Outdoor ski area identification method

Abstract

The application relates to an outdoor snowfield area identification method, which determines a potential snowfield area of a target area, then acquires multispectral remote sensing data of the area, analyzes the multispectral remote sensing data by using a preset deep learning model, deeply mines spatial and spectral characteristics of snow, and accurately determines a snow-covered area. Then, road data of the potential snowfield area is acquired, and a road accessible area in the snow-covered area is further screened out by using the road data, and finally, an outdoor snowfield area of the target area is obtained according to the road accessible area. The method discards a traditional single artificial measurement or a remote sensing threshold segmentation method based on a single spectral feature, adopts a combination of multi-source data fusion and a deep learning model, and improves the accuracy and practicability of identification.

Description

Technical Field

[0001] This application relates to the field of snowfield monitoring technology, and in particular to a method for identifying the area of ​​an outdoor snowfield. Background Technology

[0002] The area of ​​outdoor ski resorts serves as a core foundation for ski resort planning, snow resource monitoring, and the operation of the ice and snow industry. Accurately obtaining this data is of paramount importance for ski resort construction layout, operation scheduling, and dynamic management of snow resources.

[0003] Currently, there are two main methods for identifying the area of ​​outdoor ski resorts. The first is manual on-site measurement. While this method can obtain ski resort area data to some extent, it is extremely inefficient and requires significant manpower and resources, resulting in high costs. More importantly, manual measurement is severely limited by terrain and weather conditions. In complex terrain or harsh weather, measurement work is difficult to carry out smoothly, making large-scale, dynamic monitoring impossible and failing to meet the real-time, comprehensive data requirements of modern ski resort management. The second method is threshold segmentation based on remote sensing imagery. This method distinguishes between snow-covered and non-snow-covered areas solely through single spectral features. However, in practical applications, this method is highly susceptible to interference from clouds, mountain shadows, and bare ground. These interferences share similar spectral features with snow, making it difficult for traditional methods to accurately extract the spatial and spectral features of snow, leading to low accuracy in snow area identification and difficulty in accurately defining the actual usable area of ​​the ski resort. This poses numerous challenges to the scientific planning and effective operation of ski resorts. Therefore, improving the accuracy of outdoor ski resort identification has become a critical issue that urgently needs to be addressed. Summary of the Invention

[0004] Based on this, the purpose of this application is to provide a method for identifying the area of ​​an outdoor snowfield, thereby improving the accuracy of outdoor snowfield area identification and achieving efficient, accurate and dynamically monitorable snowfield area identification.

[0005] The outdoor snowfield area identification method of this application includes the following steps:

[0006] Identify potential ski resort areas within the target region; Acquire multispectral remote sensing data of the potential snowfield area; analyze the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area; Obtain road data for the potential ski resort area; based on the road data, determine the road-accessible areas within the snow-covered area; The outdoor snowfield area of ​​the target area is obtained based on the road accessibility area.

[0007] This application utilizes a pre-defined deep learning model to analyze multispectral remote sensing data of potential snowfield areas. Leveraging the feature extraction and classification capabilities of the deep learning model, it fully exploits the unique spatial and spectral characteristics of snow cover in the multispectral data. This effectively overcomes the limitations of traditional remote sensing threshold segmentation methods, which rely solely on single spectral features and are susceptible to interference from clouds, mountain shadows, and bare ground. This significantly improves the accuracy of snow-covered area identification, laying a solid foundation for precisely determining the snowfield area. Secondly, road data from potential snowfield areas is introduced to determine the road-accessible areas within the snow-covered region. This innovative step fully considers the crucial impact of traffic accessibility on the usable area of ​​the snowfield in actual operation. Only road-accessible areas are truly exploitable parts of the snowfield. This method of selection ensures that the final identified outdoor snowfield area more closely reflects actual usability, avoiding the problems caused by overly large or small identification ranges that hinder snowfield planning, operation scheduling, and dynamic snow resource management. Furthermore, the entire technical solution is based on remote sensing data and model analysis, overcoming the drawbacks of low efficiency, high cost, and limitation by terrain and climate conditions associated with manual on-site measurement. It can identify the area of ​​a large snowfield in a short time and achieve dynamic monitoring, reflecting the changes in the snowfield area over time and the environment. This provides real-time and comprehensive data support for snowfield managers, helping them make scientific and reasonable decisions, improve the operational efficiency and management level of the snowfield, and has broad application prospects and significant economic and social value in the ice and snow industry.

[0008] To better understand and implement this application, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the outdoor snowfield area identification method according to an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. Wherein, when the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements.

[0011] It should be understood that the embodiments described below do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0012] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. Furthermore, in the description of this application, unless otherwise stated, “a plurality” means two or more. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items, for example, A and / or B, which can represent: A alone, A and B together, and B alone; the character “ / ” generally indicates that the preceding and following objects are in an “or” relationship.

[0013] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms, and these terms are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. Depending on the context, the word "if" as used in this application can be interpreted as "when," "when," or "in response to determination."

[0014] This application provides a method for identifying the area of ​​outdoor snowfields, which improves the accuracy of outdoor snowfield area identification and achieves efficient, accurate and dynamically monitorable snowfield area identification.

[0015] Please refer to Figure 1 The outdoor snowfield area identification method of this application includes the following steps: S101: Obtain potential ski resort areas within the target region; S102: Acquire multispectral remote sensing data of the potential snowfield area; analyze the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area; S103: Obtain road data for the potential ski resort area; determine the accessible areas within the snow-covered area based on the road data; S104: Based on the accessible area of ​​the road, obtain the outdoor snowfield area of ​​the target area.

[0016] This application utilizes a pre-defined deep learning model to analyze multispectral remote sensing data of potential snowfield areas. Leveraging the feature extraction and classification capabilities of the deep learning model, it fully exploits the unique spatial and spectral characteristics of snow cover in the multispectral data. This effectively overcomes the limitations of traditional remote sensing threshold segmentation methods, which rely solely on single spectral features and are susceptible to interference from clouds, mountain shadows, and bare ground. This significantly improves the accuracy of snow-covered area identification, laying a solid foundation for precisely determining the snowfield area. Secondly, road data from potential snowfield areas is introduced to determine the road-accessible areas within the snow-covered region. This innovative step fully considers the crucial impact of traffic accessibility on the usable area of ​​the snowfield in actual operation. Only road-accessible areas are truly exploitable parts of the snowfield. This method of selection ensures that the final identified outdoor snowfield area more closely reflects actual usability, avoiding the problems caused by overly large or small identification ranges that hinder snowfield planning, operation scheduling, and dynamic snow resource management. Furthermore, the entire technical solution is based on remote sensing data and model analysis, overcoming the drawbacks of low efficiency, high cost, and limitation by terrain and climate conditions associated with manual on-site measurement. It can identify the area of ​​a large snowfield in a short time and achieve dynamic monitoring, reflecting the changes in the snowfield area over time and the environment. This provides real-time and comprehensive data support for snowfield managers, helping them make scientific and reasonable decisions, improve the operational efficiency and management level of the snowfield, and has broad application prospects and significant economic and social value in the ice and snow industry.

