A polar meteorological contrast observation evaluation method and system based on image data

Abstract

The application belongs to the technical field of image processing, and relates to a polar region meteorological contrast observation evaluation method and system based on image data. The method comprises the following steps: S1: inputting a polar region observation image into a self-supervised visual model, extracting a self-attention matrix of a last layer multi-attention head, obtaining a global attention graph based on the self-attention matrix, and converting the global attention graph into a feature semantic binary matrix; S2: calculating a semantic distance based on the feature semantic binary matrix and a feature tensor output by the self-supervised visual model; S3: calculating information entropy based on the global attention graph; S4: calculating the maximum value of a 4-neighborhood of each pixel, and calculating the standard deviation of all effective pixels in the polar region observation image based on the maximum value; and S5: calculating a comprehensive evaluation score based on the semantic distance, the information entropy and the standard deviation, and performing polar region meteorological contrast observation evaluation based on the comprehensive evaluation score. The method has the advantages of high calculation efficiency, strong environmental adaptability, no need for a reference image and the like, and can effectively improve the polar region meteorological contrast observation evaluation level.

Description

Technical Field

[0001] This invention belongs to the field of image processing and artificial intelligence technology, and relates to a method and system for polar meteorological contrast observation and evaluation, particularly a method and system for polar meteorological contrast observation and evaluation based on image data. Background Technology

[0002] Antarctica is the continent with the most extreme weather and climate on Earth, and the largest ice and snow cover. It is frequently accompanied by strong winds, blowing snow, fog, low clouds, and whiteout, all of which severely restrict human activities. Its unique ice and snow surface environment makes visual reference conditions extremely prone to deterioration, posing a significant safety hazard, especially to polar aviation operations. In polar meteorological observation, the analysis and evaluation of contrast conditions using images acquired by fixed cameras is of great significance for logistical support, scientific research activities, and aviation operations in polar regions. Therefore, there is an urgent need to develop polar meteorological contrast observation and evaluation methods and systems based on image data.

[0003] However, traditional image contrast evaluation methods mainly have the following problems:

[0004] (1) Reliance on whole image processing: Traditional contrast evaluation methods (such as Michelson contrast and RMS contrast) usually process the entire image, including a large number of invalid areas (such as pure white sky, pure black shadows, etc.), which leads to a waste of computing resources and bias in evaluation results.

[0005] (2) Requires reference image: Most existing methods require an ideal reference image as a benchmark, but in extreme polar environments, it is almost impossible to obtain an ideal reference image of the same scene.

[0006] (3) Poor environmental adaptability: The special environment of the polar region (whitening, low contrast, low visibility, strong scattering and reflection, fogging, etc.) makes it difficult for traditional algorithms to accurately evaluate the image quality of the actual observation area.

[0007] (4) Evaluation results of interference in invalid areas: The effective data area in polar observation images usually only accounts for 30-50% of the entire image. Traditional methods process invalid areas indiscriminately, which is easily affected by large areas of snow without information.

[0008] Based on the above problems, there is an urgent need for a method that can perform efficient and accurate contrast assessment using only the effective data area of ​​an image, in order to adapt to the special observation environment in polar regions. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention proposes a polar meteorological contrast observation and evaluation method and system based on image data. It has the advantages of high computational efficiency, strong environmental adaptability, and no need for reference images, and can effectively improve the evaluation level of polar meteorological contrast observation.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A method for evaluating polar meteorological contrast observation based on image data, characterized by the following steps:

[0012] S1: Input the polar region observation image into the self-supervised visual model, extract the self-attention matrix of the multi-attention head of the last layer of the self-supervised visual model, obtain the global attention map based on the self-attention matrix of the multi-attention head, and convert the global attention map into a feature semantic binary matrix.

[0013] S2: Based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, extract the mean features of the foreground region and the mean features of the background region, and calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region;

[0014] S3: Calculate the information entropy based on the global attention graph;

[0015] S4: Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel;

[0016] S5: Calculate the comprehensive evaluation score based on the semantic distance, information entropy, and standard deviation, and conduct polar meteorological contrast observation evaluation based on the comprehensive evaluation score.

