A snowfall area cloud identification method based on multi-dimensional information fusion

Abstract

The application discloses a snowfall area cloud identification method based on multi-dimensional information fusion, comprising the following steps: obtaining a remote sensing image of a snowfall area and selecting a wave band; pre-processing the remote sensing image of the selected wave band to obtain a pre-processed image; performing filter processing on the pre-processed image through a hybrid tuning filter technology to obtain a gray scale image; separating the gray scale image according to normal distribution characteristics by using a false anomaly screening separation model to obtain cloud body information and non-cloud body information; and eliminating a small part of false anomalies left by the false anomaly screening separation model through spatial statistical analysis of an Anscombe local Moran index. The hybrid tuning filter technology is introduced, aiming to strengthen the identification ability of the cloud body information and effectively inhibit spectral interference generated by a complex ground surface background; subsequently, the cloud layer characteristics are locked through the false anomaly screening separation model, and multi-dimensional information fusion is performed through spectral information and spatial statistical information, so that the cloud identification precision is further improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of remote sensing image cloud recognition, and particularly relates to a snowfall area cloud recognition method based on multi-dimensional information fusion. BACKGROUND

[0002] Snowfall, as an important form of precipitation in nature, its distribution, intensity and duration have a profound impact on water resource management, agricultural production, transportation and ecological balance. However, in remote sensing images, both clouds and snowfall show high reflectivity, and the spectral characteristics of clouds and snowfall in the visible and infrared bands are very similar. This spectral similarity makes it particularly difficult to distinguish clouds and snow in remote sensing images, and snowfall areas are often misjudged as clouds.

[0003] In addition, the texture characteristics of clouds and snow have certain complexity and are changeable. The shape, thickness and density of clouds will affect their texture performance in the image, and the coverage of snow and the terrain will also affect the texture characteristics of the snow layer. The complexity of the texture increases the difficulty of distinguishing clouds and snow. The confusion between cloud layers and snowfall areas not only reduces the accuracy of weather forecasting, but also affects the reliability of environmental monitoring and climate change research.

[0004] At present, the existing cloud recognition algorithms are mostly based on threshold, texture features or deep learning methods, but the applicability of these methods in different regions, different times and different weather conditions is different, and the universality is low, and the effect is not good in snowfall area. At the same time, most of the cloud recognition algorithms do not fully utilize the advantages of spatial information technology, and cannot provide rich ground information, which directly affects the accuracy and effect of cloud and snow area division. SUMMARY

[0005] To solve the above technical problems, the present application provides a snowfall area cloud recognition method based on multi-dimensional information fusion to solve the problems existing in the prior art.

[0006] To achieve the above purpose, the present application provides a snowfall area cloud recognition method based on multi-dimensional information fusion, comprising:

[0007] Obtaining a remote sensing image of a snowfall area and selecting a wave band;

[0008] Pretreating the remote sensing image of the selected wave band to obtain a pretreated image;

[0009] Filtering the spectral information of the pretreated image by a hybrid tuning filter technology to obtain a gray image;

[0010] Separating the gray image according to normal distribution characteristics by using a false anomaly screening separation model to obtain cloud body information and non-cloud body information;

[0011] Spatial statistical analysis of the cloud information was performed using the Anselm Local Moran Index to quantify the spatial autocorrelation at specific locations, eliminate false anomalies in snowfall areas, and obtain cloud identification results.

[0012] Preferably, the remote sensing images selected for the bands are nine bands of Landsat 8OLI data, specifically Band 1 to Band 9.

[0013] Preferably, the preprocessing of remote sensing images in selected bands includes:

[0014] The remote sensing images of the selected bands were resampled using a cubic convolution interpolation method, and the resolution of the Band 8 band was resampled from 15 meters to 30 meters to match the resolution of other multispectral bands.

[0015] Preferably, the process of filtering the preprocessed image using hybrid tuning filtering technology includes:

[0016] The endmember spectral information of interest to the user is extracted through a partial demixing strategy to generate a hybrid tuned filter image;

[0017] Based on the hybrid tuned filtered image, target regions with low spectral contrast between the target and the background, as well as within the target itself, are detected.

[0018] The abundance image is adjusted by applying a feasibility score to the target region to obtain a grayscale image.

[0019] Preferably, the process of separating the grayscale image based on the normal distribution characteristics using a false anomaly screening and separation model includes:

[0020] Based on the normal distribution characteristics, the grayscale image is separated using a false anomaly screening and separation model to obtain cloud information and non-cloud information with a bimodal normal distribution.

[0021] The separability between the probability distributions of cloud information and non-cloud information is assessed using the Bach distance.

[0022] Preferably, the cloud identification method for snowfall areas also includes a process of improving the false anomaly screening and separation model:

[0023] Based on the inherent differences between the target area and its background, a reasonable threshold range is calculated, and the threshold range is used to distinguish between the area to be extracted and the background area.

[0024] A local subdivision strategy was introduced, which subdivided the local data multiple times based on the specific parameters of each sub-distribution in the mixed distribution in order to extract the information of the land features of interest.

[0025] A separation assessment is introduced to determine the degree of separation between the ground features to be extracted and the background information based on the magnitude of the separation.