[0017] The outdoor snowfield area identification method of this application uses a computer as the execution subject, and the following describes each step in detail.

[0018] For step S101, obtain the potential ski resort area of ​​the target area.

[0019] The target area refers to the specific geographical range within which the outdoor ski resort area is to be identified. This range can be defined according to actual needs, such as the mountainous area where a large ski resort is located, or the area around a specific city suitable for skiing.

[0020] Potential ski area is a preliminary delineation within a target area based on prior information such as topography, climate, and historical ski area distribution, indicating areas where ski areas may potentially exist. For example, in mountainous regions, areas with high altitude, suitable slopes, and significant winter snowfall may be designated as potential ski area.

[0021] By analyzing and comprehensively assessing prior information such as topography (e.g., altitude, slope), climate conditions (e.g., average annual snowfall, winter temperature), and historical distribution of ski resorts in the target area, a preliminary area for potential ski resorts can be identified. For example, if a mountainous area has an altitude between 1500-2500 meters, a slope between 10°-30°, and has historically had multiple ski resorts, then that area could potentially be identified as a ski resort region.

[0022] In one embodiment, step S101, obtaining the potential ski resort area of ​​the target area, includes: Step S1011: Obtain ski resort distribution data for the target area; the ski resort distribution data includes several geographical coordinate points corresponding to each ski resort.

[0023] Ski resort distribution data is a collection of data on the location information of various ski resorts within a target area, including several geographic coordinate points corresponding to each ski resort. These geographic coordinate points can accurately locate the ski resorts on the Earth's surface, providing a basis for subsequent identification of potential ski resort areas.

[0024] Geographic coordinates are used to represent the location information of a specific point on the Earth's surface using longitude and latitude; they are the basic elements for determining geographical location. In this embodiment, the approximate area of ​​the ski resort can be outlined using multiple geographic coordinates.

[0025] This step is the initial stage in identifying potential ski resort areas. Existing ski resort distribution information within the target area is collected through various means, including Geographic Information System (GIS) databases, relevant tourism statistics, and official ski resort websites. The acquired ski resort distribution data includes several geographic coordinate points corresponding to each ski resort. These coordinate points accurately identify the specific location of each ski resort within the target area, laying the foundation for subsequent expansion into potential ski resort areas based on this location information. For example, the latitude and longitude coordinates of multiple ski resorts within a mountainous area can be extracted from a GIS database; these coordinates constitute an important component of the ski resort distribution data.

[0026] Step S1012: Using each geographical coordinate point in the ski resort distribution data as the center, expand outward to form a rectangular area of ​​a preset size; merge all the rectangular areas to obtain the potential ski resort area.

[0027] The pre-defined rectangular area is a rectangular region formed by expanding outwards from each geographic coordinate point in the ski resort distribution data according to pre-set length and width. Determining the pre-defined dimensions requires comprehensive consideration of factors such as the actual size of the ski resort and the surrounding terrain. For example, if a ski resort typically occupies a large area and has relatively open surrounding terrain, the length and width of the expanded rectangular area can be appropriately increased; conversely, if the ski resort is small and the surrounding terrain is complex, the dimensions of the rectangular area should be correspondingly reduced.

[0028] The potential ski area is the region obtained by merging all rectangular areas centered on various geographical coordinate points. This area is a preliminary delineation of the possible ski areas based on historical ski area distribution, providing a basic range for further precise identification of ski area size.

[0029] After obtaining the geographic coordinates from the ski resort distribution data, this step expands outwards from each coordinate point according to pre-defined length and width, forming a rectangular area of ​​a pre-defined size. All rectangular areas centered on each geographic coordinate point are then merged, removing overlapping portions and retaining the total area covered by all rectangular areas. The resulting area is the potential ski resort area. This area includes known ski resorts within the target area and their surrounding areas that are potentially suitable for skiing, providing a reasonable boundary for further precise identification of the ski resort area.

[0030] In this embodiment, existing ski resort distribution data is utilized, with the geographical coordinates of each ski resort as the basis for expansion. This approach fully leverages historical information, ensuring the high rationality and reliability of the identified potential ski resort areas. Because existing ski resort locations are often determined through multifaceted considerations, such as terrain, climate, and transportation, areas expanded from these locations are more likely to possess suitable conditions for skiing. Secondly, using pre-defined rectangular areas for expansion and merging is a simple and easy method that guarantees comprehensive coverage. Finally, the potential ski resort areas identified in this way provide a clear baseline for subsequent ski resort area identification, avoiding blind searching and analysis within excessively large target areas, thus significantly improving identification efficiency and accuracy.

[0031] In one embodiment, step S1011, obtaining snowfield distribution data for the target area, includes: Step S10111: Obtain ski resort location data from at least two platforms.

[0032] Ski resort location data refers to discrete data points that record the geographical location of ski resorts. It usually includes attribute information such as the latitude and longitude coordinates, name, and size of the ski resort and is the basic unit for constructing ski resort distribution data.

[0033] The platform refers to the source channels that provide ski resort location data, such as government geographic information public service platforms, tourism industry databases, commercial map service APIs, etc. Different platforms have different data collection methods, coverage and update frequency.

[0034] Step S10112: Merge and deduplicate the ski resort location data from each platform to obtain ski resort distribution data.

[0035] Merging and deduplication refers to the process of integrating ski resort location data obtained from multiple platforms. This involves using techniques such as coordinate matching, name comparison, or unique identifier verification to identify and remove duplicate records while retaining only the unique and valid ski resort location information. This ensures the accuracy and consistency of ski resort distribution data.

[0036] This embodiment improves the rationality and reliability of potential ski resort areas through multi-platform data acquisition and deduplication technology. Multi-source data fusion overcomes the coverage blind spots and update delays of single data sources, making ski resort distribution data closer to actual operation. Merging and deduplication processes eliminate the computational burden and regional definition deviations caused by data redundancy, ensuring that the rectangular area expanded from the geographic coordinate point can not only fully cover known ski resorts, but also reasonably extend to surrounding potential ski areas.

