Mountain area lake identification method and device based on environmental characteristics

Abstract

The invention discloses a mountain area lake identification method and device based on environmental characteristics. The method comprises the following steps: acquiring a plurality of remote sensing images of a target water body; calculating a comprehensive water body index based on the remote sensing image so as to enhance the spectral characteristics of the water body and obtain a spectral enhanced image; calculating topographic data based on the spectrum enhanced image, calculating a buffer area of the target water body according to the topographic data, and extracting a remote sensing image of the buffer area; and inputting the remote sensing image of the target water body, the remote sensing image of the buffer area and the topographic data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model. The technical problems that an existing method is limited by information representation capacity of traditional local spectrums and texture features, and accurate recognition of small water bodies in the alpine regions is difficult to achieve are solved, and therefore remote sensing recognition accuracy of lakes in the alpine regions is improved.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a method and device for identifying mountain lakes based on environmental characteristics. Background Art

[0002] Remote sensing technology has become the dominant means of monitoring lakes in alpine regions. However, remote sensing identification of lakes in alpine regions is still very difficult. This difficulty mainly stems from the following aspects:

[0003] First, the unique geographical environment of lakes in alpine regions poses a huge challenge to data acquisition. The lakes in these regions are covered with ice and snow for most of the annual cycle, and the high-altitude areas are generally cloud-covered, which greatly limits the availability of high-quality remote sensing images. The widespread presence of clouds and snow not only reduces the opportunity to obtain data, but also significantly increases the complexity of image processing, and puts forward more stringent requirements for the accurate remote sensing identification of lakes in alpine regions.

[0004] Secondly, the complex topography and water body development characteristics in alpine areas further increase the difficulty of identification. The terrain in these areas is undulating, resulting in the existence of a large number of shadows. Due to the similarities between water bodies and shadows in spectral characteristics, especially in the visible and infrared bands, it is difficult to effectively distinguish water bodies and shadows in remote sensing images based solely on spectral characteristics. More importantly, most alpine areas have small water bodies, which are often greatly affected by mixed pixels in remote sensing images and are easily overlooked or misclassified. This feature, combined with complex terrain and lighting conditions, makes high-precision identification more difficult.

[0005] To overcome these challenges, existing technologies have introduced local texture features to eliminate shadow interference, but this method faces inherent limitations. Texture information is highly dependent on the lighting conditions in the image, and the shadow area is precisely caused by insufficient lighting, which means that in the shadow area, the texture features may be weakened or even disappear. In addition, due to its unique reflective properties, water bodies also present smoother texture features in some cases, further blurring the boundary between water bodies and shadows. For small water bodies, their feature information in the image is weak, which makes it twice as difficult to distinguish.

[0006] At present, widely used methods in the field of lake extraction, such as index methods and machine learning methods based on local spectral and texture information, face serious limitations in alpine areas. These methods often have difficulty in accurately distinguishing water bodies, shadows, and seasonal meltwater in complex terrain, resulting in a large number of shadows and seasonal meltwater being incorrectly classified as lakes. This not only reduces the accuracy of the extraction results, but also often requires tedious post-processing work, greatly increasing the difficulty and workload of remote sensing lake monitoring in alpine areas.

[0007] In recent years, deep learning semantic segmentation methods, especially the most advanced Transformer model, have shown some potential in eliminating shadow interference. Deep learning can extract multi-scale environmental information through hierarchical convolution, while the self-attention mechanism of Transformer enables it to capture long-range dependencies, that is, a wider range of contextual information. These characteristics enable deep learning models to perform well in identifying large water bodies with regular shapes, clear shorelines and other obvious features, and can better distinguish them from large areas of shadows. However, deep learning methods focus more on global features rather than local details, resulting in insufficient refinement in lake boundary extraction. For small lakes that are common in alpine areas, the recognition effect of deep learning models is often unsatisfactory because they occupy fewer pixels in remote sensing images and have unclear features.

[0008] In summary, although machine learning and deep learning technologies have made significant progress in the field of water body identification, accurate identification of lakes in alpine regions is still one of the most challenging tasks facing remote sensing science. The limitations of existing technologies are mainly reflected in the inadequacies in feature extraction and characterization. Therefore, improving the accuracy of feature input and enhancing the expression of water body information have become key issues in breaking through the bottleneck of lake identification accuracy in alpine regions.

[0009] In view of this, a method and device for identifying mountain lakes based on environmental characteristics are provided to solve the technical problem that existing methods are limited by the information representation capabilities of traditional local spectral and texture features and are difficult to accurately identify small water bodies in alpine areas. Summary of the invention

[0010] To this end, the embodiments of the present invention provide a method and device for identifying mountain lakes based on environmental characteristics, in order to solve the technical problem that existing methods are limited by the information representation capabilities of traditional local spectral and texture features, making it difficult to accurately identify small water bodies in high-altitude cold areas, thereby improving the remote sensing identification accuracy of lakes in high-altitude cold areas.

