high-level temporal features.
[0094] 3) BoTNet module (Bottleneck Transformer Block): In each block, the number of channels is first reduced by 1x1 convolution, then the Transformer encoder layer is applied, and finally the original number of channels is restored by 1x1 convolution. This design maintains computational efficiency while introducing global context information.
[0095] Furthermore, this embodiment constructs a deep network by stacking multiple Bottleneck Transformer Blocks.
[0096] 4) Cyclic network structure
[0097] Forward model: Use the normal sequential data of the training set to train a BoTNet model to predict the next state or feature of the target.
[0098] Reverse model: Use the reverse data of the training set to train another BoTNet model, trying to infer the target state at different time intervals ago from the current state of the result.
[0099] In order to further improve the robustness and generalization ability of the model, this embodiment also introduces a cyclic consistency training strategy. This strategy minimizes the consistency loss of data after cyclic transformation by constructing a "forward-reverse" and "reverse-forward" cyclic process. Specifically, this embodiment first uses a forward model to perform prediction training on a set of time series data, and the results are then used as the input of the reverse model for reverse prediction. Ideally, the output of the reverse model should be as close to the original input data as possible. Similarly, the "reverse-forward" cycle follows a similar logic. By continuously optimizing the consistency loss in these two cyclic processes, it is possible to ensure that the model maintains a high degree of consistency between forward and reverse predictions, thereby improving the overall performance and stability of the model. Cyclic consistency training not only helps the model to better identify the target state under limited data conditions, but also effectively avoids the problem of model overfitting.
[0100] 5) Output layer: Use the fully connected layer to map the output of BoTNet to the segmentation map to obtain the final recognition result.
[0101] 5. Input the fused image into the glacier recognition cycle model to obtain the glacier recognition result. The specific implementation process includes:
[0102] In the glacier recognition cycle model, the fused images that integrate temporal features are first taken as input. Through the embedding operation of the input layer, these images are mapped to the multidimensional vector space, and the temporal information therein is accurately encoded at the same time, thereby generating image codes rich in time dimension information.
[0103] The feature extraction layer then receives this encoded information and performs preliminary low-level feature extraction, focusing on significant features such as the edge contour of the glacier.
[0104] Then, the BoTNet module further explores the high-level temporal features of the glacier, including detailed geographic information, spectral characteristics, and surface morphology. These high-level features together construct a complete portrait of the glacier, making the identification process more accurate.
[0105] Based on these high-level features, the model can accurately locate and identify the target glacier and further analyze its status. Then, using the unique advantages of the recurrent network structure, the model can predict and compare the changes in the state of the glacier at different time points, and continuously optimize the performance of the BoTNet module by calculating the error between the fused image at the corresponding time point in the input layer.
[0106] Finally, when the model prediction error drops to within a preset threshold, it indicates that the identification process has reached a high level of accuracy, and the output layer will confidently present the glacier identification results.
[0107] Furthermore, as a preferred solution of this embodiment, the output of this embodiment may also be a pixel-level glacier / non-glacier classification map or a classification result of the entire image.
[0108] In glacier identification, accuracy is a measure of the ratio of glacier areas to non-glacier areas correctly identified by the model. Assuming there are four cases: TP (true positive, i.e., glacier areas are correctly identified as glaciers), TN (true negative, i.e., non-glacier areas are correctly identified as non-glaciers), FP (false positive, i.e., non-glacier areas are incorrectly identified as glaciers), and FN (false negative, i.e., glacier areas are incorrectly identified as non-glaciers), the accuracy calculation formula is:
[0109]
[0110] Recall, also known as True Positive Rate (TPR), measures the model's ability to identify all glacier areas. Its calculation formula is:
[0111]
[0112] A high recall means that the model is able to find glacier areas more comprehensively, but it may sacrifice some accuracy in some cases, i.e., misidentifying non-glacier areas as glaciers.
[0113] This method is based on the mechanism of optimizing the BoTNet module based on glacier state. By calculating the target error and feeding it back to the recurrent network structure, the model is self-optimized. This dynamic optimization strategy enables the model to adapt to the needs of glacier identification in different states, further improving the accuracy of the identification results.
[0114] In summary, the glacier identification method based on remote sensing images of the present invention generates a fused image with high spatial resolution and rich spectral information by innovatively fusing high-resolution remote sensing images, digital elevation data and multispectral data, and using Gram-Schmidt transform technology for preliminary fusion. The application of the deep learning model BoTNet combines the advantages of CNN and Transformer to further improve the accuracy and generalization ability of recognition. The accuracy of georeferencing technology ensures the alignment of multi-source data in the same coordinate system and improves the reliability of analysis. The fusion optimization of the CNN model enhances the extraction of terrain features and provides accurate data support for glacier identification. The automation and intelligence of the entire processing flow reduces manual intervention and improves efficiency. The method has strong adaptability, can work stably under complex terrain and diverse climatic conditions, and has the potential for large-scale application. It has important practical application value in the fields of climate change research, water resources management, and disaster prevention and mitigation.
[0115] The remote sensing image-based glacier identification method proposed in the present invention realizes efficient and accurate identification of glaciers through advanced data fusion technology and deep learning models, providing strong technical support for research and application in related fields.
[0116] Embodiment 2
[0117] This embodiment also provides a computer terminal device, including:
[0118] one or more processors;
[0119] A memory, coupled to the processor, for storing one or more programs;
[0120] When the one or more programs are executed by the one or more processors, the one or more processors implement a glacier identification method based on remote sensing images.