[0037] For step S102, multispectral remote sensing data of the potential snowfield area is acquired; the multispectral remote sensing data is analyzed using a preset deep learning model to determine the snow-covered area.

[0038] Multispectral remote sensing data is data acquired through multispectral remote sensing technology. Multispectral remote sensing refers to dividing the electromagnetic wave information radiated or reflected by an object into several bands according to the wavelength of the spectrum, and simultaneously detecting the same ground feature to obtain information from different bands. Different ground features have different reflectance characteristics in different bands, and multispectral remote sensing data can record these characteristics, thus reflecting the features of ground features more comprehensively and accurately, playing an important role in identifying snow cover.

[0039] The preset deep learning model is a model built based on deep learning algorithms. It automatically learns complex patterns in data through multi-layer nonlinear transformations. In this embodiment, it is used to accurately identify snow-covered areas from multispectral remote sensing data. In this embodiment, the preset deep learning model has been trained on a large number of snow-covered and non-snow-covered samples, enabling it to accurately extract snow-related features from multispectral remote sensing data and perform classification, distinguishing between snow-covered and non-snow-covered areas. In one embodiment, the preset deep learning model is a pixel-level segmentation model built on an improved U-Net architecture, specifically designed for identifying snow-covered areas from multispectral remote sensing data.

[0040] After identifying potential snowfield areas, multispectral remote sensing technology was used to acquire multispectral remote sensing data of the area. This data contains reflectance information of ground objects in different bands, providing a more comprehensive picture of their characteristics. The acquired multispectral remote sensing data was then input into a pre-set deep learning model. This model, trained on a large number of snow-covered and non-snow-covered samples, has learned the unique characteristics and patterns of snow cover in multispectral data. The model automatically analyzes and processes the input data, classifying each pixel based on the learned features and patterns to determine whether it belongs to a snow-covered area, thereby accurately determining the snow cover extent within the potential snowfield area.

[0041] In one embodiment, step S102, acquiring multispectral remote sensing data of the potential snowfield area, includes: Step S1021: Obtain satellite multispectral data of the potential ski resort area within a preset time period.

[0042] The preset time period is a specific time range determined based on the research objectives and the characteristics of the ski resort, such as the winter ski season, to ensure that the data obtained can reflect the state of the ski resort during typical operating periods.

[0043] Satellite multispectral data is data acquired by satellites carrying multispectral sensors in orbit by observing the Earth's surface. It has advantages such as wide coverage and fast data updates, and is an important data source for obtaining information on large-scale ski resorts.

[0044] This step is fundamental to acquiring multispectral remote sensing data. Based on research needs, a pre-defined time period representing the typical conditions of a ski resort is determined, such as December to February of the following year. This period typically represents the peak operating season for ski resorts, with relatively stable snow cover. In one embodiment, the pre-defined time period is December to March of the following year. Multispectral satellite data covering the potential ski resort area within this time period is acquired through collaboration with satellite data providers or by utilizing publicly available satellite data platforms. This data contains reflectance information of the potential ski resort area in different spectral bands, providing a rich data source for subsequent analysis.

[0045] Step S1022: Extract several band data of different resolutions from the satellite multispectral data; resample each band data to band data of a preset resolution; and merge the band data of each preset resolution to obtain multichannel spectral data.

[0046] Band data refers to the image data corresponding to each specific spectral band in multispectral remote sensing data. Different bands respond differently to different ground features. For example, the visible light band is sensitive to vegetation color, the near-infrared band reflects vegetation health status significantly, and the shortwave infrared band plays an important role in snow identification. In this embodiment, the extracted band data includes the B2, B3, B4, and B8 bands at a resolution of 10 meters, and the B11 and B12 bands at a resolution of 20 meters.

[0047] Resolution refers to spatial resolution, which is the actual size of the ground represented by one pixel in an image. The higher the resolution, the clearer the image details.

[0048] Resampling is the process of converting image data from one resolution to another. It involves changing pixel sizes through methods such as interpolation, allowing data from different resolutions to be analyzed and processed at the same scale. In this embodiment, all bands are resampled to a 10-meter resolution to form 6-channel data.

[0049] Multichannel spectral data is a dataset composed of data from multiple different bands. Each channel corresponds to a specific band, which can comprehensively utilize information from different bands to improve the accuracy of ground feature identification and classification.

[0050] Satellite multispectral data typically contains multiple bands, each with potentially different spatial resolutions. For example, in some satellite multispectral data, the spatial resolution of the visible light band might be 30 meters, while the spatial resolution of the shortwave infrared band might be 60 meters. Extracting these different resolution band data from the acquired satellite multispectral data allows for full utilization of the information in each band. To enable analysis of band data with different resolutions at the same scale, resampling is required to convert each band data to a preset resolution. The selection of the preset resolution needs to consider factors such as research objectives, data accuracy, and computational efficiency. For example, if the primary focus is on the large-area distribution of snowfields, a relatively low resolution, such as 30 meters, can be chosen; if precise identification of snowfield boundaries and small snowdrifts is required, a higher resolution, such as 10 meters, can be selected. Commonly used interpolation methods during resampling include nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation. Different interpolation methods will have a certain impact on the resampling results, and the appropriate method needs to be selected based on the specific circumstances. The resampled band data are then merged in band order to form a multichannel spectral dataset. Each channel corresponds to a specific band. Multi-channel spectral data can comprehensively utilize information from different bands to improve the identification and classification of features such as snowfields. For example, by merging data from the visible light band, near-infrared band, and short-wave infrared band, the different reflectance characteristics of snow, vegetation, and water bodies in different bands can be used to more accurately identify snowfield areas.

[0051] Step S1023: The multi-channel spectral data is determined as the multispectral remote sensing data of the potential snowfield area.

[0052] The multi-channel spectral data obtained after the above processing steps contains information about the potential snowfield area in multiple spectral bands and has a uniform preset resolution, which can meet the needs of subsequent analysis and processing of the snowfield. Therefore, this multi-channel spectral data is identified as the multispectral remote sensing data of the potential snowfield area, providing an accurate and comprehensive data foundation for subsequent steps such as using deep learning models to analyze snow-covered areas and calculate road accessibility.