[0017] Preferably, step S1 specifically includes:

[0018] S11: Image of polar observation Input the data into the self-supervised visual model and extract the self-attention matrix of the multi-attention head in the last layer of the self-supervised visual model;

[0019] S12: Aggregate the self-attention matrices of multiple attention heads to obtain the global attention graph:

[0020]

[0021] in, Represents a global attention graph; This represents the last layer of the self-supervised visual model. A self-attention matrix for each attention head; This represents the total number of attention heads in the last layer of the self-supervised visual model;

[0022] S13: Convert the global attention map into a feature semantic binary matrix:

[0023]

[0024] in, Represents the semantic binary matrix of features The Middle Line number The element values ​​of the column; Represents a global attention graph The Middle Line number The characteristic values ​​of the column; Indicates the dynamic semantic threshold factor; Represents a global attention graph The maximum value of all eigenvalues.

[0025] Preferably, step S2 specifically includes:

[0026] S21: Based on the aforementioned feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, extract the mean features of the foreground region and the mean features of the background region:

[0027]

[0028]

[0029] in, This indicates the mean characteristics of the foreground region; and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; The output of the self-supervised visual model represents the first... Line number The characteristic tensor of the column; Represents the semantic binary matrix of features The Middle Line number The element values ​​of the column; This indicates the mean characteristics of the background region;

[0030] S22: Calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region:

[0031]

[0032] in, Indicates semantic distance.

[0033] Preferably, step S3 specifically includes:

[0034] S31: Normalize each feature value in the global attention map to obtain a normalized attention map:

[0035]

[0036] in, The first character in the normalized attention graph represents the... Line number The characteristic values ​​of the column; Represents a global attention graph The Middle Line number The characteristic values ​​of the column; and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; Indicates to Take the exponent;

[0037] S32: Calculate the proportion of each feature value in the normalized attention map:

[0038]

[0039] in, Indicates that the value is equal to The proportion of the feature values ​​in the normalized attention map; This indicates that the value in the normalized attention graph is equal to... The number of eigenvalues;

[0040] S33: Calculate the information entropy based on the proportion of each feature value:

[0041]

[0042] in, This represents information entropy.

[0043] Preferably, step S4 specifically includes:

[0044] S41: Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image:

[0045]

[0046] in, The polar region observation image represents The Middle Line number The maximum value of the 4-neighborhood of a column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel;

[0047] S42: Calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel:

[0048]

[0049]

[0050] in, The polar region observation image represents The mean of all valid pixels in the array; and These respectively represent the polar region observation images. The original height and width; The polar region observation image represents The standard deviation of all valid pixels.

[0051] Preferably, step S5 specifically includes:

[0052] S51: Calculate the comprehensive evaluation score based on the semantic distance, information entropy, and standard deviation:

[0053]

[0054] in, This indicates the overall evaluation score; Indicates semantic distance; Represents information entropy; Indicates standard deviation; ;

[0055] S52: Based on the comprehensive evaluation score, a contrast score is calculated, and the contrast score is used to evaluate the contrast observations of polar meteorological conditions.

[0056]

[0057] in, Indicates contrast rating; Indicates the steepness coefficient; This represents the midpoint threshold.

[0058] Preferably, , .

[0059] Furthermore, the present invention also provides a polar meteorological contrast observation and evaluation system based on image data, characterized in that it includes:

[0060] A global attention map and feature semantic binarization matrix construction module is used to input polar region observation images into a self-supervised visual model, extract the self-attention matrix of the multi-attention head of the last layer of the self-supervised visual model, obtain a global attention map based on the self-attention matrix of the multi-attention head, and convert the global attention map into a feature semantic binarization matrix.

[0061] The semantic distance calculation module is used to extract the mean features of the foreground region and the mean features of the background region based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, and to calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region.

[0062] Information entropy calculation module, which is used to calculate information entropy based on the global attention map;

[0063] The standard deviation calculation module is used to calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and to calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel.

[0064] The comprehensive evaluation score calculation and evaluation module is used to calculate the comprehensive evaluation score based on the semantic distance, information entropy and standard deviation, and to conduct polar meteorological contrast observation evaluation based on the comprehensive evaluation score.

[0065] Furthermore, the present invention also provides a polar meteorological contrast observation and evaluation device based on image data, characterized in that it includes:

[0066] One or more processors;

[0067] Memory, used to store one or more programs;

[0068] When the one or more programs are executed by the one or more processors, the one or more processors implement the polar meteorological contrast observation and evaluation method based on image data as described above.