[0011] In order to achieve the above purpose, the embodiment of the present invention provides the following technical solutions:

[0012] The present invention provides a method for identifying mountain lakes based on environmental characteristics, the method comprising:

[0013] Acquire multiple remote sensing images of the target water body;

[0014] Based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image;

[0015] Calculating terrain data based on the spectral enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting a remote sensing image of the buffer zone;

[0016] Inputting the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model;

[0017] The water body classification model is obtained by training a pre-built convolutional neural network using terrain data, historical image data and category labels of water body samples.

[0018] In some embodiments, based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body, specifically including:

[0019] Select images of the period without ice and snow cover from the acquired remote sensing images to obtain the target images;

[0020] Calculating water indexes of multiple dimensions of the target image;

[0021] By converting the RGB to HSV color space, the water body indices of different dimensions are combined to generate a comprehensive water body index to enhance the spectral characteristics of the water body.

[0022] In some embodiments, enhancing the spectral characteristics of the water body further comprises:

[0023] The water body area is extracted from the spectrally enhanced image by using a remote sensing index algorithm or a mask generation algorithm.

[0024] In some embodiments, calculating terrain data based on the spectrally enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting the remote sensing image of the buffer zone specifically includes:

[0025] Extracting a water body region on the spectrally enhanced image, and identifying and marking water body units in the water body region;

[0026] The spectrally enhanced images at different times are stitched together to obtain a composite image;

[0027] Calculating terrain data of the synthetic image and fusing the obtained terrain data with data of the synthetic image;

[0028] constructing a buffer zone for each of the water body units;

[0029] The water environment image of the buffer zone is extracted, and the water environment image of the buffer zone is resampled to obtain water environment characteristic parameters.

[0030] In some embodiments, calculating the terrain data of the synthetic image specifically includes:

[0031] The slope information, elevation and azimuth of each pixel in the synthetic image are calculated.

[0032] In some embodiments, the pre-built convolutional neural network is trained using the topographic data, historical image data and category labels of the water body samples to obtain the water body classification model, which specifically includes:

[0033] Obtain terrain data, historical image data, and category labels of water samples to construct a dataset;

[0034] Dividing the data set into a training set, a validation set, and a test set;

[0035] Based on a preset loss function, the pre-built convolutional neural network is trained using the training set to obtain an initial model;

[0036] The initial model is iteratively trained using the validation set to obtain the optimized water body classification model.

[0037] In some embodiments, the network architecture of the pre-built convolutional neural network includes:

[0038] Convolutional layer, which is used to extract spatial features of the input image and deepen the network's ability to identify water bodies, rivers and non-water bodies through layer-by-layer feature extraction;

[0039] A pooling layer, which is used to reduce the size of the feature map and retain the main features of the input image;

[0040] Activation function: A nonlinear activation function is used after each convolution layer to capture additional features.

[0041] The classification layer, at the end of the network, uses a fully connected layer for feature fusion, and finally outputs the classification probability of lakes, rivers, and non-water bodies through the softmax function.

[0042] The present invention also provides a device for identifying mountain lakes based on environmental characteristics, the device comprising:

[0043] A remote sensing image acquisition unit, used to acquire multiple remote sensing images of a target water body;

[0044] A spectral information enhancement unit, used to calculate a comprehensive water body index based on the remote sensing image to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image;

[0045] An environmental information extraction unit, used to calculate terrain data based on the spectral enhanced image, calculate the buffer zone of the target water body according to the terrain data, and extract a remote sensing image of the buffer zone;

[0046] A recognition result generating unit, used for inputting the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model;

[0047] The water body classification model is obtained by training a pre-built convolutional neural network using terrain data, historical image data and category labels of water body samples.

[0048] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0049] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the above method when executed by a processor.

[0050] The method and device for identifying mountain lakes based on environmental characteristics provided by the present invention obtain multiple remote sensing images of the target water body; based on the remote sensing images, calculate the comprehensive water body index to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image; calculate terrain data based on the spectrally enhanced image, calculate the buffer zone of the target water body according to the terrain data, and extract the remote sensing image of the buffer zone; input the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model, and obtain the category recognition result output by the water body classification model. This solves the technical problem that the existing methods are limited by the information representation capabilities of traditional local spectra and texture features, making it difficult to accurately identify small water bodies in high-cold areas, thereby improving the remote sensing recognition accuracy of lakes in high-cold areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0052] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0053] Figure 1 One of the flow charts of the method for identifying mountain lakes based on environmental characteristics provided by the present invention;

[0054] Figure 2 The second flowchart of the method for identifying mountain lakes based on environmental characteristics provided by the present invention;

[0055] Figure 3 The third flowchart of the method for identifying mountain lakes based on environmental characteristics provided by the present invention;

[0056] Figure 4 The fourth flowchart of the method for identifying mountain lakes based on environmental characteristics provided by the present invention;

[0057] Figure 5 This is a schematic diagram of the principle of water unit extraction;

[0058] Figure 6 Schematic diagram of the principle affecting resolution calculation;

[0059] Figure 7 This is the ConvNet architecture diagram;

[0060] Figure 8 Flowchart of the water classification model training process;

[0061] Fig. 9 A schematic diagram of the structure of a mountain lake identification device based on environmental characteristics provided by the present invention;

[0062] Fig.10 This is a structural block diagram of a computer device provided by the present invention. DETAILED DESCRIPTION