[0053] This embodiment acquires satellite multispectral data of potential ski resort areas within a preset time period to ensure the data reflects typical ski resort conditions. Then, several bands of data at different resolutions are extracted from the data, resampled to a preset resolution, and merged into multi-channel spectral data. This fully leverages the potential of satellite multispectral data; the information from different resolution bands is rationally integrated, avoiding information loss due to resolution differences and maximizing the utilization of the rich and diverse spectral information of ground features in the multispectral data. The multi-channel spectral data integrates the advantages of each band, enabling a more accurate representation of the ground feature characteristics of potential ski resort areas.

[0054] In one embodiment, before step S102, which analyzes the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area, the method further includes: Step S1024: Radiometric calibration, atmospheric correction, and image registration are sequentially performed on the multispectral remote sensing data to obtain the first preprocessed data.

[0055] Radiometric calibration refers to the process of converting the digital quantization (DN) values ​​output by a remote sensor into physical quantities such as radiance or reflectance. Since there is a certain proportional relationship between the sensor's output value and the actual received radiation energy when receiving radiation from ground objects, radiometric calibration determines this proportional relationship, thereby eliminating the sensor's inherent errors and making data acquired at different times and from different sensors comparable.

[0056] Atmospheric correction refers to the process of eliminating the influence of the atmosphere on remote sensing images. Gas molecules and aerosols in the atmosphere absorb and scatter solar radiation, resulting in atmospheric interference in the reflected radiation from ground objects received by the sensor. Atmospheric correction removes these atmospheric influences, restores the true reflectivity of ground objects, and improves the quality of remote sensing images and the accuracy of quantitative analysis.

[0057] Image registration refers to the process of spatially aligning two or more images. In remote sensing data processing, images acquired at different times and from different sensors may have problems such as geometric distortion and viewpoint differences. Image registration uses certain algorithms to find the correspondence between images, making them accurately matched in spatial location for subsequent analysis and processing.

[0058] Step S1025: Perform pixel reflectance normalization processing on each band of the first preprocessed data according to the band to obtain the second preprocessed data, which is used as the multispectral remote sensing data input to the preset deep learning model.

[0059] Pixel reflectance normalization is a process that standardizes the reflectance values ​​of pixels in different bands according to certain rules, unifying their value range to a specific interval, typically [0,1]. This eliminates the influence of differences in units and numerical ranges between different bands, ensuring that data from each band has equal importance in model analysis and improving the stability and accuracy of the model.

[0060] This embodiment performs a series of preprocessing operations on multispectral remote sensing data. Radiometric calibration eliminates errors inherent in the sensor itself, making data from different data sources comparable and providing an accurate physical quantity basis for subsequent processing. Atmospheric correction removes the influence of the atmosphere on remote sensing images, restoring the true reflectance of ground features, improving image quality and the accuracy of quantitative analysis, and making the characteristics of target features such as snow cover clearer. Image registration solves the geometric differences between different images, ensuring the consistency of data in spatial location and providing a unified spatial benchmark for band-based processing. Pixel reflectance normalization eliminates differences in dimensions and numerical ranges between different bands, making data from each band equally important in model analysis, and improving the stability and accuracy of the model. The high-quality data obtained after these preprocessing operations serves as input to a pre-set deep learning model, enabling the model to learn and analyze data features more effectively and identify snow-covered areas more accurately.

[0061] In one embodiment, after step S102, which analyzes the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area, the method further includes: Step S1026: Obtain DEM data of the potential ski area.

[0062] DEM (Digital Elevation Model) data is a digital simulation of the terrain surface using limited terrain elevation data. It is a physical ground model that represents the ground elevation in the form of an ordered numerical array and can reflect the undulation of the terrain.

[0063] The DEM data in this step contains topographic elevation information, which is the basic data for calculating slope and aspect.

[0064] Step S1027: Divide the snow-covered area into several snow patches, and determine the average slope and aspect information of each snow patch based on the DEM data.

[0065] Snow patches refer to relatively independent snow-covered areas formed within a snow-covered region due to factors such as topography and landforms.

[0066] Average slope refers to the degree of inclination of the ground in the area where a snow patch is located. It is expressed in angles and reflects the steepness or gentleness of the ground in that area.

[0067] Slope aspect information refers to the orientation of the ground in the area where snow patches are located, such as east, south, west, and north. It has an important influence on the distribution and melting of snow.

[0068] The snow-covered area was divided into several snow patches for more detailed analysis. Based on the acquired DEM data, the average slope and aspect of each snow patch were calculated using appropriate algorithms to understand the topographic features of the area where each snow patch was located.

[0069] Step S1028: Remove snow patches in the snow-covered area whose average slope is greater than a first preset slope threshold or less than a second preset slope threshold to obtain the removed snow-covered area; the first preset slope threshold is greater than the second preset slope threshold.

[0070] The first preset slope threshold and the second preset slope threshold are preset slope limits used to filter snow patches. The first preset slope threshold is greater than the second preset slope threshold and is used to determine snow patches that meet a specific slope range.

[0071] Based on the set first and second preset slope thresholds, snow patches within the snow-covered area with an average slope greater than the first preset slope threshold or less than the second preset slope threshold are removed. This is because areas with excessively steep or shallow slopes are generally unsuitable for ski resort use; for example, excessively steep slopes may pose safety hazards, while excessively shallow slopes may hinder snow retention or the skiing experience. This screening process, which removes snow-covered areas, ensures that subsequent analyses better meet the needs of ski resort construction or related activities.

[0072] Step S103, which involves determining the road accessibility area within the snow-covered area based on the road data, includes: Step S1030: Based on the road data, determine the accessible area of ​​the road within the snow-covered area after removal.

[0073] When determining road-accessible areas within snow-covered regions, analysis is conducted on the snow-covered areas after they have been removed, based on road data. This is because the snow-covered areas better reflect actual needs. Determining road-accessible areas based on these removed areas allows for a more accurate identification of regions that meet both snow cover requirements and are easily accessible by road, providing a more reasonable location reference for the subsequent construction, operation, or related activities of the ski resort.

[0074] This embodiment acquires DEM data of potential ski resort areas, meticulously divides the snow-covered area, and calculates the average slope and aspect information of each snow patch. Then, based on a preset slope threshold, unsuitable snow patches are removed, resulting in a snow-covered area more suited to the ski resort's needs. Finally, based on this area and road data, the road accessibility area is determined. This series of operations makes the final determined road accessibility area within the snow-covered area more accurate and reasonable, considering both snow distribution and the impact of terrain factors on the ski resort's suitability. Combined with road accessibility, it provides more practical and feasible area selections for ski resort-related activities, improving the efficiency and rationality of ski resort planning, construction, and operation, and contributing to an enhanced overall user experience and safety.