[0121] Embodiment 3
[0122] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a glacier identification method based on remote sensing images is implemented.
[0123] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A glacier identification method based on remote sensing images, It is characterized in that The following steps are involved: Acquiring multi-source remote sensing data with time series characteristics based on the target area, wherein the multi-source remote sensing data includes: high-resolution remote sensing images, digital elevation data and multi-spectral data; Based on the digital elevation data, the high-resolution remote sensing image and the multispectral data are fused and optimized to obtain a fused image; A glacier recognition cycle model is constructed based on the BoTNet network architecture; and the fused image is input into the glacier recognition cycle model to obtain a glacier recognition result.
2. The glacier identification method based on remote sensing images according to claim 1, It is characterized in that The process of fusion optimization based on the multi-source remote sensing data also includes multi-source remote sensing data registration; wherein the process of multi-source remote sensing data registration includes: Preprocessing the multi-source remote sensing data to obtain preprocessed data; Performing coordinate transformation on the preprocessed data to obtain high-resolution remote sensing images, digital elevation data and multispectral data in the same coordinate system; Extract feature points of high-resolution remote sensing images, digital elevation data, and multispectral data in the same coordinate system based on SIFT algorithm; Based on the feature points of the high-resolution remote sensing image, digital elevation data and multispectral data in the same coordinate system, geographic registration is performed to obtain the registered multi-source remote sensing data.
3. The glacier identification method based on remote sensing images according to claim 2, It is characterized in that The process of georeferencing based on the feature points of high-resolution remote sensing images, digital elevation data, and multispectral data in the same coordinate system includes: Based on the fast nearest neighbor search library algorithm, the corresponding relationship between the extracted feature points is established between the high-resolution remote sensing image and the digital elevation data, and between the digital elevation data and the multispectral data; Based on the corresponding relationship, a two-dimensional affine transformation matrix is used to perform geographic transformation to obtain the registered multi-source remote sensing data.
4. The glacier identification method based on remote sensing images according to claim 3, It is characterized in that The formula for geographic transformation is: In the formula, (x 1 ,y 1 ) is the original coordinate, (X 2 ,y 2 ) are the transformed coordinates, and a, b, c, d, e, f are the transformation parameters.
5. The glacier identification method based on remote sensing images according to claim 2, It is characterized in that The process of data fusion optimization based on the registered multi-source remote sensing data includes: The Gram-Schmidt transformation method is used to preliminarily fuse the registered high-resolution remote sensing image and multispectral data to obtain the fused multispectral image. Extract terrain features based on the fused multispectral images and the registered digital elevation data; The terrain features and the fused multispectral image are input into the CNN model for fusion optimization to obtain a fused image.
6. The glacier identification method based on remote sensing images according to claim 5, It is characterized in that The process of preliminary fusion using the Gram-Schmidt transform method includes: Perform principal component analysis on several bands of multispectral data to obtain the first principal component; Based on the first principal component, Gram-Schmidt orthogonalization is performed on each remaining band in the multispectral data to obtain a plurality of orthogonal components; Calculate the correlation between high-resolution remote sensing images and each orthogonal component to obtain several correlation indices; Based on the several correlation indices, weighted calculation is performed on the high-resolution remote sensing image and the first principal component to obtain a fusion component; The fused components are inversely transformed to reconstruct a fused multispectral image.
7. The glacier identification method based on remote sensing images according to claim 1, It is characterized in that The glacier identification cycle model built based on the BoTNet network architecture includes: input layer, feature extraction layer, several BoTNet modules, cycle network structure and output layer; The process of performing glacier identification based on the glacier identification cycle model includes: Input the fused image with time series characteristics into the input layer of the glacier recognition cycle model, perform image coding to obtain image coding information with time series characteristics; The feature extraction layer receives the image encoding information with time series features and performs low-level feature extraction to obtain low-level time series features of the glacier; The low-level time series features of the glacier are input into the BoTNet module to extract the high-level time series features of the glacier, and the target glacier and the glacier status are obtained based on the high-level time series features of the glacier.
8. The glacier identification method based on remote sensing images according to claim 7, It is characterized in that The glacier identification cycle model further includes optimizing a BoTNet module based on a glacier state, wherein the process of optimizing the BoTNet module based on a glacier state includes: Inputting the glacier state into a cyclic network structure to obtain a previous state target glacier and a next state target glacier; Calculating target errors between the previous state target glacier and the previous state fused image of the input layer, and between the next state target glacier and the next state fused image of the input layer respectively; Optimizing the BoTNet module based on the target error; An error threshold is set, and when the errors of the two targets are within the error threshold, the output layer outputs the glacier recognition result.
9. A computer terminal device, It is characterized in that include: one or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the glacier identification method based on remote sensing images as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the remote sensing image-based glacier identification method according to any one of claims 1 to 8 is implemented.
Citation Information
Cited By
Remote sensing image geographic element automatic identification method and system based on large model analysis
CN119445582A
Automatic Recognition Method and System for Geographical Elements of Remote Sensing Images Based on Large Model Analysis
CN119445582B
Highway construction multi-source data fusion processing system based on cloud platform
CN120071071A
Highway construction multi-source data fusion processing system based on cloud platform
CN120071071B
Glacier surface moraine identification method based on multi-source remote sensing parameters
CN120495933A