[0026] Preferably, the false anomaly screening and separation model is a mixed distribution model, and its expression is:

[0027]

[0028] Wherein, the parameter set μ = (n, μ1, ..., μ) n a1, ..., a n It contains a series of key elements for defining multiple branches, where n represents the total number of branches, and f(x, μ) i ) represents the probability density of the i-th branch, μ i For the corresponding parameter, a i Let represent the weight of the i-th branch, and

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] This invention proposes a cloud identification method for snowfall areas based on multi-dimensional information fusion. For accurate cloud identification in snowfall areas, it first uses Landsat 8OLI data as the research data and selects specific bands. Then, preprocessing techniques are employed to eliminate resolution inconsistencies between different bands, ensuring all bands have the same spatial sampling interval. Next, a hybrid tuned filtering technique is introduced to enhance the identification capability of cloud information while effectively suppressing spectral interference from complex surface backgrounds. Subsequently, a false anomaly screening and separation model is used to lock in cloud features. Finally, spatial statistical analysis using the Anselm local Moran index eliminates a small portion of residual false anomalies from the false anomaly screening and separation model, yielding the cloud identification result. This invention further improves the accuracy of cloud identification in snowfall areas by fusing multi-dimensional information from spectral and spatial statistical information.

[0031] This invention, through continuous optimization of the cloud identification algorithm, improves its accuracy and stability in snowy areas, providing more accurate and reliable data support for fields such as weather forecasting, environmental monitoring, and climate change research. Furthermore, with the continuous development of remote sensing technology and the improvement of data acquisition capabilities, it is expected to achieve higher precision, higher resolution, and wider coverage cloud identification services in the future, contributing more wisdom and strength to global climate change response and sustainable development goals. Attached Figure Description

[0032] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0033] Figure 1 The remote sensing image of region a is a representative area of ​​this invention embodiment;

[0034] Figure 2 This is a false-color composite image of a representative region a in an embodiment of the present invention;

[0035] Figure 3 This is a grayscale image of a representative region a from an embodiment of the present invention after hybrid tuning and filtering.

[0036] Figure 4 This is a remote sensing image of region b, a representative area of ​​this invention.

[0037] Figure 5 This is a representative region b of the false-color composite image in an embodiment of the present invention;

[0038] Figure 6 This is a grayscale image of the representative region b after hybrid tuning filtering in an embodiment of the present invention;

[0039] Figure 7 This is a diagram showing the results of the improved pseudo-anomaly screening and separation model for representative region a in this embodiment of the invention.

[0040] Figure 8 This is a diagram showing the results of the improved pseudo-anomaly screening and separation model for representative region b in this embodiment of the invention.

[0041] Figure 9 This is a schematic diagram of the cloud extraction result after processing the representative region a of this invention using the improved pseudo-anomaly screening and separation model.

[0042] Figure 10 This is a schematic diagram of the cloud extraction result after the representative region b of this embodiment of the invention has been processed by the improved false anomaly screening and separation model.

[0043] Figure 11 This is a schematic diagram of the final cloud extraction result after Anselm's local Moran index calculation for a representative region a in an embodiment of the present invention.

[0044] Figure 12 This is a schematic diagram of the final cloud extraction result of a representative region b in an embodiment of the present invention after calculation by the Anselm local Moran index.

[0045] Figure 13 This is a comparison diagram of the embodiments of the present invention with artificial neural networks and FMSK;

[0046] Figure 14 This is a flowchart illustrating the steps of the snowfall area cloud identification method according to an embodiment of the present invention. Detailed Implementation

[0047] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0048] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0049] Example 1

[0050] like Figure 14 As shown, this embodiment provides a method for cloud identification in snowfall areas based on multi-dimensional information fusion, wherein the multi-dimensional information includes: spectral information and spatial statistical information, and the cloud identification method includes the following steps:

[0051] S1, acquire remote sensing images of the snowfall area and select the band;

[0052] Specifically, Landsat 8OLI data is used as the research data, aiming to achieve high-precision identification of cloud layers in snowfall areas through a strategy of deep fusion of multi-dimensional information. In the process of band selection, this embodiment strictly adheres to the following four core considerations to ensure that the selected bands can comprehensively and accurately serve the research objectives of this embodiment:

[0053] ① Information comprehensiveness and feature retention: In order to ensure that the selected bands can retain the feature information of clouds and underlying surfaces to the greatest extent, this embodiment strives to ensure that the selected band combination can cover a wide and comprehensive spectral range, thereby avoiding the omission of any key information and laying a solid foundation for subsequent analysis.

[0054] ② Minimizing spectral information loss: The spectral characteristics of various land features differ under different spectral bands. Therefore, when selecting spectral bands, this embodiment pays special attention to avoiding the loss of spectral information caused by these differences, ensuring that each land feature, including snowfall areas and their clouds, can be accurately and completely represented under its most suitable spectral band.

[0055] ③ Accurate representation of the target region's spectral information: To accurately reflect the true spectral characteristics of the region of interest in subsequent research, this embodiment carefully selected bands that can clearly demonstrate the spectral features of these regions. The selection of these bands not only helps this embodiment to gain a deeper understanding of the physical properties of the target region, but also provides reliable input data for subsequent cloud recognition algorithms.

[0056] ④ Precise capture of spectral differences between clouds and snow: Although easily confused underlying surfaces such as clouds and snow exhibit high reflectivity in the visible light band, their actual reflectance spectral curves differ significantly in specific bands. These differences, such as specific absorption peaks and reflection valleys, are important diagnostic features for distinguishing between clouds and snow. Therefore, in the band selection process, this embodiment pays special attention to selecting bands that can highlight these subtle spectral differences to ensure that the cloud identification method of this embodiment can achieve accurate identification of cloud layers.