[0063] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] At present, the main methods widely used for lake extraction in alpine areas are index method, machine learning and deep learning methods. These methods are limited by the information representation ability of traditional local spectral and texture features, and it is difficult to accurately identify small water bodies in alpine areas. The key to solving this problem lies in how to effectively capture and utilize the unique characteristics of water bodies in complex alpine environments. Specifically, the following aspects need to be focused on:

[0065] Refinement of feature representation: Traditional methods often directly use raw spectra or simple indices as input, which is difficult to fully express the characteristics of water bodies in complex environments in mountainous areas. This method intends to enhance the spectral characteristics of water bodies through multi-index synthesis, retain richer texture information, and distinguish water bodies from similar objects such as shadows;

[0066] Effective use of contextual information: Although conventional deep learning models can learn complex features, they often lack explicit consideration of the local environment. This method intends to establish a water body buffer zone and explicitly introduce environmental information around the water body to provide the model with more valuable contextual clues to facilitate the identification of small water bodies;

[0067] Scale adaptability: Traditional machine learning methods perform differently when dealing with water bodies of different scales, especially when identifying small water bodies. This method intends to improve the model's ability to identify water bodies of various scales, especially the detection accuracy of small water bodies, by combining enhanced spectral information and local environmental characteristics and designing the level of detail of input features according to the size of the water body;

[0068] Anti-interference ability: The complex terrain and lighting conditions in mountainous areas often cause traditional methods to mistake shadows or seasonal meltwater for lakes. This method comprehensively considers multi-source features and environmental information (multi-environmental features: shape features, shoreline features, slope around the lake, etc.; multi-spectral features: water body index, vegetation index, etc.) to improve the model's anti-interference ability and recognition accuracy.

[0069] In view of this, the present invention provides a method for identifying mountain lakes based on environmental characteristics, which improves the characterization ability of water bodies by enhancing the water body spectrum and environmental information, realizes high-precision identification of water bodies, and further promotes the application of remote sensing science in extreme environments, providing more reliable data support for water resources monitoring, ecological environment assessment and climate change research in high-cold areas.

[0070] In a specific embodiment, Figure 1 As shown, the method for identifying mountain lakes based on environmental characteristics provided by the present invention comprises the following steps:

[0071] S110: Acquire multiple remote sensing images of the target water body;

[0072] S120: Based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image; in a specific usage scenario, such as Figure 2 As shown in the figure, when enhancing the spectral information of water bodies, firstly, images of the period without ice and snow cover are selected to calculate water body indexes such as MNDWI, NDVI, and NDWI, and then different water body indexes are combined through the conversion from RGB to HSV color space to synthesize multiple new comprehensive water body indexes to enhance the spectral characteristics of water bodies;

[0073] S130: Calculate terrain data based on the spectral enhanced image, calculate the buffer zone of the target water body according to the terrain data, and extract the remote sensing image of the buffer zone; Figure 2 Taking the scenario shown in the figure as an example, in the process of determining the scope of water body environmental information, the water body scope is first extracted based on the spectral information of remote sensing images to establish a water body buffer zone; at the same time, multi-temporal images are synthesized, and terrain parameters are calculated to construct multi-source feature parameters; then the features within the water body buffer zone are extracted and resampled according to the size of the water body to ensure that the detail of the feature information is appropriate;

[0074] S140: Inputting the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model; wherein the water body classification model is obtained by training a pre-built convolutional neural network using the terrain data, historical image data and category labels of the water body samples. Figure 2 As shown in the figure, in this specific usage scenario, during the training process of the water body classification model, three label samples of lakes, rivers, and non-water bodies are first constructed. Based on the ConvNet classification method and the synthesized high-quality multispectral images, comprehensive water body index, and terrain parameters, a water body classification model is constructed to classify water body units, ultimately achieving accurate identification of alpine lakes.

[0075] In this way, the method provided by the present invention breaks through the limitation of traditional machine learning and deep learning methods that only rely on enhanced spectral information to improve water body characterization capabilities. By establishing a water body buffer zone and integrating multiple water body indices, it enhances spectral information and local environmental characteristics, improves the utilization efficiency of contextual information, and is more suitable for small water body surface tasks in complex environments in high-altitude and cold regions.

[0076] In some embodiments, Figure 3 As shown, based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body, specifically including the following steps:

[0077] S310: Selecting images of a period without ice and snow coverage from the acquired remote sensing images to obtain a target image;

[0078] S320: Calculate the water index of multiple dimensions of the target image; specifically, including calculating the following indexes:

[0079] a) MNDWI (Modified Normalized Difference Water Index):

[0080] MNDWI=(Green-SWI R) / (Green+SWIR)

[0081] Among them, Green is the green light band, and SWI R is the short-wave infrared band;

[0082] b) NDVI (Normalized Difference Vegetation Index):

[0083] NDVI=(NIR-Red) / (NIR+Red)

[0084] Among them, NIR is the near infrared band, and Red is the red light band;

[0085] c) NDWI (Normalized Difference Water Index):

[0086] NDWI=(Green-NIR) / (Green+NIR)

[0087] S330: By converting the RGB to HSV color space, the water body indexes of different dimensions are combined to generate a comprehensive water body index to enhance the spectral characteristics of the water body. Specifically, the calculated MNDWI, NDVI and NDWI are assigned to the three channels of R, G and B respectively. At this time, a false color composite image is obtained, in which the water body will present a specific color; then, the RGB color space is converted to the HSV (Hue, Saturation, Value) color space to form a new index. The HSV color space is closer to the way humans perceive color, which helps to better distinguish and enhance specific features. This comprehensive index (Hue, Saturation, Value) will integrate the information of MNDWI, NDVI and NDWI, while taking advantage of the enhancement effect brought by the color space conversion.