[0069] Finally, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the steps of the polar meteorological contrast observation and evaluation method based on image data as described above.

[0070] Compared with existing technologies, the polar meteorological contrast observation and evaluation method based on image data of the present invention has one or more of the following beneficial technical effects:

[0071] (1) Semantic robustness: This invention uses DINOv3 as a self-supervised visual model, which has a strong structural perception capability. Even in low contrast environments, it can identify targets through geometric features, avoiding false alarms caused by the same color in the ground and space.

[0072] (2) No manual annotation required: This invention utilizes a self-supervised visual model, which eliminates the need for massive supervised fine-tuning for polar scenes and has low deployment costs.

[0073] (3) Focus on effective observation: This invention can automatically ignore pure white / pure black areas with no information, and the evaluation weight is concentrated on scenes with scientific research value. Attached Figure Description

[0074] Figure 1 This is a flowchart illustrating the polar meteorological contrast observation and evaluation method based on image data of the present invention.

[0075] Figure 2 This is a schematic diagram of the polar meteorological contrast observation and evaluation system based on image data according to the present invention;

[0076] Figure 3 This is a structural block diagram of the polar meteorological contrast observation and evaluation device based on image data according to the present invention. Detailed Implementation

[0077] Before detailing any embodiment of the invention, it should be understood that the invention, in its application, is not limited to the details of the construction and arrangement of the components set forth in the following description or illustrated in the following figures. The invention can have other embodiments and can be practiced or carried out in various ways. Furthermore, it should be understood that the wording and terminology used herein are for descriptive purposes and should not be considered limiting. The use of “comprising” or “having” and variations thereof in this invention is intended to cover the items set forth below and their equivalents, as well as any additional items. Unless otherwise specified or limited, the terms “installation,” “connection,” “support,” and “linkage,” and variations thereof are used broadly and cover both direct and indirect installation, connection, support, and linking. Moreover, “connection” and “linkage” are not limited to physical or mechanical connections or links.

[0078] Furthermore, firstly, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting the invention. Secondly, the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple. The term "a" should not be construed as a limitation on the quantity.

[0079] To address the problems existing in current technologies, this invention proposes a polar meteorological contrast observation and evaluation method that integrates the semantic understanding capabilities of a self-supervised large model. This method employs a three-stage architecture: semantic recognition, multi-dimensional contrast feature analysis, and comprehensive quality mapping. First, a pre-trained DINOv3 model is used to extract the self-attention map of the image, automatically identifying semantically valuable target regions (foreground regions) through the energy distribution of multi-head attention. Then, the relative measures of target features (foreground features) and background features are calculated within the feature manifold space. Finally, a contrast evaluation index of 0-100 points is generated by comprehensively evaluating the nonlinear function.

[0080] Figure 1 A flowchart illustrating the polar meteorological contrast observation and evaluation method based on image data of the present invention is shown. Figure 1 As shown, the polar meteorological contrast observation and evaluation method based on image data of the present invention includes the following steps:

[0081] S1: Construction of global attention map and feature semantic binary matrix.

[0082] The polar region observation image is input into a self-supervised visual model. The self-attention matrix of the last layer of the self-supervised visual model is extracted from the multi-attention head. Based on the self-attention matrix of the multi-attention head, a global attention map is obtained and the global attention map is transformed into a feature semantic binary matrix. Specifically, this includes:

[0083] 1. Polar observation images Input the data into the self-supervised visual model and extract the self-attention matrix of the multi-attention head in the last layer of the self-supervised visual model.

[0084] In this invention, DINOv3 is used as a self-supervised visual model, which possesses extremely strong structure perception capabilities. Even in low-contrast environments, it can identify targets through geometric features, avoiding false alarms caused by the similarity of ground and air colors. Polar observation images are used. Inputting the data into the DINOv3 model, the self-attention matrix of the multi-attention head of the last Transformer block is extracted. Each attention head independently calculates the attention weights between all patch pairs and outputs a self-attention matrix.

[0085] 2. Aggregate the self-attention matrices of multiple attention heads to obtain a global attention graph.

[0086]

[0087] in, Represents a global attention graph; This represents the last layer of the DINOv3 model. A self-attention matrix for each attention head; This indicates the total number of attention heads.