[0057] In summary, this patent ultimately selected nine bands (band1 to band9) from Landsat 8OLI data. These bands not only meet all the above selection criteria, but also provide a novel method for identifying snowfall area clouds based on multi-dimensional information fusion, providing strong support for research and application in related fields.

[0058] S2, preprocess the remote sensing image of the selected band to obtain the preprocessed image;

[0059] Specifically, when processing Landsat 8 satellite imagery data, the Band 8 band resolution of this satellite reaches 15 meters, while the resolution of the other multispectral bands is 30 meters. This difference causes inconvenience for subsequent data processing and analysis. To eliminate this inconsistency and ensure that all bands have the same sampling interval in space, this embodiment employs a resampling technique. Specifically, this experiment uses cubic convolutional interpolation as the resampling method, resampling the Band 8 band, originally with a resolution of 15 meters, to 30 meters to match the resolution of other bands. This method is a highly accurate resampling method. It considers the information of multiple original pixels surrounding the output pixel and uses cubic polynomial interpolation to calculate the gray value of the output pixel. This method can produce smoother and more accurate results while preserving image details.

[0060] S3, the spectral information of the preprocessed image is filtered using hybrid tuning filtering technology to obtain a grayscale image;

[0061] Specifically, the hybrid tuned filtering technique extracts endmember information of interest to the user through a partial demixing strategy. In practice, this technique first focuses on a selected set of pixels of interest, then treats the remaining unclassified pixels as background data. The average spectral characteristics of these regions of interest are calculated and used as the baseline endmember information for the hybrid tuned filtering. To ensure data accuracy, training samples were deliberately collected in a relatively flat area with stable lighting conditions, effectively avoiding potential interference from complex terrain and lighting variations.

[0062] Hybrid tuned filtering offers a highly efficient approach for detecting specific target features. Its advantage lies in its ability to accurately identify regions matching the endmember spectrum based on a high degree of matching with pre-defined sample spectra or image endmember spectra, rather than analyzing all pixel details of the entire satellite image. This characteristic makes the method exceptionally effective in reducing confusion errors caused by high-reflectivity surface cover such as snow and deserts, thereby improving the accuracy of cloud recognition in complex terrain environments. This study focuses on extracting endmembers and using them to generate hybrid tuned filtered images, an advanced matching filter method designed to reduce false anomalies. This method can also detect targets with low spectral contrast between the target and the background, as well as within the target itself. The multivariate sample endmembers are denoted as "x".

[0063] Fi(x),i=(1,2,3,...,p) (1)

[0064] Any dataset is a signal component containing background noise caused by various sensors and atmospheric factors. Therefore, the total signal has the following characteristics:

[0065] F(x)=A(x)+B(x) (2)

[0066] The signal and noise components are A(x) and B(x), respectively. Hybrid tuning filtering is used to evaluate the abundance of endmembers, and a feasibility score between 0 and 1 is used to adjust the abundance image by reducing false positives. The abundance image is a distribution map of the relative intensity or significance of certain features or information in the image after filtering.

[0067] In this embodiment, an innovative strategy is employed to construct endmember spectral information: the average spectral value of the four pixels surrounding each pixel within the training region. Compared to single-pixel spectra, this method better represents the comprehensive characteristics of the region, enhancing the stability and reliability of the spectral information. Simultaneously, averaging effectively reduces spectral noise interference, further improving data purity. Furthermore, this method significantly weakens the bright / dark zone effect caused by varying illumination conditions and solar azimuth in complex terrain areas, providing more accurate foundational data for subsequent cloud identification. To comprehensively cover cloud diversity, this study specifically selected two types of regions of interest—thick clouds and thin clouds—encompassing all spectral bands, as the analysis objects. This selection is based on the differences in the physical characteristics of clouds: thin clouds often mix with surface information to form mixed pixels, while thick clouds may be misclassified as other natural cover such as snow or desert due to spectral similarity. This step was efficiently completed on the MATLAB platform, ensuring the accuracy and convenience of data processing.

[0068] To further validate the practical application of this method in cloud identification, this embodiment selects two representative complex background regions for case analysis. These regions not only encompass diverse land cover types but also exhibit various cloud morphologies, thus comprehensively evaluating the feasibility and effectiveness of this embodiment. Future research will further explore and quantify the performance of this method in practical operations, focusing on its application potential on large-scale datasets and its accuracy evaluation.

[0069] Taking representative region a as an example, such as Figure 1 As shown, in the northern region of the image, hills are covered with thick snow, which exhibits high brightness and exists in patchy form. This feature significantly increases the complexity of automatic cloud identification. Simultaneously, irregularly shaped frozen rivers are scattered throughout the image, and their high brightness and disordered distribution further exacerbate the challenges in cloud extraction. Of particular note are the presence of fine point clouds scattered throughout the image, especially in the due south direction. These point clouds, due to their small size, are easily visually confused with the patchy snow in the northern region, further increasing the difficulty of accurate cloud extraction. In summary, the complex characteristics of this image, including the patchy distribution of snow, the disordered high brightness of the frozen rivers, and the subtle confusion caused by point clouds, collectively constitute multiple obstacles in the cloud extraction task.