[0088] Further, the spectral characteristics of water bodies are enhanced, and then the following are included:

[0089] The water body area is extracted from the spectral enhanced image by using a remote sensing index algorithm or a mask generation algorithm. That is, the water body area is preliminarily identified and extracted through pixel-level analysis, which includes:

[0090] Remote sensing index calculation: Calculate commonly used water body indices (such as normalized difference water index NDWI, improved NDWI, ratio index, etc.) to enhance the contrast between water bodies and surrounding objects. The index method distinguishes water bodies from non-water bodies by calculating reflectivity or reflectivity;

[0091] Mask generation: Water masks can use exponential or machine learning methods, depending on the size and complexity of the area. The exponential method is suitable for small areas with obvious spectral features, which is efficient and easy to implement; while the machine learning method is suitable for water extraction in large areas or complex environments, and can handle more features and background interference, provide higher extraction accuracy, generate binary water masks, mark water areas, and further help filter out background noise or irrelevant information.

[0092] It should be understood that in step S310, selecting suitable remote sensing images is the basis for ensuring the accuracy of water body identification. The image source can be images obtained from various remote sensing satellites, such as Landsat, Sentinel, etc. Selecting higher resolution images helps to more accurately identify small lakes and complex lake boundaries. Image types can be selected according to needs, such as optical remote sensing images, synthetic aperture radar (SAR) images, etc. Optical images are suitable for spectral feature analysis of water bodies, while SAR images are insensitive to clouds and weather conditions and can be used to supplement the deficiencies of optical images. When selecting phases, select phase images of months without ice and snow interference. Ice and snow coverage may obscure the true boundaries of water bodies and affect the spectral features of water bodies.

[0093] After selecting the image, you can also pre-process the image to improve image quality, eliminate interference that affects analysis, and ensure data consistency and accuracy. Pre-processing can include the following dimensions:

[0094] Geometric correction: Perform geometric correction on the image to eliminate geometric distortion caused by image acquisition angle or earth curvature. Geometric correction ensures that the objects in the image are consistent with the actual ground position.

[0095] Atmospheric correction: Perform atmospheric correction on the image to reduce the impact of atmospheric scattering and absorption on spectral information. Atmospheric correction can improve the recognizability of water bodies and other ground features;

[0096] Radiometric correction: Perform radiometric correction on images to standardize the radiometric characteristics of images acquired by different sensors or at different times. Radiometric correction helps ensure data consistency between different images.

[0097] In some embodiments, Figure 4 As shown, calculating terrain data based on the spectral enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting the remote sensing image of the buffer zone specifically includes the following steps:

[0098] S410: extracting a water body region on the spectral enhanced image, and identifying and marking water body units in the water body region;

[0099] S420: splicing the spectral enhanced images at different times to obtain a composite image;

[0100] S430: Calculating terrain data of the synthetic image (for example, calculating slope information, elevation and azimuth of each pixel in the synthetic image), and fusing the obtained terrain data with data of the synthetic image;

[0101] S440: constructing a buffer zone for each of the water body units;

[0102] S450: extracting the water environment image of the buffer zone, and resampling the water environment image of the buffer zone to obtain water environment characteristic parameters.

[0103] exist Figure 2 In the specific usage scenario shown, in the identification of alpine lakes, water environment information extraction involves multiple key steps to accurately extract water bodies and analyze their environmental characteristics. These steps include multi-temporal image synthesis, terrain parameter calculation, and water environment image extraction and image resolution calculation. Specifically, water environment information extraction includes the following steps:

[0104] S1: Water unit extraction, such as Figure 5 As shown in the figure, the initially extracted water area is further processed to identify and separate independent water units; water unit extraction specifically includes connectivity analysis, morphological processing and water unit labeling. In the connectivity analysis, the binary water mask is subjected to connectivity analysis to identify continuous water areas. This step helps to identify interconnected water areas and exclude small noise areas. In the morphological processing, morphological operations (such as dilation, corrosion, opening and closing operations, etc.) are applied to optimize the water body boundary and repair boundary missing or noise. Morphological processing helps to improve the smoothness and continuity of the water body boundary. In the water unit labeling, the connected water area is labeled to distinguish different water units (such as lakes, rivers, non-water bodies, etc.). Each water unit can be assigned a unique identifier to form a training sample for the subsequent construction of the training model.