[0088] Therefore, by taking the arithmetic mean of the self-attention matrices of all attention heads, the information from multiple attention heads can be fused into a global attention map.

[0089] 3. The global attention map is transformed into a feature semantic binarization matrix by dynamic threshold binarization.

[0090]

[0091] That is, when hour, This indicates that the attention intensity of this patch is high, belonging to the foreground / target region; otherwise, This indicates that the attention intensity of this pair of patches is low, belonging to the background / noise region. Therefore, a dynamic threshold with a relative maximum value can be used to convert continuous attention intensity values ​​into discrete binary masks, highlighting the semantic features of the foreground region and suppressing the background region.

[0092] in, Represents the semantic binary matrix of features The Middle Line number The element values ​​of the column; Represents a global attention graph The Middle Line number The characteristic values ​​of the column; This represents the dynamic semantic threshold factor, which is an adjustable coefficient between 0 and 1, used to control the strictness of binarization. Represents a global attention graph The maximum value of all features is used as the benchmark for the dynamic threshold.

[0093] Values ​​of all elements in all rows and columns Combining these elements, we can obtain the feature semantic binary matrix. .

[0094] In this invention, the dynamic semantic threshold factor It can be dynamically adjusted. For example, in the intense daylight conditions of polar regions, Automatically adjusts the brightness, retaining only areas with extremely high visibility at a value of 1, effectively filtering out strong background noise such as snow reflections and ice textures; even in low-light conditions like polar nights. Automatically lowering the brightness allows it to capture the outlines of less noticeable targets, preventing them from being completely filtered out in low light.

[0095] Therefore, when performing semantic observation region recognition, this invention uses the DINOv3 model to extract features from the input image, takes the attention matrix of the last layer's attention head, and upsamples the attention matrix to convert it into a semantic saliency mask of the input image size, i.e., a feature semantic binarization matrix. Therefore, this invention can effectively eliminate invalid areas such as pure white sky, snow, and pure black shadows, ensuring that image areas with practical observation value are obtained.

[0096] S2: Semantic distance calculation.

[0097] Based on the aforementioned feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, the mean features of the foreground region and the mean features of the background region are extracted, and the semantic distance is calculated based on the mean features of the foreground region and the mean features of the background region. Specifically, this includes:

[0098] 1. Based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, extract the mean features of the foreground region and the mean features of the background region.

[0099]

[0100]

[0101] in, The mean feature of the foreground region represents the overall semantic features of the foreground region (target region); and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; and These represent the height and width of the feature map (i.e., the number of rows and columns of the feature tensor output by the self-supervised visual model). The output of the self-supervised visual model represents the first... Line number The characteristic tensor of the column has a total of indivual; Represents the semantic binary matrix of features The Middle Line number The element values ​​of the column; The mean feature of the background region is represented by the overall semantic feature of the background region.

[0102] 2. Calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region.

[0103]

[0104] in, The semantic distance represents the distance between foreground and background regions, ranging from 0 to 2. This semantic distance quantifies the semantic differences between these regions, allowing for the assessment of target identifiability. When the value is close to 0, it means that the target and the background semantics are almost completely identical, and the target is submerged in the background (such as being obscured by dense fog or reflected light from snow). When the value is large, it indicates that the target is semantically easily identifiable; for example, when When the value is close to 2, it indicates that the semantic difference between the target and the background is extremely large, and the target is very easy to identify.

[0105] S3: Information entropy calculation.

[0106] In this invention, the calculation of information entropy based on the global attention map specifically includes:

[0107] 1. Normalize each feature value in the global attention map to obtain a normalized attention map.

[0108]

[0109] in, The first character in the normalized attention graph represents the... Line number The column's characteristic values ​​range from (0,1], and the sum of all elements is 1; Represents a global attention graph The Middle Line number The characteristic values ​​of the column; and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; Indicates to The purpose of using an index is to amplify high attention values ​​and suppress low attention values, thereby highlighting the differences in distribution.

[0110] Therefore, the original attention intensity can be transformed into a standard probability distribution, which prepares for subsequent calculation of information entropy.

[0111] 2. Calculate the proportion of each feature value in the normalized attention map.

[0112]

[0113] in, Indicates that the value is equal to The proportion of the feature values ​​in the normalized attention map; This indicates that the value in the normalized attention graph is equal to... The number of eigenvalues.