[0075] In one embodiment, the preset deep learning model includes a pre-trained feature extraction network and an encoder-decoder network decoder; The pre-trained feature extraction network is used to perform deep feature extraction on the multispectral remote sensing data to obtain snow spectral features and spatial features; and outputs feature vectors of the snow spectral features and spatial features. The encoder-decoder network decoder is used to map the feature vector into pixel features with the same resolution as the multispectral remote sensing data through upsampling operations, perform binary classification on each pixel feature to determine whether it is snow-covered or not, and obtain a binary pixel-level segmentation result; based on the binary pixel-level segmentation result, all pixels determined to be snow-covered are merged according to their geographic spatial location to obtain the snow-covered area.

[0076] The preset deep learning model is a pixel-level segmentation model built on an improved U-Net architecture. The pre-trained feature extraction network uses a ResNet network to replace the original encoder of the U-Net model. It leverages ImageNet pre-trained weights to achieve transfer learning, accelerating model convergence and improving performance. The core function of the pre-trained feature extraction network is to perform deep feature extraction on the input 6-channel multispectral remote sensing data, mining the unique spectral and spatial features of snow cover, and outputting a high-dimensional feature vector that fuses the two types of features.

[0077] The encoder-decoder network decoder maps high-dimensional feature vectors to pixel features with the same resolution as the original multispectral remote sensing data through upsampling operations. Then, it uses the Sigmoid function to perform binary classification of each pixel feature as either snow-covered or non-snow-covered, outputting a binary pixel-level segmentation result. Finally, all pixels classified as snow-covered are merged according to their geographic location to obtain a continuous snow-covered area.

[0078] Deep feature extraction uses multi-layer convolutional operations of convolutional neural networks to mine abstract, high-level semantic features in data. Unlike traditional methods that rely on manually designed shallow features, it can more comprehensively characterize the complex representation patterns of snow in multispectral data.

[0079] Feature vectors are multidimensional numerical arrays output by pre-trained feature extraction networks. They contain semantic feature representations obtained by compressing input data after multiple layers of convolution and pooling, and serve as the basis for subsequent upsampling and classification by the decoder.

[0080] Upsampling is the process of expanding a low-resolution feature map to a high-resolution one. It increases the number of pixels by transposing convolution or interpolation, aligning the spatial size of the feature map with the original input data, and providing a spatial correspondence for pixel-level classification.

[0081] Binary classification decision is to make a binary classification decision for each pixel feature, whether it is snow-covered or not. The probability value is output by the Sigmoid or Softmax function, and pixel-level segmentation is achieved by dividing the pixel by a threshold (such as 0.5).

[0082] The result of binarized pixel-level segmentation is a binary mask image that converts the classification probability into 0 / 1, where 1 represents snow-covered pixels and 0 represents non-snow-covered pixels, directly corresponding to the snow distribution location in geographic space.

[0083] Geospatial location merging refers to combining adjacent or consecutive snow-covered pixels into a complete snow-covered area based on the geographic coordinate information of pixels, forming a recognition result with spatial continuity.

[0084] This embodiment achieves high-precision identification of snow-covered areas from multispectral remote sensing data through the collaborative design of a pre-trained feature extraction network and an encoder-decoder network. The pre-trained feature extraction network, with its powerful feature learning capabilities, can deeply mine the spectral-spatial joint features of snow in multispectral data, overcoming the misjudgment problem of complex surface backgrounds (such as mountain shadows and bright white features) in traditional methods (e.g., NDSI index). Especially in mountainous snowfield scenes, it can accurately distinguish the differences in reflectance characteristics between snow, rocks, and vegetation. The decoder, through upsampling and skip connection mechanisms, restores spatial resolution while fusing shallow detail information, ensuring the clarity and internal continuity of pixel-level segmentation results, avoiding the "salt-and-pepper noise" and rough boundaries generated by traditional methods. After merging the binarized pixel-level segmentation results with geospatial locations, the resulting snow-covered areas highly match the actual snowfield distribution, providing accurate basic data for subsequent calculation of road accessibility areas and snowfield area identification. This model, through an end-to-end deep learning architecture, achieves fully automated processing from multispectral data input to snow-covered area output, significantly improving recognition efficiency and accuracy. It also has good generalization ability and can adapt to the snow resort recognition needs of different regions and seasons, providing strong technical support for outdoor snow resort planning, operation and resource management.

[0085] In one embodiment, the preset deep learning model is trained through the following steps: Step S201: Obtain a training dataset; wherein the training dataset includes several training samples and corresponding training labels; the several training samples include multispectral remote sensing images of outdoor snowfields and non-snowfield areas in different scenes and time periods; the training labels are binarized pixel-level segmentation labels corresponding to the multispectral remote sensing images; the binarized pixel-level segmentation labels are used to label the snow or non-snow category of each pixel.

[0086] Multispectral remote sensing images covering outdoor snowfields and non-snowfield areas at different times (e.g., winter snow cover period, summer melting period) are collected from publicly available data sources (such as Sentinel-2 satellite) or historical archives. For each image, each pixel is manually labeled with its snow or non-snow cover category, generating corresponding binary pixel-level segmentation labels (1 for snow cover, 0 for non-snow cover). The labels must be verified and corrected to ensure accuracy. For example, for mountain snowfield images, misclassification due to mountain shadows must be eliminated, and only pixels with actual snow cover should be labeled.

[0087] Step S202: Input the training dataset into the preset deep learning model to obtain the binarized pixel-level segmentation results and predicted snow cover areas corresponding to each training sample.

[0088] Multispectral remote sensing images from the training dataset are input into a pre-defined deep learning model. The model extracts high-dimensional features from the images (such as the spectral reflectance characteristics and spatial texture of snow) through a pre-trained feature extraction network (such as ResNet). Then, the encoder-decoder network performs upsampling and classification prediction, outputting the snow probability of each pixel. The predicted pixel-level segmentation result is obtained by binarization through a threshold (such as 0.5). At the same time, consecutive snow pixels are merged into predicted snow regions to form spatially continuous recognition results.

[0089] Step S203: Calculate pixel binary classification loss based on the binarized pixel-level segmentation result and the corresponding binarized pixel-level segmentation label; calculate region matching loss based on the predicted snow area and the real snow area determined by the binarized pixel-level segmentation label; obtain the total loss value of model training based on the pixel binary classification loss and the region matching loss; when the total loss value is greater than or equal to a preset threshold, adjust the weights of the pre-trained feature extraction network and the decoder weights of the encoder-decoder network of the preset deep learning model, and re-input the training dataset into the preset deep learning model; until the total loss value is less than the preset threshold, the model training is considered complete.