[0105] S2: Multi-temporal image synthesis: By synthesizing multi-temporal images, the clarity and accuracy of images can be improved, the interference of clouds, snow, ice and shadows can be reduced, and a better basis for water body extraction and environmental analysis can be provided. When synthesizing multi-temporal images, it includes image selection, phase selection (selecting image periods without cloud or snow cover to improve image clarity), cloud filtering and cloud mask, cloud detection (using cloud detection algorithms such as band-based cloud detection and cloud mask generation methods to identify cloud and cloud shadow areas in images. Common methods include threshold-based cloud detection, machine learning methods or existing cloud mask data), cloud mask (generate cloud masks to mask cloud and cloud shadow areas in images for subsequent processing. Cloud masks can be synthesized through cloud detection results of multi-temporal images to ensure that the masking of cloud areas is as comprehensive as possible), image synthesis, image stitching (stitching images from different times into a complete large-scale image. Use image fusion technology (such as pixel-based fusion and feature-based fusion) to synthesize clear images and reduce interference from clouds, snow and ice) and clarity enhancement (enhance the synthesized images, such as contrast adjustment and color enhancement, to improve the visibility of environmental details).

[0106] S3: Terrain parameter calculation, by calculating terrain parameters (such as slope) to assist in distinguishing water bodies from mountain shadows and provide key information for the identification of lakes and rivers. Terrain parameter calculation mainly includes the following steps:

[0107] Digital Elevation Model (DEM) Processing:

[0108] DEM acquisition: Get high-resolution DEM data to provide surface elevation information. DEM data can be obtained from remote sensing satellites (such as SRTM, ASTER) or ground measurements.

[0109] DEM preprocessing: Correct and filter DEM data to eliminate noise and errors.

[0110] Calculation of terrain parameters:

[0111] Slope Calculation: Calculates the slope of each pixel in the image. Slope information can help distinguish between hillshade and water bodies. Steeper slopes usually correspond to mountains, while flatter slopes may correspond to water bodies.

[0112]

[0113] The size of its grid cells is Δx and Δy, and the elevation value is z.

[0114] Other terrain features: Other terrain parameters (such as elevation z and azimuth) can be calculated as a supplement to improve the ability to distinguish between water bodies and mountains.

[0115]

[0116] Terrain parameter fusion:

[0117] The calculated terrain parameters (such as slope) are fused with remote sensing image data to enhance the recognition of water bodies and mountains.

[0118] S4: Water environment image extraction and image resolution calculation. In the identification of alpine lakes, water environment image extraction and image resolution calculation of adaptive scale are key steps, such as Figure 6 As shown in the figure, the core of this step is to collect and process water environment information according to the scale of water units, ensuring that small lakes can extract high-precision environmental features while summarizing unnecessary details of large lakes. This differentiated processing can optimize the accuracy and efficiency of lake identification. It specifically includes the following steps:

[0119] Establishment of water body unit buffer: First, the buffer refers to establishing a 20-pixel buffer for each water body unit. The buffer is used to extract the environmental information around the water body unit and reduce the boundary effect.

[0120] Buffer extraction: Use geographic information system (GIS) tools or programming tools such as the geopandas library in Python to create and extract buffers.

[0121] Image extraction and resampling: Extract remote sensing image data from the buffer zone. The extracted area includes water units and their surroundings to provide contextual information. Extract DEM data from the buffer zone and calculate relevant terrain parameters (slope, orientation).

[0122] Image resolution adjustment: Resample the extracted buffer images and DEM data to a uniform resolution, usually 515x515 pixels. This step is to standardize the spatial resolution of the data to ensure consistency of data for different water units. Select an appropriate resampling method (such as bilinear interpolation, cubic convolution interpolation) to maintain the spatial accuracy of the image and DEM data.

[0123] In some embodiments, the pre-built convolutional neural network is trained using the topographic data, historical image data and category labels of the water body samples to obtain the water body classification model, which specifically includes:

[0124] Obtain terrain data, historical image data, and category labels of water samples to construct a dataset;

[0125] Dividing the data set into a training set, a validation set, and a test set;

[0126] Based on a preset loss function, the pre-built convolutional neural network is trained using the training set to obtain an initial model;

[0127] The initial model is iteratively trained using the validation set to obtain the optimized water body classification model.

[0128] Among them, the network architecture of the pre-built convolutional neural network includes:

[0129] Convolutional layer, which is used to extract spatial features of the input image and deepen the network's ability to identify water bodies, rivers and non-water bodies through layer-by-layer feature extraction;

[0130] A pooling layer, which is used to reduce the size of the feature map and retain the main features of the input image;

[0131] Activation function: A nonlinear activation function is used after each convolution layer to capture additional features.

[0132] The classification layer, at the end of the network, uses a fully connected layer for feature fusion, and finally outputs the classification probability of lakes, rivers, and non-water bodies through the softmax function.

[0133] exist Figure 2 In the specific usage scenario shown, accurate identification of lakes is required. Specifically, in the process of identifying alpine lakes, accurate identification of lakes is the final key step. Although water body identification based on pixel-level spectrum can identify the fine contours of water bodies, there are a lot of errors (shadows, ice and snow melt, etc.). This step constructs three types of label samples of lakes, rivers, and non-water bodies, uses the convolutional neural network (ConvNet) classification method, and combines multi-source data (such as synthetic optical remote sensing images, calculated comprehensive water body index, and terrain parameters calculated by DEM) to achieve accurate classification of water body units, and finally achieve the goal of accurate identification of alpine lakes.