[0070] Next, this embodiment will employ false-color compositing technology for image a. In this process, this embodiment selects bands 7, 6, and 3 from the Landsat 8OLI dataset, and assigns them to the red, green, and blue channels of the false-color image for compositing, respectively. Through this processing, this embodiment obtains a completely new false-color image, the display effect of which is shown in the attached figure. Figure 2 As shown.

[0071] From this false-color composite image, this embodiment clearly observes a wide distribution of intricate frozen rivers across the entire image area, meandering irregularly. Simultaneously, numerous tiny point clouds are scattered throughout the image. These point clouds, due to their minute size, might be difficult to detect in the original true-color image, but become more prominent in the false-color composite. These image features not only enrich the understanding of the study area in this embodiment but also provide a more intuitive reference for subsequent cloud extraction.

[0072] After hybrid tuning filtering, the resulting grayscale image clearly demonstrates the filtering effect, as shown below. Figure 3As shown in the image, the high-mixed-tuned filter values ​​are mainly concentrated on the clouds, indicating that the method effectively highlights cloud information. Meanwhile, the previously large-scale, irregularly distributed frozen rivers almost completely disappear after filtering, fully demonstrating the superior ability of mixed-tuned filtering to suppress the spectral response of frozen rivers and significantly reducing the complexity of cloud information extraction. Furthermore, the filtering process also enhances the spectral response of some smaller point clouds, making these previously difficult-to-identify cloud features more apparent, further reducing the difficulty of cloud information extraction.

[0073] However, despite the positive results achieved, some patchy high-value pixel remnants are still visible in the high-altitude snow-covered areas of the northern region. These patchy high-value areas may be visually confused with smaller point clouds in the image, posing a challenge to the accurate extraction of cloud information. Therefore, additional measures are needed in subsequent processing to distinguish these patchy snow cover from point clouds to ensure the accuracy and reliability of cloud information extraction.

[0074] To evaluate the impact of hazy, thin clouds on image filtering, this embodiment selects a representative region b as the research object, such as... Figure 4 As shown, a striking feature of this image is the vast expanse of snow cover in its northeastern region, sprawling across the continuous mountain peaks. Due to differences in slope aspect, the snow exhibits drastically different characteristics under varying lighting conditions: on the sun-facing side, the snow's reflectivity is significantly enhanced, and combined with the unique lighting conditions of high latitudes, these areas appear exceptionally bright; conversely, on the side facing away from the sun, the snow appears dim due to the mountains blocking sunlight, creating a stark contrast—the sunny slopes are dazzlingly bright, while the shady slopes are dark and devoid of light. This natural phenomenon results in a patchy landscape of light and dark across the entire high-altitude snow-covered area, greatly complicating the automatic cloud extraction process. Furthermore, the cloud morphology within the selected area is inherently unclear, particularly in the southwestern region of the image, where the clouds exhibit typical hazy characteristics with blurred edges, making precise cloud extraction even more challenging.

[0075] To more intuitively demonstrate the distribution of clouds and snow, this embodiment employs false-color compositing technology to process the true-color image of the first representative area, and presents the composite image as follows: Figure 5 As shown in the image, the clouds in the southwestern region exhibit a typical foggy state with blurred boundaries and indistinct features, which undoubtedly poses a significant challenge to the accurate extraction of the clouds.

[0076] The grayscale image obtained after processing with the hybrid tuning filter method, such as Figure 6As shown, this grayscale image is generated based on mixed-tuned filter values, where the filter values ​​follow a normal distribution with a mean of zero. In this grayscale image, lower pixel values ​​represent background areas (i.e., areas that do not contain the target component), while higher pixel values ​​indicate areas containing a higher proportion of the target component. These are the key objects extracted in this embodiment. It can be clearly seen that high mixed-tuned filter values ​​are mainly concentrated on clouds, especially the thick clouds in the southeast, demonstrating the high sensitivity of this method to clouds. At the same time, surface information in the overall bright background is effectively suppressed, significantly reducing the complexity of cloud extraction. It is particularly noteworthy that the spectral characteristics of the fog-like clouds in the southwest are significantly enhanced after filtering, greatly simplifying the extraction process of such cloud information. However, in the snow-covered area in the northeast, although most of it is in a patchy bright state, there are still a few cases with relatively high pixel values. This again illustrates that the snow in high-altitude areas, due to the complex lighting conditions and slope aspect differences, results in complex and diverse brightness variations in remote sensing images, bringing additional challenges to cloud identification.

[0077] In summary, the hybrid tuned filtering technique demonstrates unique advantages: it effectively suppresses the spectral response of bright background regions while significantly enhancing cloud information, especially for hazy, thin clouds and tiny point clouds that are difficult to capture using traditional methods. Given that the hybrid tuned filtering results exhibit a normal distribution, this experiment further introduces a false anomaly screening and separation model, aiming to extract more refined cloud information for complex terrain areas. This strategy not only improves the accuracy of cloud identification but also enhances the model's adaptability to complex environments.

[0078] S4. Using a false anomaly screening and separation model, the grayscale image is separated according to the normal distribution characteristics to obtain cloud information and non-cloud information.