[0090] The cross-entropy between the predicted pixel category and the ground truth label is calculated, reflecting the classification accuracy of a single pixel. The Dice coefficient or IoU value between the predicted snow area and the actual snow area is calculated, reflecting the degree of matching between the spatial location and shape of the region. The weighted sum of the pixel binary classification loss and the region matching loss is used as the objective function for model optimization. If the total loss value is greater than or equal to a preset threshold, the weights of the pre-trained feature extraction network and the encoder-decoder network decoder are adjusted using the backpropagation algorithm. The data is then re-input into the training dataset for prediction and loss calculation until the total loss value is less than the threshold, at which point model training is complete.

[0091] This embodiment constructs training datasets across multiple scenarios and time periods, and achieves high-precision identification of snow-covered areas in complex terrain using a pre-defined deep learning model through joint optimization of pixel-level classification loss and region matching loss. Specifically, the diversity of the training dataset enables the model to learn the generalized representation patterns of snow in multispectral data, overcoming the sensitivity issues of traditional methods (such as the NDSI index) to thin clouds, mountain shadows, and bright white features. Pixel binary classification loss ensures the accuracy of individual pixel classification, while region matching loss further constrains the spatial continuity and shape rationality of the predicted snow-covered area, avoiding fragmented or over-expanded identification results. By iteratively optimizing the weights of the pre-trained feature extraction network and the encoder-decoder network decoder, the model can automatically adjust the parameters of feature extraction and classification prediction, improving its adaptability to complex scenarios. The final trained model can output snow-covered areas with clear boundaries and internal continuity, providing reliable technical support for outdoor snowfield area identification, dynamic monitoring, and resource management, significantly outperforming traditional methods based on thresholds or single loss functions.

[0092] For step S103, obtain road data of the potential snowfield area; based on the road data, determine the road accessible area within the snow-covered area.

[0093] Road data includes information such as the geographical location, direction, and width of roads within the target area. This data can be obtained through Geographic Information System (GIS) databases, satellite imagery interpretation, and other methods. Road data reflects the accessibility of the area and is crucial for determining the actual usable area of ​​the ski resort.

[0094] The accessible area is the snow-covered area that can be reached by road. Because only the accessible area can skiers easily access the area, and can the facilities (such as chairlifts, rest areas, etc.) be properly arranged, the accessible area is the actual usable part of the ski resort.

[0095] To further determine the actual usable area of ​​the ski resort, road data for the potential ski resort area needs to be obtained. This data can be acquired through various means, such as extraction from Geographic Information System (GIS) databases or through the interpretation and analysis of satellite imagery. After acquiring the road data, the spatial analysis functions of GIS are used to overlay and analyze the road data with the snow-covered area. By determining whether each location within the snow-covered area is connected to a road, the road-accessible area within the snow-covered area is determined. For example, using the buffer analysis and network analysis functions of GIS, areas within the snow-covered area that are within a certain range of roads and can be reached via the road network can be identified; these areas are the road-accessible areas.

[0096] In one embodiment, the road data is road vector data.

[0097] Road vector data is geographic information data that represents roads in vector form. Vector data uses geometric elements such as points, lines, and polygons to accurately describe the location, shape, and spatial relationships of geographic entities. For roads, road vector data typically uses a series of ordered coordinate points connected to form a line to represent the road's centerline. It can also include attribute information such as the road's name, grade, and width. It has advantages such as high data accuracy and strong spatial analysis capabilities, and can accurately express the direction and distribution of roads.

[0098] Step S103, which involves determining the accessible area of ​​the road within the snow-covered area based on the road data, includes: Step S1031: Divide the snow-covered area into several sub-regions; calculate the road network density of each sub-region based on the road vector data within each sub-region and the area of ​​each sub-region.

[0099] Subregions are smaller areas into which the entire snow-covered area is divided according to certain rules for ease of calculation and management during the analysis of snow-covered areas. These subregions can be regular grids (such as square or rectangular grids) or irregular divisions based on geographical features, administrative regions, etc.

[0100] Road network density refers to the ratio of the total length of the road network within a certain area to the area of ​​that region. It is an important indicator for measuring the level of development of road construction and the convenience of transportation in a region. The higher the road network density, the denser the road distribution in the region and the more convenient the transportation; conversely, the lower the road network density, the sparser the road distribution and the potentially poorer the accessibility.

[0101] This step involves dividing the snow-covered area into several sub-regions based on actual needs and research precision, using an appropriate method. For example, a regular grid division method can be used, dividing the snow-covered area into square or rectangular grid sub-regions of the same size at certain intervals; alternatively, irregular division can be made based on geographical features such as natural boundaries like mountains and rivers, or administrative boundaries. The purpose of dividing into sub-regions is to analyze the road distribution within the snow-covered area in greater detail, facilitating subsequent calculations of road network density. For each sub-region, the total length of roads within that sub-region is obtained using road vector data. Road vector data stores road information in the form of lines; the total road length is obtained by calculating the sum of the lengths of these lines. Simultaneously, the area of ​​the sub-region is also obtained. Then, according to the definition of road network density, the total road length is divided by the area of ​​the sub-region to calculate the road network density of each sub-region. For example, if the total road length in a sub-region is 10 kilometers and the area of ​​the sub-region is 5 square kilometers, then the road network density of that sub-region is 10 ÷ 5 = 2 kilometers per square kilometer.

[0102] Step S1032: Merge the sub-regions within the snow-covered area whose road network density is greater than or equal to a preset road network density threshold to obtain the road-accessible area within the snow-covered area.

[0103] A preset road network density threshold is a road network density value pre-set according to actual needs and research objectives, used to determine whether a sub-region belongs to a road-accessible area. When the road network density of a sub-region is greater than or equal to the threshold, the sub-region is considered to have good road accessibility and belongs to a road-accessible area; otherwise, the sub-region is considered to have poor road accessibility and does not belong to a road-accessible area.

[0104] In this step, a suitable road network density threshold is pre-set based on the research objectives and actual conditions, such as the size of the ski resort and the travel needs of tourists. For example, for a large outdoor ski resort, a higher road network density threshold, such as 1.5 km / km², might be set to ensure that tourists can easily access all areas; while for a small ski resort, the threshold might be relatively lower. All sub-regions are traversed, and sub-regions with a road network density greater than or equal to the preset threshold are merged. During merging, algorithms such as spatial connectivity and region fusion can be used to combine these eligible sub-regions into a continuous region, which is the road-accessible area within the snow-covered area. For example, if multiple adjacent sub-regions have a road network density greater than the preset threshold, they can be merged into a large road-accessible area, indicating that the area has a dense road distribution, convenient transportation, and is easily accessible to tourists.