[0134] In the process of model training, we first need to construct label samples to generate sample data sets for training and verifying classification models, including three categories: lakes, rivers, and non-water bodies. Specifically, in sample collection, the data source is to extract lake, river, and non-water body samples from known geographic information data, historical image data, and manually annotated data; then class balance is performed to ensure that the number of samples of the three categories is relatively balanced to avoid the deviation caused by class imbalance to model training; samples are manually annotated, and obvious lake areas are annotated as lake samples. Lakes usually have smooth boundaries and stable water surface features; when selecting river samples, pay attention to distinguishing the linear morphology and water body characteristics of rivers; non-water body samples include categories such as mountains, vegetation, and bare land. These areas are annotated as non-water body samples to help the model distinguish water bodies from other surface features; after collecting samples, the samples are expanded, specifically by expanding the sample set through data enhancement techniques such as rotation, scaling, and flipping to improve the robustness and generalization ability of the model; finally, multi-temporal samples are integrated, and multi-temporal image samples are used to ensure that the model can adapt to water body recognition under different seasons and lighting conditions.

[0135] When building the ConvNet classification model, a classification model is designed based on a convolutional neural network (ConvNet), and the image and terrain data are used as input to classify water units. In the input data preparation stage, the synthetic optical remote sensing image, terrain parameters (such as slope, elevation, etc.) and environmental information in the water unit buffer are integrated into the input data. Data from different sources can be processed through multi-channel input to ensure the effective fusion of terrain and image information. After data fusion, data standardization is performed to standardize the input data to ensure that the data of each channel (such as image bands, terrain parameters) are processed at the same scale to avoid the influence of numerical differences on model training.

[0136] like Figure 7 As shown in the figure, the ConvNet architecture network layer is a convolutional neural network architecture with moderate depth, including multiple convolutional layers, pooling layers, fully connected layers, and the final softmax classification layer. The specific network depth and structure can be adjusted according to the amount of data and task complexity. Among them, the spatial features of the image are extracted through the convolutional layer, and the network's ability to identify water bodies, rivers, and non-water bodies is deepened through layer-by-layer feature extraction; pooling layers (such as maximum pooling and average pooling) are used to reduce the size of feature maps while retaining the main features to improve the computational efficiency and noise resistance of the model; nonlinear activation functions (such as ReLU) are used after each layer of convolution to increase the expressive power of the model and capture more complex features; the classification layer is set at the end of the network, and a fully connected layer is used for feature fusion, and finally the classification probability of lakes, rivers, and non-water bodies is output through the softmax function.

[0137] During model training, Figure 8As shown in the figure, select a suitable loss function (such as cross entropy loss) to guide the training of the model and minimize the classification error; use optimization algorithms (such as Adam, SGD) to iteratively train the model, and continuously adjust the network weights through back propagation to improve the classification accuracy; adjust hyperparameters such as learning rate, batch size, and number of training rounds to optimize the convergence speed and final performance of the model.

[0138] During the model verification and testing process, the validation set is used to evaluate the model performance in real time during the training process, and the model structure and parameters are adjusted to avoid overfitting; after the model training is completed, an independent test set is used to evaluate the generalization ability of the model and verify its performance in practical applications.

[0139] In the process of water body unit classification, the trained ConvNet classification model is used to classify each water body unit to achieve accurate identification of alpine lakes. When classifying environmental information in the buffer zone, the classification process is to input the image and terrain data of each water body unit and its buffer zone into the trained ConvNet classification model to predict its category (lake, river, non-water body); the classification result is that the model outputs the probability distribution of three categories, and the final category of the water body unit is determined according to the maximum probability value. When the classification results are integrated, the classification results are mapped back to the geographic space and integrated into a complete water body classification layer to display the classification of each water body unit. In the process of lake identification accuracy verification, the accuracy evaluation indicators such as overall accuracy, Kappa coefficient, and F1-score are used to evaluate the accuracy of lake identification and compared with the actual lake distribution for verification. Finally, the sources of classification errors (such as image noise and shadow interference) are analyzed and the model or input data are adjusted to further improve the recognition accuracy.

[0140] In the above specific implementation, the method for identifying mountain lakes based on environmental characteristics provided by the present invention obtains multiple remote sensing images of the target water body; based on the remote sensing images, calculates the comprehensive water body index to enhance the spectral characteristics of the water body and obtains a spectrally enhanced image; calculates terrain data based on the spectrally enhanced image, calculates the buffer zone of the target water body according to the terrain data, and extracts the remote sensing image of the buffer zone; inputs the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model, and obtains the category recognition result output by the water body classification model. This solves the technical problem that the existing methods are limited by the information representation capabilities of traditional local spectra and texture features, making it difficult to accurately identify small water bodies in high-cold areas, thereby improving the remote sensing recognition accuracy of lakes in high-cold areas.

[0141] In some embodiments, the method provided by the present invention enhances water body information from multiple levels of spectral and environmental information through a multi-level water body information enhancement strategy. After multi-index fusion, a water body buffer zone is established, and rich multi-source feature parameters are constructed by combining multi-temporal image synthesis and terrain parameter calculation, thereby effectively taking the local environmental information of the water body into consideration and improving the model's ability to recognize complex terrain and small water bodies. This multi-level information enhancement strategy breaks through the limitations of traditional methods that rely only on single spectral information, and provides a more comprehensive and richer feature representation for water body identification.