[0079] In the analysis of remote sensing images, the presence of various land cover types constitutes a complex multi-modal dataset, which is essentially a mixture of information from different land cover types. From a mathematical perspective, this mixed distribution phenomenon of multiple land cover types can be accurately described using the concept of a false anomaly screening and separation model. Among these, the normal distribution, as one of the key mathematical tools for distinguishing different land cover characteristics, is undeniably important, providing a strong basis for identifying land cover features in remote sensing images.

[0080] The statistical characteristics of remote sensing images typically exhibit a multi-modal distribution pattern formed by the superposition of multiple independent populations. In most cases, the ground cover information in the images closely follows the normal distribution law. Even those few ground cover information that deviate from or are significantly inconsistent with the normal distribution can be effectively transformed into a data form that conforms to a normal distribution through in-depth statistical analysis, combined with logarithmic transformation or other types of mathematical transformation techniques.

[0081] Therefore, this embodiment can treat cloud information as an instance of a normal distribution and use a mathematical model for false anomaly screening and separation to accurately characterize it. In practical applications, this embodiment often assumes that the sample data in remote sensing images follow a mixed distribution model, which integrates the distribution characteristics of various ground features, laying a solid theoretical foundation for subsequent cloud information extraction and analysis.

[0082]

[0083] Wherein, the parameter set μ = (n, μ1, ..., μ) n a1, ..., a n It contains a series of key elements for defining multiple branches. Specifically, n represents the total number of branches, and f(x, μ) i ) represents the probability density of the i-th branch, μ i For the corresponding parameter, a i Let represent the weight of the i-th branch, and have

[0084] This model can extract specific land cover information more accurately. Furthermore, this embodiment improves model performance by using the separation parameter between sub-distributions. Based on in-depth analysis of representative training data for each target category, it quantifies the distinguishability between different target categories. In statistics, especially under the assumption that the data follows a normal probability distribution, the Bhattacharyya Distance (L) has proven to be an effective and applicable metric for evaluating the separability between two probability distributions.

[0085] The Bhattacharyya distance not only reflects the probability of misclassification in classification tasks but is also closely related to the optimal classification performance under Bayesian decision theory. Therefore, it has a solid theoretical foundation as a measure of separability. The specific formula for calculating the Bhattacharyya distance is as follows. This formula can be directly applied to the model in this embodiment to quantify and optimize the separation effect between sub-distributions, thereby enhancing the accuracy and reliability of the model in extracting information about features of interest.

[0086]

[0087] Where α1 and α2 represent the means of two distribution characteristics, and Let represent the variances of the two distributions. If the means of the two distributions are exactly the same, the first term in the formula will automatically become zero, because the difference in means is the basis for calculating this term, and zero difference naturally leads to this term being zero. On the other hand, if the variances of the two distributions are completely identical, the second term in the formula will also disappear to zero, because the difference in variance is a key factor in evaluating this part, and equal variances mean there is no difference, thus making this term zero. That is, the equality of the means eliminates the influence of the first term, while the equivalence of the variances makes the second term no longer contribute to the overall measure of separability. This characteristic helps this embodiment to more clearly understand how different statistical characteristics independently affect the degree of separability between two distributions.

[0088] When dealing with complex classification scenarios containing easily confused features, the SD distance is introduced as a means of describing separation, demonstrating significant advantages over the previously mentioned L separation method. The SD distance not only provides a clear and finite dynamic range, but this characteristic also makes comparing the separation results of different features more intuitive and effective. Through the SD distance, this embodiment can more accurately identify features with good separability, which is crucial for improving classification accuracy. The SD distance varies within its domain (a scale range of 0 to 2), providing a standardized perspective for evaluating the separability between two categories. Specifically, the L parameter, as part of the SD distance calculation or directly as its metric, is used to quantify the degree of separability between categories, thereby helping this embodiment make more scientific and reasonable decisions in practical applications.

[0089] SD = 2(1-e -L (5)

[0090] Ideally, if the features of two sub-distributions exhibit perfect separability, their SD value will approach 2. This indicates that the two categories can be completely distinguished along a specific feature dimension, and theoretically, no misclassification will occur when using this feature for classification. Conversely, as the SD value decreases, the separability gradually decreases, meaning that the confusion between sub-distributions increases, and the possibility of classification errors also rises. In the extreme case, SD = 0 indicates that the two categories are almost indistinguishable on that feature, exhibiting a high degree of overlap.

[0091] Based on the above analysis, this embodiment optimizes the false anomaly screening and separation model and implements it on the MATLAB platform. The data processed by the model comes from the hybrid tuning values ​​generated by the hybrid tuning filtering technique. In this embodiment, after performing false anomaly screening and separation on the representative region 'a', the results show ( Figure 7The overall data exhibits a bimodal normal distribution. Meanwhile, the hybrid tuned filtering effectively enhances the representation of cloud information in the grayscale image and weakens the interference of other non-cloud information, thus naturally dividing the entire image into two categories: clouds and non-clouds. Figure 7 These two clusters correspond to two distinct normal distribution patterns: the more concentrated portion on the left is defined as Normal Distribution I, while the relatively dispersed portion on the right is labeled as Normal Distribution II. This further demonstrates the inherent structural complexity of the data and provides important evidence for subsequent data analysis and interpretation.