[0114] In actual calculations, due to Since the values ​​are continuous, we can first discretize them into bins (e.g., divide them into 256 intervals), then calculate the mean of the element values ​​in each interval and use this mean as the average value. .

[0115] 3. Calculate the information entropy based on the proportion of each feature value.

[0116]

[0117] in, Represents information entropy, with a value range of . It is used to quantify the degree of concentration of attention distribution in order to assess image sharpness. Specifically, when the image is sharp, attention is highly focused on the edges and key structures of the foreground area, the distribution is sharp, and the entropy value H is small; when the image is blurred due to heavy snow or frost, the attention distribution becomes diffuse, and the entropy value H increases.

[0118] S4: Standard deviation calculation.

[0119] Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel, specifically including:

[0120] 1. Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image.

[0121]

[0122] in, The polar region observation image represents The Middle Line number The maximum value of the 4-neighborhood of a column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel.

[0123] Therefore, it is possible to capture the strongest edge / texture information of each pixel and quantize the intensity of local details in the image. matrix.

[0124] 2. Calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel.

[0125]

[0126]

[0127] in, The polar region observation image represents The average value of all valid pixels in the image reflects the overall average texture intensity of the image; and These respectively represent the polar region observation images. The original height and width; The polar region observation image represents The standard deviation of all valid pixels in the dataset ranges from [0, 127.5]. The polar region observation image represents The total number of pixels.

[0128] Therefore, the standard deviation can be used to... To evaluate the microscopic sharpness of an image (texture detail, edge sharpness, etc.). Among these, The larger the value, the greater the difference in gradient intensity in the image, indicating the presence of obvious edges and textures (such as clear target outlines and areas with rich details), and high microscopic clarity. The smaller the value, the more uniform and lower the overall gradient intensity of the image, and the more details are obscured by blurring / noise (such as frost, fog, snow scenes, etc.), resulting in low microscopic clarity.

[0129] S5: Calculation and evaluation of comprehensive assessment score.

[0130] A comprehensive evaluation score is calculated based on the semantic distance, information entropy, and standard deviation, and a polar meteorological contrast observation evaluation is conducted based on the comprehensive evaluation score. This specifically includes:

[0131] 1. Calculate the comprehensive evaluation score based on the semantic distance, information entropy, and standard deviation.

[0132]

[0133] in, This indicates the overall evaluation score; Indicates semantic distance; Represents information entropy; Indicates standard deviation; It can adjust according to local conditions (e.g., strong / weak light in polar regions, different weather conditions, etc.), for example, it can improve semantic distance under low light conditions. weight Strong light can improve the microscopic clarity index, that is, the standard deviation. weight .

[0134] In this invention, due to semantic distance The value range is [0,2]. Dividing by 2 normalizes it to [0,1]. A larger value indicates a greater semantic difference between the foreground and background regions, making them easier to identify. Due to information entropy... The value range is [0, 8], which is normalized to [0, 1] after dividing by 8. The larger the standard deviation, the clearer the image; The value range is [0, 127.5]. After dividing by 127.5, it is normalized to [0, 1]. The larger the value, the richer the image details and the higher the microscopic clarity.

[0135] Therefore, in this invention, when performing multi-dimensional contrast feature analysis, on the one hand, the standard deviation of the gradient magnitude of the pixel neighborhood is calculated at the microscale; on the other hand, the semantic region of actual observation value is analyzed from the pixel dimension to a high-dimensional feature manifold space, extracting the mean features of the foreground region and the mean features of the background region, and calculating the difference in semantic features between the two through cosine distance, thus solving the problem that traditional physical contrast cannot measure the discernibility of targets in complex environments; furthermore, the feature value distribution entropy value of the semantic region of observation value is statistically analyzed. The features of the three dimensions are fused through adaptive weights to form a comprehensive contrast feature vector. Among them, the weight coefficients can be dynamically adjusted according to the polar illumination conditions, emphasizing micro features during the daytime and macro features during windy, snowy, and nighttime periods.

[0136] 2. Based on the comprehensive evaluation score, a contrast score is calculated, and the contrast score is used to evaluate the contrast observation of polar meteorological conditions.