[0142] In some embodiments, the method provided by the present invention performs adaptive feature resampling to adaptively resample the extracted features according to the size of the water body, which is particularly suitable for the identification of small water bodies in alpine areas. For small water bodies, high-resolution fine features are retained, while for large water bodies, detailed information is appropriately summarized to highlight the overall structural features; this differentiated processing strategy ensures that the detail of the feature information matches the scale of the water body, and especially strengthens the detailed description of small water bodies. In alpine areas, small glacial lakes and other tiny water bodies are common but difficult to identify. This adaptive resampling method significantly improves the model's sensitivity to these small water bodies; by retaining the high-precision features of the environment surrounding the small water body, this method effectively overcomes the limitations of traditional recognition algorithms in dealing with small water bodies in complex terrain in alpine areas, and greatly improves the detection accuracy of small water bodies.

[0143] In some embodiments, the method provided by the present invention designs a classification model for lakes in high-altitude cold regions, and constructs a classification model including three categories of lakes, rivers, and non-water bodies, rather than a simple lake / non-lake binary classification, which is more in line with the complex characteristics of lakes in high-altitude cold regions and is conducive to improving the accuracy of identification; using deep learning technology, combined with enhanced multispectral images, comprehensive water body index and terrain parameters, a classification model is constructed that can fully utilize multi-source information, fully considering the special needs of lake identification in high-altitude cold regions, and through adaptive feature processing and targeted model design, the adaptability and accuracy of the method in complex environments are improved.

[0144] In addition to the above method, the present invention also provides a mountain lake identification device based on environmental characteristics, such as Fig. 9 As shown, the device comprises:

[0145] A remote sensing image acquisition unit 910 is used to acquire multiple remote sensing images of a target water body;

[0146] A spectral information enhancement unit 920 is used to calculate a comprehensive water body index based on the remote sensing image to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image;

[0147] An environmental information extraction unit 930 is used to calculate terrain data based on the spectral enhancement image, calculate the buffer zone of the target water body according to the terrain data, and extract a remote sensing image of the buffer zone;

[0148] The recognition result generating unit 940 is used to input the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model;

[0149] The water body classification model is obtained by training a pre-built convolutional neural network using terrain data, historical image data and category labels of water body samples.

[0150] In some embodiments, based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body, specifically including:

[0151] Select images of the period without ice and snow cover from the acquired remote sensing images to obtain the target images;

[0152] Calculating water indexes of multiple dimensions of the target image;

[0153] By converting the RGB to HSV color space, the water body indices of different dimensions are combined to generate a comprehensive water body index to enhance the spectral characteristics of the water body.

[0154] In some embodiments, enhancing the spectral characteristics of the water body further comprises:

[0155] The water body area is extracted from the spectrally enhanced image by using a remote sensing index algorithm or a mask generation algorithm.

[0156] In some embodiments, calculating terrain data based on the spectrally enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting the remote sensing image of the buffer zone specifically includes:

[0157] Extracting a water body region on the spectrally enhanced image, and identifying and marking water body units in the water body region;

[0158] The spectrally enhanced images at different times are stitched together to obtain a composite image;

[0159] Calculating terrain data of the synthetic image and fusing the obtained terrain data with data of the synthetic image;

[0160] constructing a buffer zone for each of the water body units;

[0161] The water environment image of the buffer zone is extracted, and the water environment image of the buffer zone is resampled to obtain water environment characteristic parameters.

[0162] In some embodiments, calculating the terrain data of the synthetic image specifically includes:

[0163] The slope information, elevation and azimuth of each pixel in the synthetic image are calculated.

[0164] In some embodiments, the pre-built convolutional neural network is trained using the topographic data, historical image data and category labels of the water body samples to obtain the water body classification model, which specifically includes:

[0165] Obtain terrain data, historical image data, and category labels of water samples to construct a dataset;

[0166] Dividing the data set into a training set, a validation set, and a test set;

[0167] Based on a preset loss function, the pre-built convolutional neural network is trained using the training set to obtain an initial model;

[0168] The initial model is iteratively trained using the validation set to obtain the optimized water body classification model.

[0169] In some embodiments, the network architecture of the pre-built convolutional neural network includes:

[0170] Convolutional layer, which is used to extract spatial features of the input image and deepen the network's ability to identify water bodies, rivers and non-water bodies through layer-by-layer feature extraction;

[0171] A pooling layer, which is used to reduce the size of the feature map and retain the main features of the input image;

[0172] Activation function: A nonlinear activation function is used after each convolution layer to capture additional features.

[0173] The classification layer, at the end of the network, uses a fully connected layer for feature fusion, and finally outputs the classification probability of lakes, rivers, and non-water bodies through the softmax function.

[0174] In the above specific implementation, the mountain lake identification device based on environmental characteristics provided by the present invention obtains multiple remote sensing images of the target water body; based on the remote sensing image, calculates the comprehensive water body index to enhance the spectral characteristics of the water body and obtains a spectrally enhanced image; calculates terrain data based on the spectrally enhanced image, calculates the buffer zone of the target water body according to the terrain data, and extracts the remote sensing image of the buffer zone; inputs the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model, and obtains the category recognition result output by the water body classification model. This solves the technical problem that the existing methods are limited by the information representation capabilities of traditional local spectra and texture features, making it difficult to accurately identify small water bodies in high-cold areas, thereby improving the remote sensing recognition accuracy of lakes in high-cold areas.