[0105] This embodiment divides snow-covered areas into sub-regions and calculates road network density to determine road accessibility, providing scientific and accurate technical support for the planning and operation of outdoor ski resorts. Dividing snow-covered areas into sub-regions allows for a more detailed analysis of road distribution within each region, avoiding the loss of detail that may occur in overall analysis and improving the accuracy of the analysis. Calculating the road network density of each sub-region quantitatively measures the road development and traffic convenience of each sub-region, making the assessment of road accessibility more objective and accurate. By setting a road network density threshold and merging sub-regions that meet the criteria, the road accessibility within snow-covered areas can be quickly and effectively determined, identifying areas with better traffic conditions and facilitating ski resort managers to rationally plan the layout of ski resort facilities.

[0106] In one embodiment, step S1031, which calculates the road network density of each sub-region based on the road vector data and the area of ​​each sub-region, includes: Step S10311: Obtain the traffic level of each road vector data; divide each road vector data into several levels of roads according to the traffic level.

[0107] Traffic levels are a grading system based on road design standards, traffic capacity, and service functions, such as high-grade roads (expressways / highways), medium-grade roads (arterial roads / secondary arterial roads), and basic-grade roads (local roads / rural roads). Different levels contribute significantly to regional accessibility. In one embodiment, several road levels include high-grade roads, medium-grade roads, and basic-grade roads.

[0108] This step extracts traffic level attributes from road vector data and classifies roads into several levels based on design standards, such as high, medium, and basic levels. For example, highways are classified as high-level due to their high traffic efficiency, while rural roads are classified as basic-level due to their limited traffic capacity, ensuring that the level classification matches the actual traffic function.

[0109] Step S10312: Assign corresponding preset weights to roads of each level in sequence.

[0110] The preset weights are quantitative coefficients assigned according to the road grade. Specifically, the weights are assigned in descending order of road grade, with higher-grade roads receiving higher weights and lower-grade roads receiving lower weights, reflecting the advantages of higher-grade roads in terms of traffic efficiency and coverage.

[0111] This step assigns preset weights in descending order of road accessibility, with higher-level roads having the highest weight and basic-level roads having the lowest. The weighting is based on actual traffic survey data to ensure that higher-level roads dominate the road network density calculation, meeting the ski resort's operational needs for rapid accessibility.

[0112] Step S10313: For each sub-region, calculate the total length of roads of each level within the sub-region; calculate the weighted total length of roads in the sub-region by summing the total length of roads of each level and their corresponding preset weights; and calculate the road network density of the sub-region by dividing the weighted total length of roads in the sub-region by the area of ​​the sub-region.

[0113] The weighted total road length is the sum of the products of the length of each road level and its corresponding weight, comprehensively reflecting the traffic capacity and importance of the road network.

[0114] This step calculates the total length of roads of each grade for each sub-region, multiplies it by the corresponding weight, sums the weighted total road lengths, and then divides it by the sub-region area to calculate the road network density. For example, in a sub-region, high-grade roads are 5km long (weight 1.0), medium-grade roads are 10km long (weight 0.7), and basic roads are 20km long (weight 0.3). The weighted total length is 5×1.0+10×0.7+20×0.3=18km, and the road network density is 1.8km / km² when the area is 10km².

[0115] This embodiment achieves a refined upgrade in road network density calculation through a traffic level classification and weight allocation mechanism. The weighted total road length comprehensively considers the actual traffic capacity and importance of roads of different levels, making the road network density index more accurately reflect the level of regional traffic accessibility. Compared with the traditional equal-weight calculation method, this scheme highlights the key role of high-level roads in ski resort accessibility through a weight-decreasing design, avoiding the dilution effect of redundant length of low-level roads on the density index. The road network density threshold determined in this way is more in line with the actual needs of ski resort operation, and the merged road accessibility area can accurately cover the core areas that are easily accessible to tourists and easy to deploy facilities.

[0116] For step S104, the outdoor snowfield area of ​​the target area is obtained based on the road-accessible area.

[0117] The accessible area represents the portion of the ski resort that can actually be developed and utilized, as these are the only areas where skiers can easily reach them and where ski facilities can be rationally arranged. By calculating the area of ​​the accessible area, the outdoor ski resort area of ​​the target region can be obtained. Area calculation can be performed using area calculation tools in a Geographic Information System (GIS), accurately determining the size of the area based on the boundary information of the accessible area.

[0118] In one embodiment, step S104, which involves obtaining the outdoor snowfield area of ​​the target area based on the road-accessible area, includes: Step S1041: Perform morphological opening and closing operations sequentially on the road reachable area to complete denoising and hole filling, and obtain the processed road reachable area.

[0119] Morphological opening is an image processing operation based on mathematical morphology. It involves first performing erosion on the image, followed by dilation. Erosion can eliminate small protrusions and remove isolated small noise points; dilation can restore some of the main structure that has been reduced in size due to erosion. Opening is often used to remove noise from images while maintaining the overall shape and size of the target region.

[0120] Morphological closing operations are also based on mathematical morphology, first performing dilation followed by erosion. Dilation fills small voids within the target area and connects adjacent fractures; erosion removes edge burrs caused by dilation. Closing operations are primarily used to fill voids within the target area, making the region more complete and continuous.

[0121] Step S1042: Convert the processed road reachable area into vector polygon data; calculate the area based on the vector polygon data to obtain the outdoor snowfield area of ​​the target area.

[0122] Vector polygon data is geographic information data of polygons represented in vector form. Vector data accurately describes the shape and location of polygons by recording the coordinates of their vertices, offering advantages such as high data accuracy, small storage space, and ease of spatial analysis and geometric operations. Vector polygon data can accurately represent the boundaries and shapes of road-accessible areas.

[0123] This step converts the morphologically processed road accessibility area from raster data or other formats into vector polygon data. Vector polygon data accurately describes the boundaries of the road accessibility area by recording the coordinates of the polygon vertices to construct the shape of the area. For example, relevant tools in GIS (Geographic Information System) software can be used to convert the processed area into vector format, facilitating subsequent spatial analysis and calculations. Furthermore, based on the converted vector polygon data, the area covered by the polygon is calculated using the area calculation function in the GIS software. The GIS software will accurately calculate the area of ​​the polygon using mathematical formulas based on the coordinates of its vertices. This area represents the outdoor ski resort area of ​​the target region, because after the previous processing, the road accessibility area can accurately reflect the area suitable for outdoor ski resort activities.