[0092] After processing the data within the representative region using the pseudo-anomaly screening and separation model, the key parameter values ​​for each sub-distribution were obtained, as detailed in Table 1. Specifically, the mean of the first normal distribution reached 87098.37, while the mean of the second normal distribution was significantly higher at 272010.42. Their weights in the dataset are roughly equal, at 0.48 and 0.52 respectively, indicating that background information and cloud information are similar in total quantity, with background information slightly dominant. To distinguish between these two sub-distributions, the algorithm automatically set corresponding threshold ranges, dividing the dataset into two parts: one from the minimum value to 182059.38, and the other from 182059.38 to the maximum value. This division strategy is based on the natural distribution characteristics of the data.

[0093] Further analysis revealed a difference in the separation of the two sub-distributions. The separation of the first sub-distribution was 1.07, which was higher than that of the second sub-distribution (0.79). This indicates that the first sub-distribution is more independent and easier to distinguish in the dataset. In contrast, the second sub-distribution is slightly more difficult to separate.

[0094] Furthermore, the overlap rates between the two normal distributions were calculated, which were 33.57% and 30.23%, respectively. This metric reveals the proportion of overlap between the two distributions in the dataset, and is of great significance for understanding the complexity of the data structure and subsequent data processing strategies. The parameter values ​​of each sub-distribution of the false anomaly screening separation model are shown in Table 1.

[0095] Table 1

[0096]

[0097] For representative region b, an optimized false anomaly screening and separation model is used for analysis, such as... Figure 8As shown, this embodiment obtains detailed sub-distribution parameter values, which accurately characterize the properties of each independent component in the mixed distribution. Here, this embodiment defines the normal distribution on the left as "Normal Distribution I," and the normal distribution on the right as "Normal Distribution II." Through the parameter values ​​calculated by this model, this embodiment can gain a deep understanding of the specific information of each sub-distribution, providing a solid foundation for further refinement of data processing. The parameter values ​​of each sub-distribution in the false anomaly screening and separation model are shown in Table 2.

[0098] Table 2

[0099]

[0100] The averages listed in the table refer to the central tendency of the datasets within their respective normal distributions. Specifically, the averages of the two sub-distributions are 2783.79 and 14790.59, respectively. The weights of the two sub-distributions are 0.38 and 0.62, respectively, directly reflecting that background information accounts for a larger proportion of the overall data, while cloud information accounts for a relatively smaller proportion. The false anomaly screening and separation model, with its intelligent characteristics, can automatically define an appropriate threshold range, thereby achieving accurate feature extraction. For the two normal distributions, the model sets the threshold ranges from the minimum value to 5305.43, and from 5305.43 to the maximum value, respectively. The separation degrees of the two sub-distributions differ; the separation degree of the second sub-distribution reaches 1.09, higher than the 0.73 of the first sub-distribution, indicating that the second sub-distribution is easier to distinguish in the overall data. Furthermore, the overlap rate reveals the degree of data overlap between the sub-distributions, which is crucial for understanding the complexity of the data structure. The overlapping data of the two sub-distributions accounted for 47.27% and 35.01% of the total data, respectively, which provides a valuable reference for subsequent data processing and analysis.

[0101] To intuitively understand the classification effect of the improved false anomaly screening and separation model, this embodiment displays the cloud information automatically extracted by the improved false anomaly screening and separation model in the visualization software ArcGIS 10.8. Figure 9 The image shows the cloud extraction results for representative region a, where yellow represents extracted cloud information and black represents non-cloud underlying surface areas. This image was synthesized using false-color techniques. Figure 2 The comparison with the cloud extraction results shows that thick cloud areas were completely extracted, and even small point clouds were not missed, proving the effectiveness of the algorithm for extracting information from small, difficult-to-extract cloud bodies. However, the patchy snow accumulation on the hills in the northern part of the image generated some false anomalies, and some ice and snow pixels, as well as irregularly distributed frozen rivers and other high-reflectivity surface pixels in the background area, were misidentified as clouds.

[0102] Figure 10 This is the automatic cloud extraction result for the second representative region b. The yellow area represents clouds identified automatically, and the black area represents non-cloud areas. This image is synthesized using false-color techniques. Figure 5 The results of cloud extraction show that thick clouds were completely extracted in the false-color composite image. Figure 5 The almost invisible, hazy clouds in the lower left corner were also detected relatively completely, thus proving the practical feasibility of the algorithm. However, due to the difficulty in extracting this scene image, it is inevitable that confusing false anomalies will be generated. In the snow and ice area in the northern part of the image, some snow and ice were misidentified as clouds, and the same situation exists in the high-reflectivity underlying surface area.

[0103] Based on the above analysis, this embodiment demonstrates that the false anomaly screening and separation model exhibits high accuracy in extracting cloud information, especially when facing a series of complex or challenging image scenarios. These scenarios include, but are not limited to: (1) extremely complex underlying backgrounds, such as mottled cloud areas formed by high-altitude snow cover, or areas covered with irregular, highly reflective frozen rivers, which traditional cloud identification methods often struggle to identify; (2) foggy cloud information with blurred shapes and indistinct contour features, making it difficult for traditional identification methods to capture; and (3) minute cloud information that is almost imperceptible due to its small size. In these cases, the false anomaly screening and separation model, with its unique algorithmic advantages, can extract cloud information more effectively, thus demonstrating significant application advantages.