[0175] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a model prediction. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The model prediction of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0176] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0177] Corresponding to the above embodiment, the embodiment of the present invention further provides a computer storage medium, which contains one or more program instructions, wherein the one or more program instructions are used to execute the above method.

[0178] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the above method.

[0179] In the embodiment of the present invention, the processor may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0180] The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware.

[0181] The storage medium may be a memory, which may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0182] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable ROM (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0183] The volatile memory may be a random access memory (RAM) which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DRRAM).

[0184] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0185] Those skilled in the art will appreciate that in one or more of the above examples, the functions described in the present invention can be implemented using a combination of hardware and software. When software is used, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any media that facilitates the transmission of computer programs from one place to another. Storage media can be any available media that can be accessed by general or special-purpose computers.

[0186] The above specific implementation methods further illustrate the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying mountain lakes based on environmental characteristics, characterized in that: The method comprises: Acquire multiple remote sensing images of the target water body; Based on the remote sensing image, a comprehensive water body index is calculated to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image; Calculating terrain data based on the spectral enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting a remote sensing image of the buffer zone; Inputting the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model; The water body classification model is obtained by training a pre-built convolutional neural network using terrain data, historical image data and category labels of water body samples.

2. The method for identifying mountain lakes based on environmental characteristics according to claim 1, characterized in that: Based on the remote sensing image, a comprehensive water index is calculated to enhance the spectral characteristics of the water body, specifically including: Select images of the period without ice and snow cover from the acquired remote sensing images to obtain the target images; Calculating water indexes of multiple dimensions of the target image; By converting the RGB to HSV color space, the water body indices of different dimensions are combined to generate a comprehensive water body index to enhance the spectral characteristics of the water body.

3. The method for identifying mountain lakes based on environmental characteristics according to claim 2, characterized in that: Enhance the spectral characteristics of water bodies, and then include: The water body area is extracted from the spectrally enhanced image by using a remote sensing index algorithm or a mask generation algorithm.

4. The method for identifying mountain lakes based on environmental characteristics according to claim 1, characterized in that: Calculating terrain data based on the spectral enhanced image, calculating the buffer zone of the target water body according to the terrain data, and extracting the remote sensing image of the buffer zone, specifically includes: Extracting a water body region on the spectrally enhanced image, and identifying and marking water body units in the water body region; The spectrally enhanced images at different times are stitched together to obtain a composite image; Calculating terrain data of the synthetic image and fusing the obtained terrain data with data of the synthetic image; constructing a buffer zone for each of the water body units; The water environment image of the buffer zone is extracted, and the water environment image of the buffer zone is resampled to obtain water environment characteristic parameters.

5. The method for identifying mountain lakes based on environmental characteristics according to claim 4, characterized in that: Calculating terrain data of the synthetic image, specifically comprising: The slope information, elevation and azimuth of each pixel in the synthetic image are calculated.

6. The method for identifying mountain lakes based on environmental characteristics according to claim 1, characterized in that: The pre-built convolutional neural network is trained using the topographic data, historical image data and category labels of the water body samples to obtain the water body classification model, which specifically includes: Obtain terrain data, historical image data, and category labels of water samples to construct a dataset; Dividing the data set into a training set, a validation set, and a test set; Based on a preset loss function, the pre-built convolutional neural network is trained using the training set to obtain an initial model; The initial model is iteratively trained using the validation set to obtain the optimized water body classification model.

7. The method for identifying mountain lakes based on environmental characteristics according to claim 6, characterized in that: The network architectures of the pre-built convolutional neural networks include: Convolutional layer, which is used to extract spatial features of the input image and deepen the network's ability to identify water bodies, rivers and non-water bodies through layer-by-layer feature extraction; A pooling layer, which is used to reduce the size of the feature map and retain the main features of the input image; Activation function: A nonlinear activation function is used after each convolution layer to capture additional features. The classification layer,at the end of the network, uses a fully connected layer for feature fusion,and finally outputs the classification probabilities of lakes, rivers, and non-water bodies through the softmax function.

8. A device for identifying mountain lakes based on environmental characteristics, characterized in that: The device comprises: A remote sensing image acquisition unit, used to acquire multiple remote sensing images of a target water body; A spectral information enhancement unit, used to calculate a comprehensive water body index based on the remote sensing image to enhance the spectral characteristics of the water body and obtain a spectrally enhanced image; An environmental information extraction unit, used to calculate terrain data based on the spectral enhanced image, calculate the buffer zone of the target water body according to the terrain data, and extract a remote sensing image of the buffer zone; A recognition result generating unit, used for inputting the remote sensing image of the target water body, the remote sensing image of the buffer zone and the terrain data into a pre-trained water body classification model to obtain a category recognition result output by the water body classification model; The water body classification model is obtained by training a pre-built convolutional neural network using terrain data, historical image data and category labels of water body samples.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images