[0137]

[0138] in, Indicates contrast rating; This represents the steepness coefficient, used to control the steepness of the Sigmoid curve. The larger the value, the faster the score changes around the midpoint, and the stronger the discrimination. This represents the midpoint threshold, which is the dividing point for scoring.

[0139] Preferably, , It enhances the distinction of medium contrast scenes.

[0140] Therefore, this invention can be based on polar region observation images. The semantic distance, information entropy, and standard deviation are used to extract contrast features, and a nonlinear mapping model is constructed to evaluate the comprehensive score of [0,1]. Contrast rating mapped to [0,100] points This enables a quantitative assessment of contrast observations.

[0141] With the aforementioned contrast rating The contrast rating can then be used. Conduct polar meteorological contrast observation assessments. For example, when the contrast score... When the contrast score is below a preset threshold (e.g., 40 points), it is recorded as low contrast; when the contrast score is... A score between 40 and 70 is recorded as medium contrast; when the contrast score is... A score above 70 is recorded as high contrast.

[0142] Figure 2 A schematic diagram of the polar meteorological contrast observation and evaluation system based on image data of the present invention is shown. As shown, the polar meteorological contrast observation and evaluation system based on image data of the present invention includes:

[0143] 1. Global attention map and feature semantic binarization matrix construction module.

[0144] The global attention map and feature semantic binarization matrix construction module is used to input polar region observation images into a self-supervised vision model, extract the self-attention matrix of the multi-attention head of the last layer of the self-supervised vision model, obtain a global attention map based on the self-attention matrix of the multi-attention head, and convert the global attention map into a feature semantic binarization matrix.

[0145] 2. Semantic distance calculation module.

[0146] The semantic distance calculation module is used to extract the mean features of the foreground region and the mean features of the background region based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, and to calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region.

[0147] 3. Information entropy calculation module.

[0148] The information entropy calculation module is used to calculate information entropy based on the global attention map.

[0149] 4. Standard deviation calculation module.

[0150] The standard deviation calculation module is used to calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and to calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel.

[0151] 5. Comprehensive evaluation score calculation and evaluation module.

[0152] The comprehensive evaluation score calculation and evaluation module is used to calculate the comprehensive evaluation score based on the semantic distance, information entropy and standard deviation, and to conduct polar meteorological contrast observation evaluation based on the comprehensive evaluation score.

[0153] Furthermore, this invention also provides a polar meteorological contrast observation and evaluation device based on image data. For example... Figure 3As shown, the polar meteorological contrast observation and evaluation device based on image data of the present invention includes: a memory 11 for storing one or more programs; one or more processors 12; when the one or more programs are executed by the one or more processors 12, the one or more processors 12 implement the polar meteorological contrast observation and evaluation method based on image data of the present invention.

[0154] Finally, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the polar meteorological contrast observation and evaluation method based on image data in the present invention.

[0155] The computer-readable storage medium includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. As defined in this invention, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0156] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0157] The steps of the methods or algorithms described in conjunction with the embodiments disclosed in this invention can be implemented in hardware, software modules executed by a processor, or a combination of both. The software modules can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention based on the concept of the present invention, without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for evaluating polar meteorological contrast based on image data, characterized in that, Includes the following steps: S1: Input the polar region observation image into the self-supervised visual model, extract the self-attention matrix of the multi-attention head of the last layer of the self-supervised visual model, obtain the global attention map based on the self-attention matrix of the multi-attention head, and convert the global attention map into a feature semantic binary matrix. S2: Based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, extract the mean features of the foreground region and the mean features of the background region, and calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region; S3: Calculate the information entropy based on the global attention graph; S4: Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel; S5: Calculate the comprehensive evaluation score based on the semantic distance, information entropy, and standard deviation, and conduct polar meteorological contrast observation evaluation based on the comprehensive evaluation score.

2. The polar meteorological contrast observation and evaluation method based on image data according to claim 1, characterized in that, Step S1 specifically includes: S11: Image of polar observation Input the data into the self-supervised visual model and extract the self-attention matrix of the multi-attention head in the last layer of the self-supervised visual model; S12: Aggregate the self-attention matrices of multiple attention heads to obtain the global attention graph: in, Represents a global attention graph; This represents the last layer of the self-supervised visual model. A self-attention matrix for each attention head; This represents the total number of attention heads in the last layer of the self-supervised visual model; S13: Convert the global attention map into a feature semantic binary matrix: in, Represents the semantic binary matrix of features The Middle Line 1 The element values ​​of the column; Represents a global attention graph The Middle Line 1 The characteristic values ​​of the column; Indicates the dynamic semantic threshold factor; Represents a global attention graph The maximum value of all eigenvalues.