[0104] The False Anomaly Screening and Separation Model has achieved the following key improvements and enhancements in cloud screening and separation: (1) Automatic Threshold Calculation: The model can intelligently calculate a reasonable threshold range based on the inherent differences between the target area and its background, which is used to accurately distinguish whether each pixel in the satellite image belongs to the area to be extracted or the background area. This innovation not only significantly reduces human intervention and subjectivity in the threshold judgment process, but also greatly improves the efficiency and accuracy of target feature extraction, ensuring the objectivity and reliability of the results. (2) Local Subdivision Optimization: In response to the problem that global threshold classification may lead to the obscuring of the internal structure of the data and poor classification effect, the False Anomaly Screening and Separation Model introduces a local subdivision strategy. It performs secondary or even multiple subdivision processing on the local data based on the specific parameters of each sub-distribution in the mixed distribution. This refined classification method effectively avoids the limitations brought by the global threshold, so that the information of interest can be extracted more accurately, improving the detail and accuracy of the classification. (3) Introduction of Separation Evaluation: In order to more intuitively reflect the degree of distinction between the feature to be extracted and the background information, the model introduces the separation index. The degree of separation directly reflects the separability between the two elements: a high degree of separation means that the boundary between the extracted feature and the background information is clear and easy to separate; conversely, a low degree of separation indicates that there is a large amount of confusion between the two, making them difficult to distinguish effectively. The introduction of this indicator provides a quantitative basis for evaluating the extraction effect, helping users to more scientifically assess model performance and optimize parameter settings.

[0105] S5. Spatial statistical analysis of the cloud information is performed using the Anselm Local Moran Index to quantify the spatial autocorrelation at a specific location.

[0106] Specifically, to eliminate a small portion of false anomalies remaining in the false anomaly screening and separation model, this embodiment incorporates the Anselm Local Moran Index (AMI) into the accurate cloud identification model for snowfall areas. As a spatial statistical tool, the AMI is specifically designed to quantify spatial autocorrelation at a specific location, a characteristic particularly important in geographic data analysis, environmental science, and social science research. The AMI can identify spatially clustered high- or low-value feature areas, i.e., so-called spatial clusters. Furthermore, it has the ability to identify spatial data points that significantly deviate from the overall pattern in spatial distribution, exhibiting anomalies. The statistical calculation of the AMI is shown below:

[0107]

[0108] Where, x i It is an attribute of feature i. w represents the mean of the corresponding attribute. i,j Let i be the spatial weight between features i and j.

[0109]

[0110] Where n represents the total number of feature elements.

[0111] In this embodiment, the Anselm Local Moran Index is used to eliminate potential false anomalies. Through its sophisticated calculation mechanism, this embodiment can quantify the spatial clustering and dispersion of clouds and potential false anomalies. This index is calculated independently for each detection point to assess its spatial distribution pattern, whether it presents as a tight cluster, a loose dispersion, or a random distribution. The calculated I value, as the core output of the Anselm Local Moran Index, spans a continuous range from +1 (indicating extremely strong positive spatial autocorrelation) to -1 (indicating extremely strong negative spatial autocorrelation). When the I value is 0, it means that the spatial distribution of the region exhibits randomness and no significant autocorrelation. The sign and magnitude of the I value directly map the clustering or dispersion of spatial entities in different geographical regions, providing a quantitative basis for understanding spatial structure.

[0112] Therefore, in this embodiment, regions with positive I values ​​are interpreted as spatial clusters of clouds or related features, a finding that is of great significance for further spatial analysis and decision-making.

[0113] A visualization of representative region a after Anselm's local Moran index analysis is shown in the attached figure. Figure 11 As shown in the figure, the red areas represent clouds, while the yellow areas are potential false anomaly regions. This process effectively quantifies the spatial clustering and dispersion characteristics of the entire dataset, providing an intuitive and scientific assessment of the spatial distribution pattern of the data.

[0114] It is worth noting that the I-value of snow cover is typically much lower than that of clouds. Even when clouds are relatively thin, their I-value remains above 0, indicating that clouds tend to form a dense, aggregated structure in space. In contrast, due to topographical segmentation caused by ridgelines or significant temperature differences between sunny and shady slopes, snow cover often appears as patches, resulting in its lower I-value. By applying the Anselm Local Moran Index, this embodiment reclassifies some false anomalies in the image as noise anomalies and effectively distinguishes between salt-and-pepper noise generated by snow-covered areas and high-brightness complex underlying surfaces. The results clearly show that these previously indistinguishable noise points have now been effectively separated from cloud information, further improving the accuracy and reliability of cloud identification. This process not only demonstrates the powerful ability of the Anselm Local Moran Index to distinguish spatial structural differences in complex environments but also provides a solid foundation for subsequent spatial data processing and analysis.

[0115] The results of the Anselm local Moran index analysis in representative region b are as follows: Figure 12As shown, this embodiment discovered that a false anomaly of snow cover, originally misidentified due to its similar spectral characteristics to clouds, located in the northern hilly area, was effectively eliminated after this analysis. This achievement not only improves the accuracy of data processing but also further verifies the ability of the Anselm Local Moran Index to distinguish between real cloud cover and false anomaly information in complex environments.