[0124] This embodiment processes the road accessibility area using morphological opening and closing operations, effectively removing noise and filling voids, thus improving the quality and accuracy of the road accessibility area data. The processed area is then converted into vector polygon data, facilitating precise spatial analysis and geometric calculations. The outdoor ski resort area calculated based on this vector polygon data has high accuracy and reliability, providing accurate area data support for the planning, construction, and management of outdoor ski resorts.

[0125] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for identifying the area of ​​an outdoor snowfield, characterized in that, Includes the following steps: Identify potential ski resort areas within the target region; Acquire multispectral remote sensing data of the potential snowfield area; analyze the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area; Obtain road data for the potential ski resort area; based on the road data, determine the road-accessible areas within the snow-covered area; The outdoor snowfield area of ​​the target area is obtained based on the road accessibility area.

2. The method for identifying the area of ​​an outdoor snowfield according to claim 1, characterized in that, The steps to identify potential ski resort areas for a target region include: Acquire ski resort distribution data for the target area; the ski resort distribution data includes several geographic coordinate points corresponding to each ski resort. Using each geographical coordinate point in the ski resort distribution data as the center, a rectangular area of ​​a preset size is formed by expanding outward; all the rectangular areas are merged to obtain the potential ski resort area.

3. The method for identifying the area of ​​an outdoor snowfield according to claim 2, characterized in that, The steps to obtain ski resort distribution data for the target area include: Obtain ski resort location data from at least two platforms; The ski resort location data from various platforms were merged and deduplicated to obtain ski resort distribution data.

4. The method for identifying the area of ​​an outdoor snowfield according to claim 1, characterized in that, The steps for acquiring multispectral remote sensing data of the potential snowfield area include: Acquire satellite multispectral data of the potential ski resort area within a preset time period; Several band data of different resolutions are extracted from the satellite multispectral data; each band data is resampled to a band data of a preset resolution, and the band data of each preset resolution are merged to obtain multichannel spectral data. The multi-channel spectral data is identified as the multispectral remote sensing data of the potential snowfield area.

5. The method for identifying the area of ​​an outdoor snowfield according to claim 1, characterized in that, After analyzing the multispectral remote sensing data using a preset deep learning model to determine the snow-covered area, the method further includes: Obtain DEM data of the potential ski area; The snow-covered area is divided into several snow patches, and the average slope and aspect information of each snow patch is determined based on the DEM data; Remove snow patches in the snow-covered area whose average slope is greater than a first preset slope threshold or less than a second preset slope threshold to obtain the removed snow-covered area; the first preset slope threshold is greater than the second preset slope threshold. The step of determining the road accessibility area within the snow-covered area based on the road data includes: Based on the road data, determine the accessible areas of the roads within the snow-covered area after removal.

6. The method for identifying the area of ​​an outdoor snowfield according to any one of claims 1 to 5, characterized in that, The preset deep learning model includes a pre-trained feature extraction network and an encoder-decoder network; The pre-trained feature extraction network is used to perform deep feature extraction on the multispectral remote sensing data to obtain snow spectral features and spatial features; Output the feature vectors of the snow spectral features and the spatial features; The encoder-decoder network decoder is used to map the feature vector into pixel features with the same resolution as the multispectral remote sensing data through upsampling operations, and to perform binary classification of each pixel feature as snow-covered or non-snow-covered to obtain a binary pixel-level segmentation result. Based on the binarized pixel-level segmentation results, all pixels identified as snow are merged according to their geographic location to obtain the snow-covered area.

7. The method for identifying the area of ​​an outdoor snowfield according to claim 6, characterized in that, The preset deep learning model is trained through the following steps: Obtain a training dataset; wherein the training dataset includes several training samples and corresponding training labels; the several training samples include multispectral remote sensing images of outdoor snowfields and non-snowfield areas in different scenes and time periods; the training labels are binarized pixel-level segmentation labels corresponding to the multispectral remote sensing images; the binarized pixel-level segmentation labels are used to label the snow or non-snow category of each pixel; The training dataset is input into the preset deep learning model to obtain the binarized pixel-level segmentation results and predicted snow cover areas corresponding to each training sample. Pixel binary classification loss is calculated based on the binarized pixel-level segmentation results and corresponding binarized pixel-level segmentation labels. Region matching loss is calculated based on the predicted snow area and the actual snow area determined by the binarized pixel-level segmentation labels. The total loss value of model training is obtained based on the pixel binary classification loss and the region matching loss. When the total loss value is greater than or equal to a preset threshold, the weights of the pre-trained feature extraction network and the decoder weights of the encoder-decoder network of the preset deep learning model are adjusted, and the training dataset is re-input into the preset deep learning model. The process continues until the total loss value is less than the preset threshold, at which point model training is considered complete.

8. The method for identifying the area of ​​an outdoor snowfield according to any one of claims 1 to 5, characterized in that, The road data is road vector data; The step of determining the road accessibility area within the snow-covered area based on the road data includes: The snow-covered area is divided into several sub-regions; the road network density of each sub-region is calculated based on the road vector data and the area of ​​each sub-region. Sub-regions within the snow-covered area whose road network density is greater than or equal to a preset road network density threshold are merged to obtain the road-accessible area within the snow-covered area.

9. The method for identifying the area of ​​an outdoor snowfield according to claim 8, characterized in that, The step of calculating the road network density of each sub-region based on the road vector data and the area of ​​each sub-region includes: Obtain the traffic level of each road vector data; divide each road vector data into several road levels according to the traffic level. Each road of different levels is assigned a corresponding preset weight in sequence; For each sub-region, the total length of roads of each level within the sub-region is calculated; based on the total length of roads of each level and their corresponding preset weights, a weighted sum is obtained to get the weighted total road length of the sub-region; the road network density of the sub-region is calculated by dividing the weighted total road length of the sub-region by the area of ​​the sub-region.

10. The method for identifying the area of ​​an outdoor snowfield according to any one of claims 1 to 5, characterized in that, The step of obtaining the outdoor ski area of ​​the target region based on the road-accessible area includes: Morphological opening and closing operations are sequentially performed on the road reachable region to complete noise reduction and hole filling, resulting in the processed road reachable region. The processed road-accessible area is converted into vector polygon data; the area is calculated based on the vector polygon data to obtain the outdoor snowfield area of ​​the target area.

Images