3. The polar meteorological contrast observation and evaluation method based on image data according to claim 1, characterized in that, Step S2 specifically includes: S21: Based on the aforementioned feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, extract the mean features of the foreground region and the mean features of the background region: in, This indicates the mean characteristics of the foreground region; and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; The output of the self-supervised visual model represents the first... Line 1 The characteristic tensor of the column; Represents the semantic binary matrix of features The Middle Line 1 The element values ​​of the column; This indicates the mean characteristics of the background region; S22: Calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region: in, Indicates semantic distance.

4. The polar meteorological contrast observation and evaluation method based on image data according to claim 1, characterized in that, Step S3 specifically includes: S31: Normalize each feature value in the global attention map to obtain a normalized attention map: in, The first character in the normalized attention graph represents the... Line 1 The characteristic values ​​of the column; Represents a global attention graph The Middle Line 1 The characteristic values ​​of the column; and These represent polar observation images. The original height and width; This represents the block size of the self-supervised visual model; Indicates to Take the exponent; S32: Calculate the proportion of each feature value in the normalized attention map: in, Indicates that the value is equal to The proportion of the feature values ​​in the normalized attention map; This indicates that the value in the normalized attention graph is equal to... The number of eigenvalues; S33: Calculate the information entropy based on the proportion of each feature value: in, This represents information entropy.

5. The polar meteorological contrast observation and evaluation method based on image data according to claim 1, characterized in that, Step S4 specifically includes: S41: Calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image: in, The polar region observation image represents The Middle Line number The maximum value of the 4-neighborhood of a column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; The polar region observation image represents The Middle Line number The pixel value of the column pixel; S42: Calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel: in, The polar region observation image represents The mean of all valid pixels in the array; and These respectively represent the polar region observation images. The original height and width; The polar region observation image represents The standard deviation of all valid pixels in the dataset.

6. The polar meteorological contrast observation and evaluation method based on image data according to claim 1, characterized in that, Step S5 specifically includes: S51: Calculate the comprehensive evaluation score based on the semantic distance, information entropy, and standard deviation: in, This indicates the overall evaluation score; Indicates semantic distance; Represents information entropy; Indicates standard deviation; ; S52: Based on the comprehensive evaluation score, a contrast score is calculated, and the contrast score is used to evaluate the contrast observations of polar meteorological conditions. in, Indicates contrast rating; Indicates the steepness coefficient; This represents the midpoint threshold.

7. The polar meteorological contrast observation and evaluation method based on image data according to claim 6, characterized in that, ， 。 8. A polar meteorological contrast observation and evaluation system based on image data, characterized in that, include: A global attention map and feature semantic binarization matrix construction module is used to input polar region observation images into a self-supervised visual model, extract the self-attention matrix of the multi-attention head of the last layer of the self-supervised visual model, obtain a global attention map based on the self-attention matrix of the multi-attention head, and convert the global attention map into a feature semantic binarization matrix. The semantic distance calculation module is used to extract the mean features of the foreground region and the mean features of the background region based on the feature semantic binarization matrix and the feature tensor output by the self-supervised visual model, and to calculate the semantic distance based on the mean features of the foreground region and the mean features of the background region. Information entropy calculation module, which is used to calculate information entropy based on the global attention map; The standard deviation calculation module is used to calculate the maximum value of the 4-neighborhood of each pixel in the polar region observation image, and to calculate the standard deviation of all valid pixels in the polar region observation image based on the maximum value of the 4-neighborhood of each pixel. The comprehensive evaluation score calculation and evaluation module is used to calculate the comprehensive evaluation score based on the semantic distance, information entropy and standard deviation, and to conduct polar meteorological contrast observation evaluation based on the comprehensive evaluation score.

9. A polar meteorological contrast observation and evaluation device based on image data, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the polar meteorological contrast observation and evaluation method based on image data as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the polar meteorological contrast observation and evaluation method based on image data as described in any one of claims 1-7.

Images