[0116] This embodiment provides a detailed analysis of the core mechanism and implementation process of a cloud-snow separation algorithm based on multi-dimensional information fusion. For accurate cloud identification in snowfall areas, this embodiment introduces a hybrid tuned filtering technique to enhance the identification capability of cloud information while effectively suppressing spectral interference caused by complex surface backgrounds. Subsequently, a false anomaly screening separation model is used to lock in cloud features, further improving identification accuracy. Finally, this embodiment utilizes the Anselm local Moran index to explore the spatial correlation between clouds and potential noise false anomalies, accurately eliminating misjudgments caused by complex terrain. This step not only enhances the robustness of the algorithm but also ensures the purity of the cloud-snow separation results. The results show that the algorithm exhibits excellent cloud-snow separation performance in complex scenarios such as complex terrain, diverse underlying surface background information, and blurred cloud outlines and fog-like distributions, achieving satisfactory separation results. This discovery not only broadens the application scope of cloud identification technology but also provides strong technical support for subsequent remote sensing data processing and analysis.

[0117] To verify the effectiveness of the proposed method, this embodiment compares the proposed method with artificial neural networks and the Fmask method, both of which have good performance in cloud extraction. Figure 13 Four regions of interest from the cloud extraction results are shown. Figure 13 The first row shows the original image, covering various terrain environments including bright backgrounds, mountains, ice, and snow. The second row is a false-color composite image, using bands 7, 4, and 2 to represent red, green, and blue, respectively, helping the viewer visually distinguish between snow / ice and clouds. Rows 3-5 show the results of cloud extraction using ANN, Fmask, and the proposed method, respectively, where yellow indicates extracted cloud information and black indicates non-cloudy background areas.

[0118] In comparison, such as Figure 13 As shown, both the ANN and Fmask methods over-identify cloud information, which verifies the fact that images with snow and ice in the background are easily misclassified as clouds. The comparison figures demonstrate that the proposed method performs better in extracting cloud information and effectively reduces the confusion between clouds and snow / ice. The table below presents a quantitative representation of the recognition results. This embodiment uses three different accuracy evaluation methods to assess the accuracy of the algorithm results. Table 3 compares the producer accuracy, user accuracy, and overall accuracy of this embodiment with those of the artificial neural network and FMSK algorithms.

[0119] Table 3

[0120]

[0121] The results of this embodiment outperform artificial neural network algorithms and the Fmask method, especially in terms of user accuracy, which is significantly improved compared to the other two methods. This is because the method effectively improves the phenomenon of misidentifying snow and ice as clouds and reduces the occurrence of cloud-snow confusion. In areas where cloud extraction is difficult, the method achieves high-precision cloud identification. The results show that the producer accuracy exceeds 95%, the user accuracy exceeds 83%, and the overall accuracy exceeds 90%. Therefore, this method can solve the problem of misidentifying snow and ice in the background as clouds, as well as the problem of missing point clouds with extremely small morphology, and also achieves better results in cloud and snow separation in complex terrain areas.

[0122] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying clouds in snowfall areas based on multi-dimensional information fusion, characterized in that, Includes the following steps: Acquire remote sensing images of the snowfall area and select the band; Preprocessing is performed on remote sensing images of selected bands to obtain preprocessed images; The spectral information of the preprocessed image is filtered using a hybrid tuning filtering technique to obtain a grayscale image; The grayscale image is separated using a false anomaly screening and separation model based on the normal distribution characteristics to obtain cloud information and non-cloud information; Spatial statistical analysis of the cloud information was performed using the Anselm Local Moran Index to quantify the spatial autocorrelation at specific locations, eliminate false anomalies in snowfall areas, and obtain cloud identification results. The process of filtering the preprocessed image using hybrid tuning filtering technology includes: The endmember spectral information of interest to the user is extracted through a partial demixing strategy to generate a hybrid tuned filter image; Based on the hybrid tuned filtered image, target regions with low spectral contrast between the target and the background, as well as within the target itself, are detected. The abundance image of the target region is adjusted using a feasibility score to obtain a grayscale image; The process of separating the grayscale image based on the normal distribution characteristics using the false anomaly screening and separation model includes: Based on the normal distribution characteristics, the grayscale image is separated using a false anomaly screening and separation model to obtain cloud information and non-cloud information with a bimodal normal distribution. The separability between the probability distributions of cloud information and non-cloud information is assessed using Bach distance. The false anomaly screening and separation model is a mixed distribution model, and its expression is: Among them, parameter set It contains a series of key elements for defining multiple branches, where n represents the total number of branches. This represents the probability density of the i-th branch. For the corresponding parameters, Let represent the weight of the i-th branch, and , >

0.

2. The method for identifying snowfall area clouds based on multi-dimensional information fusion according to claim 1, characterized in that, The selected bands are the nine bands of Landsat 8 OLI data, specifically Band 1 to Band 9.

3. The method for identifying snowfall area clouds based on multi-dimensional information fusion according to claim 1, characterized in that, The preprocessing of remote sensing images in selected bands includes: The remote sensing images of the selected bands were resampled using a cubic convolution interpolation method, and the resolution of the Band 8 band was resampled from 15 meters to 30 meters to match the resolution of other multispectral bands.

4. The method for identifying snowfall area clouds based on multi-dimensional information fusion according to claim 1, characterized in that, It also includes the process of improving the false anomaly screening and separation model: Based on the inherent differences between the target area and its background, a reasonable threshold range is calculated, and the threshold range is used to distinguish between the area to be extracted and the background area. A local subdivision strategy was introduced, which subdivided the local data multiple times based on the specific parameters of each sub-distribution in the mixed distribution in order to extract the information of the land features of interest. A separation assessment is introduced to determine the degree of separation between the ground features to be extracted and the background information based on the magnitude of the separation.

Images